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> |
2d1d43f6 | 16 | #include <linux/writeback.h> |
3947be19 DH |
17 | #include <linux/slab.h> |
18 | #include <linux/sysctl.h> | |
19 | #include <linux/cpu.h> | |
20 | #include <linux/memory.h> | |
4b94ffdc | 21 | #include <linux/memremap.h> |
3947be19 | 22 | #include <linux/memory_hotplug.h> |
3947be19 | 23 | #include <linux/vmalloc.h> |
0a547039 | 24 | #include <linux/ioport.h> |
0c0e6195 KH |
25 | #include <linux/delay.h> |
26 | #include <linux/migrate.h> | |
27 | #include <linux/page-isolation.h> | |
71088785 | 28 | #include <linux/pfn.h> |
6ad696d2 | 29 | #include <linux/suspend.h> |
6d9c285a | 30 | #include <linux/mm_inline.h> |
d96ae530 | 31 | #include <linux/firmware-map.h> |
60a5a19e | 32 | #include <linux/stop_machine.h> |
c8721bbb | 33 | #include <linux/hugetlb.h> |
c5320926 | 34 | #include <linux/memblock.h> |
698b1b30 | 35 | #include <linux/compaction.h> |
b15c8726 | 36 | #include <linux/rmap.h> |
8581fd40 | 37 | #include <linux/module.h> |
3947be19 DH |
38 | |
39 | #include <asm/tlbflush.h> | |
40 | ||
1e5ad9a3 | 41 | #include "internal.h" |
e900a918 | 42 | #include "shuffle.h" |
1e5ad9a3 | 43 | |
2d1f649c AK |
44 | enum { |
45 | MEMMAP_ON_MEMORY_DISABLE = 0, | |
46 | MEMMAP_ON_MEMORY_ENABLE, | |
47 | MEMMAP_ON_MEMORY_FORCE, | |
48 | }; | |
49 | ||
50 | static int memmap_mode __read_mostly = MEMMAP_ON_MEMORY_DISABLE; | |
51 | ||
52 | static inline unsigned long memory_block_memmap_size(void) | |
53 | { | |
54 | return PHYS_PFN(memory_block_size_bytes()) * sizeof(struct page); | |
55 | } | |
56 | ||
57 | static inline unsigned long memory_block_memmap_on_memory_pages(void) | |
58 | { | |
59 | unsigned long nr_pages = PFN_UP(memory_block_memmap_size()); | |
60 | ||
61 | /* | |
62 | * In "forced" memmap_on_memory mode, we add extra pages to align the | |
63 | * vmemmap size to cover full pageblocks. That way, we can add memory | |
64 | * even if the vmemmap size is not properly aligned, however, we might waste | |
65 | * memory. | |
66 | */ | |
67 | if (memmap_mode == MEMMAP_ON_MEMORY_FORCE) | |
68 | return pageblock_align(nr_pages); | |
69 | return nr_pages; | |
70 | } | |
71 | ||
6e02c46b | 72 | #ifdef CONFIG_MHP_MEMMAP_ON_MEMORY |
e3a9d9fc OS |
73 | /* |
74 | * memory_hotplug.memmap_on_memory parameter | |
75 | */ | |
2d1f649c AK |
76 | static int set_memmap_mode(const char *val, const struct kernel_param *kp) |
77 | { | |
78 | int ret, mode; | |
79 | bool enabled; | |
80 | ||
81 | if (sysfs_streq(val, "force") || sysfs_streq(val, "FORCE")) { | |
82 | mode = MEMMAP_ON_MEMORY_FORCE; | |
83 | } else { | |
84 | ret = kstrtobool(val, &enabled); | |
85 | if (ret < 0) | |
86 | return ret; | |
87 | if (enabled) | |
88 | mode = MEMMAP_ON_MEMORY_ENABLE; | |
89 | else | |
90 | mode = MEMMAP_ON_MEMORY_DISABLE; | |
91 | } | |
92 | *((int *)kp->arg) = mode; | |
93 | if (mode == MEMMAP_ON_MEMORY_FORCE) { | |
94 | unsigned long memmap_pages = memory_block_memmap_on_memory_pages(); | |
95 | ||
96 | pr_info_once("Memory hotplug will waste %ld pages in each memory block\n", | |
97 | memmap_pages - PFN_UP(memory_block_memmap_size())); | |
98 | } | |
99 | return 0; | |
100 | } | |
101 | ||
102 | static int get_memmap_mode(char *buffer, const struct kernel_param *kp) | |
103 | { | |
104 | if (*((int *)kp->arg) == MEMMAP_ON_MEMORY_FORCE) | |
105 | return sprintf(buffer, "force\n"); | |
106 | return param_get_bool(buffer, kp); | |
107 | } | |
108 | ||
109 | static const struct kernel_param_ops memmap_mode_ops = { | |
110 | .set = set_memmap_mode, | |
111 | .get = get_memmap_mode, | |
112 | }; | |
113 | module_param_cb(memmap_on_memory, &memmap_mode_ops, &memmap_mode, 0444); | |
114 | MODULE_PARM_DESC(memmap_on_memory, "Enable memmap on memory for memory hotplug\n" | |
115 | "With value \"force\" it could result in memory wastage due " | |
116 | "to memmap size limitations (Y/N/force)"); | |
6e02c46b | 117 | |
66361095 | 118 | static inline bool mhp_memmap_on_memory(void) |
6e02c46b | 119 | { |
2d1f649c | 120 | return memmap_mode != MEMMAP_ON_MEMORY_DISABLE; |
6e02c46b | 121 | } |
66361095 MS |
122 | #else |
123 | static inline bool mhp_memmap_on_memory(void) | |
124 | { | |
125 | return false; | |
126 | } | |
e3a9d9fc | 127 | #endif |
a08a2ae3 | 128 | |
e83a437f DH |
129 | enum { |
130 | ONLINE_POLICY_CONTIG_ZONES = 0, | |
131 | ONLINE_POLICY_AUTO_MOVABLE, | |
132 | }; | |
133 | ||
ac62554b | 134 | static const char * const online_policy_to_str[] = { |
e83a437f DH |
135 | [ONLINE_POLICY_CONTIG_ZONES] = "contig-zones", |
136 | [ONLINE_POLICY_AUTO_MOVABLE] = "auto-movable", | |
137 | }; | |
138 | ||
139 | static int set_online_policy(const char *val, const struct kernel_param *kp) | |
140 | { | |
141 | int ret = sysfs_match_string(online_policy_to_str, val); | |
142 | ||
143 | if (ret < 0) | |
144 | return ret; | |
145 | *((int *)kp->arg) = ret; | |
146 | return 0; | |
147 | } | |
148 | ||
149 | static int get_online_policy(char *buffer, const struct kernel_param *kp) | |
150 | { | |
151 | return sprintf(buffer, "%s\n", online_policy_to_str[*((int *)kp->arg)]); | |
152 | } | |
153 | ||
154 | /* | |
155 | * memory_hotplug.online_policy: configure online behavior when onlining without | |
156 | * specifying a zone (MMOP_ONLINE) | |
157 | * | |
158 | * "contig-zones": keep zone contiguous | |
159 | * "auto-movable": online memory to ZONE_MOVABLE if the configuration | |
160 | * (auto_movable_ratio, auto_movable_numa_aware) allows for it | |
161 | */ | |
162 | static int online_policy __read_mostly = ONLINE_POLICY_CONTIG_ZONES; | |
163 | static const struct kernel_param_ops online_policy_ops = { | |
164 | .set = set_online_policy, | |
165 | .get = get_online_policy, | |
166 | }; | |
167 | module_param_cb(online_policy, &online_policy_ops, &online_policy, 0644); | |
168 | MODULE_PARM_DESC(online_policy, | |
169 | "Set the online policy (\"contig-zones\", \"auto-movable\") " | |
170 | "Default: \"contig-zones\""); | |
171 | ||
172 | /* | |
173 | * memory_hotplug.auto_movable_ratio: specify maximum MOVABLE:KERNEL ratio | |
174 | * | |
175 | * The ratio represent an upper limit and the kernel might decide to not | |
176 | * online some memory to ZONE_MOVABLE -- e.g., because hotplugged KERNEL memory | |
177 | * doesn't allow for more MOVABLE memory. | |
178 | */ | |
179 | static unsigned int auto_movable_ratio __read_mostly = 301; | |
180 | module_param(auto_movable_ratio, uint, 0644); | |
181 | MODULE_PARM_DESC(auto_movable_ratio, | |
182 | "Set the maximum ratio of MOVABLE:KERNEL memory in the system " | |
183 | "in percent for \"auto-movable\" online policy. Default: 301"); | |
184 | ||
185 | /* | |
186 | * memory_hotplug.auto_movable_numa_aware: consider numa node stats | |
187 | */ | |
188 | #ifdef CONFIG_NUMA | |
189 | static bool auto_movable_numa_aware __read_mostly = true; | |
190 | module_param(auto_movable_numa_aware, bool, 0644); | |
191 | MODULE_PARM_DESC(auto_movable_numa_aware, | |
192 | "Consider numa node stats in addition to global stats in " | |
193 | "\"auto-movable\" online policy. Default: true"); | |
194 | #endif /* CONFIG_NUMA */ | |
195 | ||
9d0ad8ca DK |
196 | /* |
197 | * online_page_callback contains pointer to current page onlining function. | |
198 | * Initially it is generic_online_page(). If it is required it could be | |
199 | * changed by calling set_online_page_callback() for callback registration | |
200 | * and restore_online_page_callback() for generic callback restore. | |
201 | */ | |
202 | ||
9d0ad8ca | 203 | static online_page_callback_t online_page_callback = generic_online_page; |
bfc8c901 | 204 | static DEFINE_MUTEX(online_page_callback_lock); |
9d0ad8ca | 205 | |
3f906ba2 | 206 | DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock); |
bfc8c901 | 207 | |
3f906ba2 TG |
208 | void get_online_mems(void) |
209 | { | |
210 | percpu_down_read(&mem_hotplug_lock); | |
211 | } | |
bfc8c901 | 212 | |
3f906ba2 TG |
213 | void put_online_mems(void) |
214 | { | |
215 | percpu_up_read(&mem_hotplug_lock); | |
216 | } | |
bfc8c901 | 217 | |
4932381e MH |
218 | bool movable_node_enabled = false; |
219 | ||
8604d9e5 | 220 | #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE |
1adf8b46 | 221 | int mhp_default_online_type = MMOP_OFFLINE; |
8604d9e5 | 222 | #else |
1adf8b46 | 223 | int mhp_default_online_type = MMOP_ONLINE; |
8604d9e5 | 224 | #endif |
31bc3858 | 225 | |
86dd995d VK |
226 | static int __init setup_memhp_default_state(char *str) |
227 | { | |
1adf8b46 | 228 | const int online_type = mhp_online_type_from_str(str); |
5f47adf7 DH |
229 | |
230 | if (online_type >= 0) | |
1adf8b46 | 231 | mhp_default_online_type = online_type; |
86dd995d VK |
232 | |
233 | return 1; | |
234 | } | |
235 | __setup("memhp_default_state=", setup_memhp_default_state); | |
236 | ||
30467e0b | 237 | void mem_hotplug_begin(void) |
20d6c96b | 238 | { |
3f906ba2 TG |
239 | cpus_read_lock(); |
240 | percpu_down_write(&mem_hotplug_lock); | |
20d6c96b KM |
241 | } |
242 | ||
30467e0b | 243 | void mem_hotplug_done(void) |
bfc8c901 | 244 | { |
3f906ba2 TG |
245 | percpu_up_write(&mem_hotplug_lock); |
246 | cpus_read_unlock(); | |
bfc8c901 | 247 | } |
20d6c96b | 248 | |
357b4da5 JG |
249 | u64 max_mem_size = U64_MAX; |
250 | ||
45e0b78b | 251 | /* add this memory to iomem resource */ |
7b7b2721 DH |
252 | static struct resource *register_memory_resource(u64 start, u64 size, |
253 | const char *resource_name) | |
45e0b78b | 254 | { |
2794129e DH |
255 | struct resource *res; |
256 | unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY; | |
7b7b2721 DH |
257 | |
258 | if (strcmp(resource_name, "System RAM")) | |
7cf603d1 | 259 | flags |= IORESOURCE_SYSRAM_DRIVER_MANAGED; |
357b4da5 | 260 | |
bca3feaa AK |
261 | if (!mhp_range_allowed(start, size, true)) |
262 | return ERR_PTR(-E2BIG); | |
263 | ||
f3cd4c86 BH |
264 | /* |
265 | * Make sure value parsed from 'mem=' only restricts memory adding | |
266 | * while booting, so that memory hotplug won't be impacted. Please | |
267 | * refer to document of 'mem=' in kernel-parameters.txt for more | |
268 | * details. | |
269 | */ | |
270 | if (start + size > max_mem_size && system_state < SYSTEM_RUNNING) | |
357b4da5 JG |
271 | return ERR_PTR(-E2BIG); |
272 | ||
2794129e DH |
273 | /* |
274 | * Request ownership of the new memory range. This might be | |
275 | * a child of an existing resource that was present but | |
276 | * not marked as busy. | |
277 | */ | |
278 | res = __request_region(&iomem_resource, start, size, | |
279 | resource_name, flags); | |
280 | ||
281 | if (!res) { | |
282 | pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n", | |
283 | start, start + size); | |
6f754ba4 | 284 | return ERR_PTR(-EEXIST); |
45e0b78b KM |
285 | } |
286 | return res; | |
287 | } | |
288 | ||
289 | static void release_memory_resource(struct resource *res) | |
290 | { | |
291 | if (!res) | |
292 | return; | |
293 | release_resource(res); | |
294 | kfree(res); | |
45e0b78b KM |
295 | } |
296 | ||
943189db | 297 | static int check_pfn_span(unsigned long pfn, unsigned long nr_pages) |
7ea62160 DW |
298 | { |
299 | /* | |
300 | * Disallow all operations smaller than a sub-section and only | |
301 | * allow operations smaller than a section for | |
302 | * SPARSEMEM_VMEMMAP. Note that check_hotplug_memory_range() | |
303 | * enforces a larger memory_block_size_bytes() granularity for | |
304 | * memory that will be marked online, so this check should only | |
305 | * fire for direct arch_{add,remove}_memory() users outside of | |
306 | * add_memory_resource(). | |
307 | */ | |
308 | unsigned long min_align; | |
309 | ||
310 | if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP)) | |
311 | min_align = PAGES_PER_SUBSECTION; | |
312 | else | |
313 | min_align = PAGES_PER_SECTION; | |
943189db | 314 | if (!IS_ALIGNED(pfn | nr_pages, min_align)) |
7ea62160 | 315 | return -EINVAL; |
7ea62160 DW |
316 | return 0; |
317 | } | |
318 | ||
9f605f26 DW |
319 | /* |
320 | * Return page for the valid pfn only if the page is online. All pfn | |
321 | * walkers which rely on the fully initialized page->flags and others | |
322 | * should use this rather than pfn_valid && pfn_to_page | |
323 | */ | |
324 | struct page *pfn_to_online_page(unsigned long pfn) | |
325 | { | |
326 | unsigned long nr = pfn_to_section_nr(pfn); | |
1f90a347 | 327 | struct dev_pagemap *pgmap; |
9f9b02e5 DW |
328 | struct mem_section *ms; |
329 | ||
330 | if (nr >= NR_MEM_SECTIONS) | |
331 | return NULL; | |
332 | ||
333 | ms = __nr_to_section(nr); | |
334 | if (!online_section(ms)) | |
335 | return NULL; | |
336 | ||
337 | /* | |
338 | * Save some code text when online_section() + | |
339 | * pfn_section_valid() are sufficient. | |
340 | */ | |
341 | if (IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) && !pfn_valid(pfn)) | |
342 | return NULL; | |
343 | ||
344 | if (!pfn_section_valid(ms, pfn)) | |
345 | return NULL; | |
9f605f26 | 346 | |
1f90a347 DW |
347 | if (!online_device_section(ms)) |
348 | return pfn_to_page(pfn); | |
349 | ||
350 | /* | |
351 | * Slowpath: when ZONE_DEVICE collides with | |
352 | * ZONE_{NORMAL,MOVABLE} within the same section some pfns in | |
353 | * the section may be 'offline' but 'valid'. Only | |
354 | * get_dev_pagemap() can determine sub-section online status. | |
355 | */ | |
356 | pgmap = get_dev_pagemap(pfn, NULL); | |
357 | put_dev_pagemap(pgmap); | |
358 | ||
359 | /* The presence of a pgmap indicates ZONE_DEVICE offline pfn */ | |
360 | if (pgmap) | |
361 | return NULL; | |
362 | ||
9f9b02e5 | 363 | return pfn_to_page(pfn); |
9f605f26 DW |
364 | } |
365 | EXPORT_SYMBOL_GPL(pfn_to_online_page); | |
366 | ||
7ea62160 | 367 | int __ref __add_pages(int nid, unsigned long pfn, unsigned long nr_pages, |
f5637d3b | 368 | struct mhp_params *params) |
4edd7cef | 369 | { |
6cdd0b30 DH |
370 | const unsigned long end_pfn = pfn + nr_pages; |
371 | unsigned long cur_nr_pages; | |
9a845030 | 372 | int err; |
f5637d3b | 373 | struct vmem_altmap *altmap = params->altmap; |
4b94ffdc | 374 | |
6366238b | 375 | if (WARN_ON_ONCE(!pgprot_val(params->pgprot))) |
bfeb022f LG |
376 | return -EINVAL; |
377 | ||
bca3feaa | 378 | VM_BUG_ON(!mhp_range_allowed(PFN_PHYS(pfn), nr_pages * PAGE_SIZE, false)); |
dca4436d | 379 | |
4b94ffdc DW |
380 | if (altmap) { |
381 | /* | |
382 | * Validate altmap is within bounds of the total request | |
383 | */ | |
7ea62160 | 384 | if (altmap->base_pfn != pfn |
4b94ffdc DW |
385 | || vmem_altmap_offset(altmap) > nr_pages) { |
386 | pr_warn_once("memory add fail, invalid altmap\n"); | |
7ea62160 | 387 | return -EINVAL; |
4b94ffdc DW |
388 | } |
389 | altmap->alloc = 0; | |
390 | } | |
391 | ||
943189db | 392 | if (check_pfn_span(pfn, nr_pages)) { |
50135045 | 393 | WARN(1, "Misaligned %s start: %#lx end: %#lx\n", __func__, pfn, pfn + nr_pages - 1); |
943189db AK |
394 | return -EINVAL; |
395 | } | |
7ea62160 | 396 | |
6cdd0b30 DH |
397 | for (; pfn < end_pfn; pfn += cur_nr_pages) { |
398 | /* Select all remaining pages up to the next section boundary */ | |
399 | cur_nr_pages = min(end_pfn - pfn, | |
400 | SECTION_ALIGN_UP(pfn + 1) - pfn); | |
e3246d8f JM |
401 | err = sparse_add_section(nid, pfn, cur_nr_pages, altmap, |
402 | params->pgmap); | |
ba72b4c8 DW |
403 | if (err) |
404 | break; | |
f64ac5e6 | 405 | cond_resched(); |
4edd7cef | 406 | } |
c435a390 | 407 | vmemmap_populate_print_last(); |
4edd7cef DR |
408 | return err; |
409 | } | |
4edd7cef | 410 | |
815121d2 | 411 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ |
d09b0137 | 412 | static unsigned long find_smallest_section_pfn(int nid, struct zone *zone, |
815121d2 YI |
413 | unsigned long start_pfn, |
414 | unsigned long end_pfn) | |
415 | { | |
49ba3c6b | 416 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) { |
7ce700bf | 417 | if (unlikely(!pfn_to_online_page(start_pfn))) |
815121d2 YI |
418 | continue; |
419 | ||
420 | if (unlikely(pfn_to_nid(start_pfn) != nid)) | |
421 | continue; | |
422 | ||
9b05158f | 423 | if (zone != page_zone(pfn_to_page(start_pfn))) |
815121d2 YI |
424 | continue; |
425 | ||
426 | return start_pfn; | |
427 | } | |
428 | ||
429 | return 0; | |
430 | } | |
431 | ||
432 | /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ | |
d09b0137 | 433 | static unsigned long find_biggest_section_pfn(int nid, struct zone *zone, |
815121d2 YI |
434 | unsigned long start_pfn, |
435 | unsigned long end_pfn) | |
436 | { | |
815121d2 YI |
437 | unsigned long pfn; |
438 | ||
439 | /* pfn is the end pfn of a memory section. */ | |
440 | pfn = end_pfn - 1; | |
49ba3c6b | 441 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) { |
7ce700bf | 442 | if (unlikely(!pfn_to_online_page(pfn))) |
815121d2 YI |
443 | continue; |
444 | ||
445 | if (unlikely(pfn_to_nid(pfn) != nid)) | |
446 | continue; | |
447 | ||
9b05158f | 448 | if (zone != page_zone(pfn_to_page(pfn))) |
815121d2 YI |
449 | continue; |
450 | ||
451 | return pfn; | |
452 | } | |
453 | ||
454 | return 0; | |
455 | } | |
456 | ||
457 | static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, | |
458 | unsigned long end_pfn) | |
459 | { | |
815121d2 | 460 | unsigned long pfn; |
815121d2 YI |
461 | int nid = zone_to_nid(zone); |
462 | ||
5d12071c | 463 | if (zone->zone_start_pfn == start_pfn) { |
815121d2 YI |
464 | /* |
465 | * If the section is smallest section in the zone, it need | |
466 | * shrink zone->zone_start_pfn and zone->zone_spanned_pages. | |
467 | * In this case, we find second smallest valid mem_section | |
468 | * for shrinking zone. | |
469 | */ | |
470 | pfn = find_smallest_section_pfn(nid, zone, end_pfn, | |
5d12071c | 471 | zone_end_pfn(zone)); |
815121d2 | 472 | if (pfn) { |
5d12071c | 473 | zone->spanned_pages = zone_end_pfn(zone) - pfn; |
815121d2 | 474 | zone->zone_start_pfn = pfn; |
950b68d9 DH |
475 | } else { |
476 | zone->zone_start_pfn = 0; | |
477 | zone->spanned_pages = 0; | |
815121d2 | 478 | } |
5d12071c | 479 | } else if (zone_end_pfn(zone) == end_pfn) { |
815121d2 YI |
480 | /* |
481 | * If the section is biggest section in the zone, it need | |
482 | * shrink zone->spanned_pages. | |
483 | * In this case, we find second biggest valid mem_section for | |
484 | * shrinking zone. | |
485 | */ | |
5d12071c | 486 | pfn = find_biggest_section_pfn(nid, zone, zone->zone_start_pfn, |
815121d2 YI |
487 | start_pfn); |
488 | if (pfn) | |
5d12071c | 489 | zone->spanned_pages = pfn - zone->zone_start_pfn + 1; |
950b68d9 DH |
490 | else { |
491 | zone->zone_start_pfn = 0; | |
492 | zone->spanned_pages = 0; | |
493 | } | |
815121d2 | 494 | } |
815121d2 YI |
495 | } |
496 | ||
00d6c019 | 497 | static void update_pgdat_span(struct pglist_data *pgdat) |
815121d2 | 498 | { |
00d6c019 DH |
499 | unsigned long node_start_pfn = 0, node_end_pfn = 0; |
500 | struct zone *zone; | |
501 | ||
502 | for (zone = pgdat->node_zones; | |
503 | zone < pgdat->node_zones + MAX_NR_ZONES; zone++) { | |
6c922cf7 | 504 | unsigned long end_pfn = zone_end_pfn(zone); |
00d6c019 DH |
505 | |
506 | /* No need to lock the zones, they can't change. */ | |
656d5711 DH |
507 | if (!zone->spanned_pages) |
508 | continue; | |
509 | if (!node_end_pfn) { | |
510 | node_start_pfn = zone->zone_start_pfn; | |
6c922cf7 | 511 | node_end_pfn = end_pfn; |
656d5711 DH |
512 | continue; |
513 | } | |
514 | ||
6c922cf7 ML |
515 | if (end_pfn > node_end_pfn) |
516 | node_end_pfn = end_pfn; | |
00d6c019 DH |
517 | if (zone->zone_start_pfn < node_start_pfn) |
518 | node_start_pfn = zone->zone_start_pfn; | |
815121d2 YI |
519 | } |
520 | ||
00d6c019 DH |
521 | pgdat->node_start_pfn = node_start_pfn; |
522 | pgdat->node_spanned_pages = node_end_pfn - node_start_pfn; | |
815121d2 YI |
523 | } |
524 | ||
feee6b29 DH |
525 | void __ref remove_pfn_range_from_zone(struct zone *zone, |
526 | unsigned long start_pfn, | |
527 | unsigned long nr_pages) | |
815121d2 | 528 | { |
b7e3debd | 529 | const unsigned long end_pfn = start_pfn + nr_pages; |
815121d2 | 530 | struct pglist_data *pgdat = zone->zone_pgdat; |
27cacaad | 531 | unsigned long pfn, cur_nr_pages; |
815121d2 | 532 | |
d33695b1 | 533 | /* Poison struct pages because they are now uninitialized again. */ |
b7e3debd BW |
534 | for (pfn = start_pfn; pfn < end_pfn; pfn += cur_nr_pages) { |
535 | cond_resched(); | |
536 | ||
537 | /* Select all remaining pages up to the next section boundary */ | |
538 | cur_nr_pages = | |
539 | min(end_pfn - pfn, SECTION_ALIGN_UP(pfn + 1) - pfn); | |
540 | page_init_poison(pfn_to_page(pfn), | |
541 | sizeof(struct page) * cur_nr_pages); | |
542 | } | |
d33695b1 | 543 | |
7ce700bf DH |
544 | /* |
545 | * Zone shrinking code cannot properly deal with ZONE_DEVICE. So | |
546 | * we will not try to shrink the zones - which is okay as | |
547 | * set_zone_contiguous() cannot deal with ZONE_DEVICE either way. | |
548 | */ | |
5ef5f810 | 549 | if (zone_is_zone_device(zone)) |
7ce700bf | 550 | return; |
7ce700bf | 551 | |
feee6b29 DH |
552 | clear_zone_contiguous(zone); |
553 | ||
815121d2 | 554 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); |
00d6c019 | 555 | update_pgdat_span(pgdat); |
feee6b29 DH |
556 | |
557 | set_zone_contiguous(zone); | |
815121d2 YI |
558 | } |
559 | ||
ea01ea93 | 560 | /** |
feee6b29 | 561 | * __remove_pages() - remove sections of pages |
7ea62160 | 562 | * @pfn: starting pageframe (must be aligned to start of a section) |
ea01ea93 | 563 | * @nr_pages: number of pages to remove (must be multiple of section size) |
e8b098fc | 564 | * @altmap: alternative device page map or %NULL if default memmap is used |
ea01ea93 BP |
565 | * |
566 | * Generic helper function to remove section mappings and sysfs entries | |
567 | * for the section of the memory we are removing. Caller needs to make | |
568 | * sure that pages are marked reserved and zones are adjust properly by | |
569 | * calling offline_pages(). | |
570 | */ | |
feee6b29 DH |
571 | void __remove_pages(unsigned long pfn, unsigned long nr_pages, |
572 | struct vmem_altmap *altmap) | |
ea01ea93 | 573 | { |
52fb87c8 DH |
574 | const unsigned long end_pfn = pfn + nr_pages; |
575 | unsigned long cur_nr_pages; | |
ea01ea93 | 576 | |
943189db | 577 | if (check_pfn_span(pfn, nr_pages)) { |
50135045 | 578 | WARN(1, "Misaligned %s start: %#lx end: %#lx\n", __func__, pfn, pfn + nr_pages - 1); |
7ea62160 | 579 | return; |
943189db | 580 | } |
ea01ea93 | 581 | |
52fb87c8 | 582 | for (; pfn < end_pfn; pfn += cur_nr_pages) { |
dd33ad7b | 583 | cond_resched(); |
52fb87c8 | 584 | /* Select all remaining pages up to the next section boundary */ |
a11b9419 DH |
585 | cur_nr_pages = min(end_pfn - pfn, |
586 | SECTION_ALIGN_UP(pfn + 1) - pfn); | |
bd5f79ab | 587 | sparse_remove_section(pfn, cur_nr_pages, altmap); |
ea01ea93 | 588 | } |
ea01ea93 | 589 | } |
ea01ea93 | 590 | |
9d0ad8ca DK |
591 | int set_online_page_callback(online_page_callback_t callback) |
592 | { | |
593 | int rc = -EINVAL; | |
594 | ||
bfc8c901 VD |
595 | get_online_mems(); |
596 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
597 | |
598 | if (online_page_callback == generic_online_page) { | |
599 | online_page_callback = callback; | |
600 | rc = 0; | |
601 | } | |
602 | ||
bfc8c901 VD |
603 | mutex_unlock(&online_page_callback_lock); |
604 | put_online_mems(); | |
9d0ad8ca DK |
605 | |
606 | return rc; | |
607 | } | |
608 | EXPORT_SYMBOL_GPL(set_online_page_callback); | |
609 | ||
610 | int restore_online_page_callback(online_page_callback_t callback) | |
611 | { | |
612 | int rc = -EINVAL; | |
613 | ||
bfc8c901 VD |
614 | get_online_mems(); |
615 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
616 | |
617 | if (online_page_callback == callback) { | |
618 | online_page_callback = generic_online_page; | |
619 | rc = 0; | |
620 | } | |
621 | ||
bfc8c901 VD |
622 | mutex_unlock(&online_page_callback_lock); |
623 | put_online_mems(); | |
9d0ad8ca DK |
624 | |
625 | return rc; | |
626 | } | |
627 | EXPORT_SYMBOL_GPL(restore_online_page_callback); | |
628 | ||
18db1491 | 629 | void generic_online_page(struct page *page, unsigned int order) |
9d0ad8ca | 630 | { |
c87cbc1f VB |
631 | /* |
632 | * Freeing the page with debug_pagealloc enabled will try to unmap it, | |
633 | * so we should map it first. This is better than introducing a special | |
634 | * case in page freeing fast path. | |
635 | */ | |
77bc7fd6 | 636 | debug_pagealloc_map_pages(page, 1 << order); |
a9cd410a AK |
637 | __free_pages_core(page, order); |
638 | totalram_pages_add(1UL << order); | |
a9cd410a | 639 | } |
18db1491 | 640 | EXPORT_SYMBOL_GPL(generic_online_page); |
a9cd410a | 641 | |
aac65321 | 642 | static void online_pages_range(unsigned long start_pfn, unsigned long nr_pages) |
3947be19 | 643 | { |
b2c2ab20 DH |
644 | const unsigned long end_pfn = start_pfn + nr_pages; |
645 | unsigned long pfn; | |
b2c2ab20 DH |
646 | |
647 | /* | |
23baf831 | 648 | * Online the pages in MAX_ORDER aligned chunks. The callback might |
aac65321 DH |
649 | * decide to not expose all pages to the buddy (e.g., expose them |
650 | * later). We account all pages as being online and belonging to this | |
651 | * zone ("present"). | |
a08a2ae3 OS |
652 | * When using memmap_on_memory, the range might not be aligned to |
653 | * MAX_ORDER_NR_PAGES - 1, but pageblock aligned. __ffs() will detect | |
654 | * this and the first chunk to online will be pageblock_nr_pages. | |
b2c2ab20 | 655 | */ |
a08a2ae3 | 656 | for (pfn = start_pfn; pfn < end_pfn;) { |
59f876fb KS |
657 | int order; |
658 | ||
659 | /* | |
660 | * Free to online pages in the largest chunks alignment allows. | |
661 | * | |
662 | * __ffs() behaviour is undefined for 0. start == 0 is | |
663 | * MAX_ORDER-aligned, Set order to MAX_ORDER for the case. | |
664 | */ | |
665 | if (pfn) | |
666 | order = min_t(int, MAX_ORDER, __ffs(pfn)); | |
667 | else | |
668 | order = MAX_ORDER; | |
a08a2ae3 OS |
669 | |
670 | (*online_page_callback)(pfn_to_page(pfn), order); | |
671 | pfn += (1UL << order); | |
672 | } | |
2d070eab | 673 | |
b2c2ab20 DH |
674 | /* mark all involved sections as online */ |
675 | online_mem_sections(start_pfn, end_pfn); | |
75884fb1 KH |
676 | } |
677 | ||
d9713679 LJ |
678 | /* check which state of node_states will be changed when online memory */ |
679 | static void node_states_check_changes_online(unsigned long nr_pages, | |
680 | struct zone *zone, struct memory_notify *arg) | |
681 | { | |
682 | int nid = zone_to_nid(zone); | |
d9713679 | 683 | |
98fa15f3 AK |
684 | arg->status_change_nid = NUMA_NO_NODE; |
685 | arg->status_change_nid_normal = NUMA_NO_NODE; | |
d9713679 | 686 | |
8efe33f4 OS |
687 | if (!node_state(nid, N_MEMORY)) |
688 | arg->status_change_nid = nid; | |
689 | if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY)) | |
d9713679 | 690 | arg->status_change_nid_normal = nid; |
d9713679 LJ |
691 | } |
692 | ||
693 | static void node_states_set_node(int node, struct memory_notify *arg) | |
694 | { | |
695 | if (arg->status_change_nid_normal >= 0) | |
696 | node_set_state(node, N_NORMAL_MEMORY); | |
697 | ||
83d83612 OS |
698 | if (arg->status_change_nid >= 0) |
699 | node_set_state(node, N_MEMORY); | |
d9713679 LJ |
700 | } |
701 | ||
f1dd2cd1 MH |
702 | static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn, |
703 | unsigned long nr_pages) | |
704 | { | |
705 | unsigned long old_end_pfn = zone_end_pfn(zone); | |
706 | ||
707 | if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn) | |
708 | zone->zone_start_pfn = start_pfn; | |
709 | ||
710 | zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn; | |
711 | } | |
712 | ||
713 | static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn, | |
714 | unsigned long nr_pages) | |
715 | { | |
716 | unsigned long old_end_pfn = pgdat_end_pfn(pgdat); | |
717 | ||
718 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) | |
719 | pgdat->node_start_pfn = start_pfn; | |
720 | ||
721 | pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn; | |
f1dd2cd1 | 722 | |
3fccb74c | 723 | } |
1f90a347 | 724 | |
ed7802dd | 725 | #ifdef CONFIG_ZONE_DEVICE |
1f90a347 DW |
726 | static void section_taint_zone_device(unsigned long pfn) |
727 | { | |
728 | struct mem_section *ms = __pfn_to_section(pfn); | |
729 | ||
730 | ms->section_mem_map |= SECTION_TAINT_ZONE_DEVICE; | |
731 | } | |
ed7802dd MS |
732 | #else |
733 | static inline void section_taint_zone_device(unsigned long pfn) | |
734 | { | |
735 | } | |
736 | #endif | |
1f90a347 | 737 | |
3fccb74c DH |
738 | /* |
739 | * Associate the pfn range with the given zone, initializing the memmaps | |
740 | * and resizing the pgdat/zone data to span the added pages. After this | |
741 | * call, all affected pages are PG_reserved. | |
d882c006 DH |
742 | * |
743 | * All aligned pageblocks are initialized to the specified migratetype | |
744 | * (usually MIGRATE_MOVABLE). Besides setting the migratetype, no related | |
745 | * zone stats (e.g., nr_isolate_pageblock) are touched. | |
3fccb74c | 746 | */ |
a99583e7 | 747 | void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn, |
d882c006 DH |
748 | unsigned long nr_pages, |
749 | struct vmem_altmap *altmap, int migratetype) | |
f1dd2cd1 MH |
750 | { |
751 | struct pglist_data *pgdat = zone->zone_pgdat; | |
752 | int nid = pgdat->node_id; | |
df429ac0 | 753 | |
f1dd2cd1 MH |
754 | clear_zone_contiguous(zone); |
755 | ||
fa004ab7 WY |
756 | if (zone_is_empty(zone)) |
757 | init_currently_empty_zone(zone, start_pfn, nr_pages); | |
f1dd2cd1 | 758 | resize_zone_range(zone, start_pfn, nr_pages); |
f1dd2cd1 | 759 | resize_pgdat_range(pgdat, start_pfn, nr_pages); |
f1dd2cd1 | 760 | |
1f90a347 DW |
761 | /* |
762 | * Subsection population requires care in pfn_to_online_page(). | |
763 | * Set the taint to enable the slow path detection of | |
764 | * ZONE_DEVICE pages in an otherwise ZONE_{NORMAL,MOVABLE} | |
765 | * section. | |
766 | */ | |
767 | if (zone_is_zone_device(zone)) { | |
768 | if (!IS_ALIGNED(start_pfn, PAGES_PER_SECTION)) | |
769 | section_taint_zone_device(start_pfn); | |
770 | if (!IS_ALIGNED(start_pfn + nr_pages, PAGES_PER_SECTION)) | |
771 | section_taint_zone_device(start_pfn + nr_pages); | |
772 | } | |
773 | ||
f1dd2cd1 MH |
774 | /* |
775 | * TODO now we have a visible range of pages which are not associated | |
776 | * with their zone properly. Not nice but set_pfnblock_flags_mask | |
777 | * expects the zone spans the pfn range. All the pages in the range | |
778 | * are reserved so nobody should be touching them so we should be safe | |
779 | */ | |
ab28cb6e | 780 | memmap_init_range(nr_pages, nid, zone_idx(zone), start_pfn, 0, |
d882c006 | 781 | MEMINIT_HOTPLUG, altmap, migratetype); |
f1dd2cd1 MH |
782 | |
783 | set_zone_contiguous(zone); | |
784 | } | |
785 | ||
e83a437f DH |
786 | struct auto_movable_stats { |
787 | unsigned long kernel_early_pages; | |
788 | unsigned long movable_pages; | |
789 | }; | |
790 | ||
791 | static void auto_movable_stats_account_zone(struct auto_movable_stats *stats, | |
792 | struct zone *zone) | |
793 | { | |
794 | if (zone_idx(zone) == ZONE_MOVABLE) { | |
795 | stats->movable_pages += zone->present_pages; | |
796 | } else { | |
797 | stats->kernel_early_pages += zone->present_early_pages; | |
798 | #ifdef CONFIG_CMA | |
799 | /* | |
800 | * CMA pages (never on hotplugged memory) behave like | |
801 | * ZONE_MOVABLE. | |
802 | */ | |
803 | stats->movable_pages += zone->cma_pages; | |
804 | stats->kernel_early_pages -= zone->cma_pages; | |
805 | #endif /* CONFIG_CMA */ | |
806 | } | |
807 | } | |
3fcebf90 DH |
808 | struct auto_movable_group_stats { |
809 | unsigned long movable_pages; | |
810 | unsigned long req_kernel_early_pages; | |
811 | }; | |
e83a437f | 812 | |
3fcebf90 DH |
813 | static int auto_movable_stats_account_group(struct memory_group *group, |
814 | void *arg) | |
815 | { | |
816 | const int ratio = READ_ONCE(auto_movable_ratio); | |
817 | struct auto_movable_group_stats *stats = arg; | |
818 | long pages; | |
819 | ||
820 | /* | |
821 | * We don't support modifying the config while the auto-movable online | |
822 | * policy is already enabled. Just avoid the division by zero below. | |
823 | */ | |
824 | if (!ratio) | |
825 | return 0; | |
826 | ||
827 | /* | |
828 | * Calculate how many early kernel pages this group requires to | |
829 | * satisfy the configured zone ratio. | |
830 | */ | |
831 | pages = group->present_movable_pages * 100 / ratio; | |
832 | pages -= group->present_kernel_pages; | |
833 | ||
834 | if (pages > 0) | |
835 | stats->req_kernel_early_pages += pages; | |
836 | stats->movable_pages += group->present_movable_pages; | |
837 | return 0; | |
838 | } | |
839 | ||
840 | static bool auto_movable_can_online_movable(int nid, struct memory_group *group, | |
841 | unsigned long nr_pages) | |
e83a437f | 842 | { |
e83a437f | 843 | unsigned long kernel_early_pages, movable_pages; |
3fcebf90 DH |
844 | struct auto_movable_group_stats group_stats = {}; |
845 | struct auto_movable_stats stats = {}; | |
e83a437f DH |
846 | pg_data_t *pgdat = NODE_DATA(nid); |
847 | struct zone *zone; | |
848 | int i; | |
849 | ||
850 | /* Walk all relevant zones and collect MOVABLE vs. KERNEL stats. */ | |
851 | if (nid == NUMA_NO_NODE) { | |
852 | /* TODO: cache values */ | |
853 | for_each_populated_zone(zone) | |
854 | auto_movable_stats_account_zone(&stats, zone); | |
855 | } else { | |
856 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
857 | zone = pgdat->node_zones + i; | |
858 | if (populated_zone(zone)) | |
859 | auto_movable_stats_account_zone(&stats, zone); | |
860 | } | |
861 | } | |
862 | ||
863 | kernel_early_pages = stats.kernel_early_pages; | |
864 | movable_pages = stats.movable_pages; | |
865 | ||
3fcebf90 DH |
866 | /* |
867 | * Kernel memory inside dynamic memory group allows for more MOVABLE | |
868 | * memory within the same group. Remove the effect of all but the | |
869 | * current group from the stats. | |
870 | */ | |
871 | walk_dynamic_memory_groups(nid, auto_movable_stats_account_group, | |
872 | group, &group_stats); | |
873 | if (kernel_early_pages <= group_stats.req_kernel_early_pages) | |
874 | return false; | |
875 | kernel_early_pages -= group_stats.req_kernel_early_pages; | |
876 | movable_pages -= group_stats.movable_pages; | |
877 | ||
878 | if (group && group->is_dynamic) | |
879 | kernel_early_pages += group->present_kernel_pages; | |
880 | ||
e83a437f DH |
881 | /* |
882 | * Test if we could online the given number of pages to ZONE_MOVABLE | |
883 | * and still stay in the configured ratio. | |
884 | */ | |
885 | movable_pages += nr_pages; | |
886 | return movable_pages <= (auto_movable_ratio * kernel_early_pages) / 100; | |
887 | } | |
888 | ||
c246a213 MH |
889 | /* |
890 | * Returns a default kernel memory zone for the given pfn range. | |
891 | * If no kernel zone covers this pfn range it will automatically go | |
892 | * to the ZONE_NORMAL. | |
893 | */ | |
c6f03e29 | 894 | static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn, |
c246a213 MH |
895 | unsigned long nr_pages) |
896 | { | |
897 | struct pglist_data *pgdat = NODE_DATA(nid); | |
898 | int zid; | |
899 | ||
d6aad201 | 900 | for (zid = 0; zid < ZONE_NORMAL; zid++) { |
c246a213 MH |
901 | struct zone *zone = &pgdat->node_zones[zid]; |
902 | ||
903 | if (zone_intersects(zone, start_pfn, nr_pages)) | |
904 | return zone; | |
905 | } | |
906 | ||
907 | return &pgdat->node_zones[ZONE_NORMAL]; | |
908 | } | |
909 | ||
e83a437f DH |
910 | /* |
911 | * Determine to which zone to online memory dynamically based on user | |
912 | * configuration and system stats. We care about the following ratio: | |
913 | * | |
914 | * MOVABLE : KERNEL | |
915 | * | |
916 | * Whereby MOVABLE is memory in ZONE_MOVABLE and KERNEL is memory in | |
917 | * one of the kernel zones. CMA pages inside one of the kernel zones really | |
918 | * behaves like ZONE_MOVABLE, so we treat them accordingly. | |
919 | * | |
920 | * We don't allow for hotplugged memory in a KERNEL zone to increase the | |
921 | * amount of MOVABLE memory we can have, so we end up with: | |
922 | * | |
923 | * MOVABLE : KERNEL_EARLY | |
924 | * | |
925 | * Whereby KERNEL_EARLY is memory in one of the kernel zones, available sinze | |
926 | * boot. We base our calculation on KERNEL_EARLY internally, because: | |
927 | * | |
928 | * a) Hotplugged memory in one of the kernel zones can sometimes still get | |
929 | * hotunplugged, especially when hot(un)plugging individual memory blocks. | |
930 | * There is no coordination across memory devices, therefore "automatic" | |
931 | * hotunplugging, as implemented in hypervisors, could result in zone | |
932 | * imbalances. | |
933 | * b) Early/boot memory in one of the kernel zones can usually not get | |
934 | * hotunplugged again (e.g., no firmware interface to unplug, fragmented | |
935 | * with unmovable allocations). While there are corner cases where it might | |
936 | * still work, it is barely relevant in practice. | |
937 | * | |
3fcebf90 DH |
938 | * Exceptions are dynamic memory groups, which allow for more MOVABLE |
939 | * memory within the same memory group -- because in that case, there is | |
940 | * coordination within the single memory device managed by a single driver. | |
941 | * | |
e83a437f DH |
942 | * We rely on "present pages" instead of "managed pages", as the latter is |
943 | * highly unreliable and dynamic in virtualized environments, and does not | |
944 | * consider boot time allocations. For example, memory ballooning adjusts the | |
945 | * managed pages when inflating/deflating the balloon, and balloon compaction | |
946 | * can even migrate inflated pages between zones. | |
947 | * | |
948 | * Using "present pages" is better but some things to keep in mind are: | |
949 | * | |
950 | * a) Some memblock allocations, such as for the crashkernel area, are | |
951 | * effectively unused by the kernel, yet they account to "present pages". | |
952 | * Fortunately, these allocations are comparatively small in relevant setups | |
953 | * (e.g., fraction of system memory). | |
954 | * b) Some hotplugged memory blocks in virtualized environments, esecially | |
955 | * hotplugged by virtio-mem, look like they are completely present, however, | |
956 | * only parts of the memory block are actually currently usable. | |
957 | * "present pages" is an upper limit that can get reached at runtime. As | |
958 | * we base our calculations on KERNEL_EARLY, this is not an issue. | |
959 | */ | |
445fcf7c DH |
960 | static struct zone *auto_movable_zone_for_pfn(int nid, |
961 | struct memory_group *group, | |
962 | unsigned long pfn, | |
e83a437f DH |
963 | unsigned long nr_pages) |
964 | { | |
445fcf7c DH |
965 | unsigned long online_pages = 0, max_pages, end_pfn; |
966 | struct page *page; | |
967 | ||
e83a437f DH |
968 | if (!auto_movable_ratio) |
969 | goto kernel_zone; | |
970 | ||
445fcf7c DH |
971 | if (group && !group->is_dynamic) { |
972 | max_pages = group->s.max_pages; | |
973 | online_pages = group->present_movable_pages; | |
974 | ||
975 | /* If anything is !MOVABLE online the rest !MOVABLE. */ | |
976 | if (group->present_kernel_pages) | |
977 | goto kernel_zone; | |
978 | } else if (!group || group->d.unit_pages == nr_pages) { | |
979 | max_pages = nr_pages; | |
980 | } else { | |
981 | max_pages = group->d.unit_pages; | |
982 | /* | |
983 | * Take a look at all online sections in the current unit. | |
984 | * We can safely assume that all pages within a section belong | |
985 | * to the same zone, because dynamic memory groups only deal | |
986 | * with hotplugged memory. | |
987 | */ | |
988 | pfn = ALIGN_DOWN(pfn, group->d.unit_pages); | |
989 | end_pfn = pfn + group->d.unit_pages; | |
990 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
991 | page = pfn_to_online_page(pfn); | |
992 | if (!page) | |
993 | continue; | |
994 | /* If anything is !MOVABLE online the rest !MOVABLE. */ | |
07252dfe | 995 | if (!is_zone_movable_page(page)) |
445fcf7c DH |
996 | goto kernel_zone; |
997 | online_pages += PAGES_PER_SECTION; | |
998 | } | |
999 | } | |
1000 | ||
1001 | /* | |
1002 | * Online MOVABLE if we could *currently* online all remaining parts | |
1003 | * MOVABLE. We expect to (add+) online them immediately next, so if | |
1004 | * nobody interferes, all will be MOVABLE if possible. | |
1005 | */ | |
1006 | nr_pages = max_pages - online_pages; | |
3fcebf90 | 1007 | if (!auto_movable_can_online_movable(NUMA_NO_NODE, group, nr_pages)) |
e83a437f DH |
1008 | goto kernel_zone; |
1009 | ||
1010 | #ifdef CONFIG_NUMA | |
1011 | if (auto_movable_numa_aware && | |
3fcebf90 | 1012 | !auto_movable_can_online_movable(nid, group, nr_pages)) |
e83a437f DH |
1013 | goto kernel_zone; |
1014 | #endif /* CONFIG_NUMA */ | |
1015 | ||
1016 | return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
1017 | kernel_zone: | |
1018 | return default_kernel_zone_for_pfn(nid, pfn, nr_pages); | |
1019 | } | |
1020 | ||
c6f03e29 MH |
1021 | static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn, |
1022 | unsigned long nr_pages) | |
e5e68930 | 1023 | { |
c6f03e29 MH |
1024 | struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn, |
1025 | nr_pages); | |
1026 | struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
1027 | bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages); | |
1028 | bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages); | |
e5e68930 MH |
1029 | |
1030 | /* | |
c6f03e29 MH |
1031 | * We inherit the existing zone in a simple case where zones do not |
1032 | * overlap in the given range | |
e5e68930 | 1033 | */ |
c6f03e29 MH |
1034 | if (in_kernel ^ in_movable) |
1035 | return (in_kernel) ? kernel_zone : movable_zone; | |
9f123ab5 | 1036 | |
c6f03e29 MH |
1037 | /* |
1038 | * If the range doesn't belong to any zone or two zones overlap in the | |
1039 | * given range then we use movable zone only if movable_node is | |
1040 | * enabled because we always online to a kernel zone by default. | |
1041 | */ | |
1042 | return movable_node_enabled ? movable_zone : kernel_zone; | |
9f123ab5 MH |
1043 | } |
1044 | ||
7cf209ba | 1045 | struct zone *zone_for_pfn_range(int online_type, int nid, |
445fcf7c | 1046 | struct memory_group *group, unsigned long start_pfn, |
e5e68930 | 1047 | unsigned long nr_pages) |
f1dd2cd1 | 1048 | { |
c6f03e29 MH |
1049 | if (online_type == MMOP_ONLINE_KERNEL) |
1050 | return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages); | |
f1dd2cd1 | 1051 | |
c6f03e29 MH |
1052 | if (online_type == MMOP_ONLINE_MOVABLE) |
1053 | return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; | |
df429ac0 | 1054 | |
e83a437f | 1055 | if (online_policy == ONLINE_POLICY_AUTO_MOVABLE) |
445fcf7c | 1056 | return auto_movable_zone_for_pfn(nid, group, start_pfn, nr_pages); |
e83a437f | 1057 | |
c6f03e29 | 1058 | return default_zone_for_pfn(nid, start_pfn, nr_pages); |
e5e68930 MH |
1059 | } |
1060 | ||
a08a2ae3 OS |
1061 | /* |
1062 | * This function should only be called by memory_block_{online,offline}, | |
1063 | * and {online,offline}_pages. | |
1064 | */ | |
836809ec DH |
1065 | void adjust_present_page_count(struct page *page, struct memory_group *group, |
1066 | long nr_pages) | |
f9901144 | 1067 | { |
4b097002 | 1068 | struct zone *zone = page_zone(page); |
836809ec | 1069 | const bool movable = zone_idx(zone) == ZONE_MOVABLE; |
4b097002 DH |
1070 | |
1071 | /* | |
1072 | * We only support onlining/offlining/adding/removing of complete | |
1073 | * memory blocks; therefore, either all is either early or hotplugged. | |
1074 | */ | |
1075 | if (early_section(__pfn_to_section(page_to_pfn(page)))) | |
1076 | zone->present_early_pages += nr_pages; | |
f9901144 | 1077 | zone->present_pages += nr_pages; |
f9901144 | 1078 | zone->zone_pgdat->node_present_pages += nr_pages; |
836809ec DH |
1079 | |
1080 | if (group && movable) | |
1081 | group->present_movable_pages += nr_pages; | |
1082 | else if (group && !movable) | |
1083 | group->present_kernel_pages += nr_pages; | |
f9901144 DH |
1084 | } |
1085 | ||
a08a2ae3 OS |
1086 | int mhp_init_memmap_on_memory(unsigned long pfn, unsigned long nr_pages, |
1087 | struct zone *zone) | |
1088 | { | |
1089 | unsigned long end_pfn = pfn + nr_pages; | |
66361095 | 1090 | int ret, i; |
a08a2ae3 OS |
1091 | |
1092 | ret = kasan_add_zero_shadow(__va(PFN_PHYS(pfn)), PFN_PHYS(nr_pages)); | |
1093 | if (ret) | |
1094 | return ret; | |
1095 | ||
1096 | move_pfn_range_to_zone(zone, pfn, nr_pages, NULL, MIGRATE_UNMOVABLE); | |
1097 | ||
66361095 MS |
1098 | for (i = 0; i < nr_pages; i++) |
1099 | SetPageVmemmapSelfHosted(pfn_to_page(pfn + i)); | |
1100 | ||
a08a2ae3 OS |
1101 | /* |
1102 | * It might be that the vmemmap_pages fully span sections. If that is | |
1103 | * the case, mark those sections online here as otherwise they will be | |
1104 | * left offline. | |
1105 | */ | |
1106 | if (nr_pages >= PAGES_PER_SECTION) | |
1107 | online_mem_sections(pfn, ALIGN_DOWN(end_pfn, PAGES_PER_SECTION)); | |
1108 | ||
1109 | return ret; | |
1110 | } | |
1111 | ||
1112 | void mhp_deinit_memmap_on_memory(unsigned long pfn, unsigned long nr_pages) | |
1113 | { | |
1114 | unsigned long end_pfn = pfn + nr_pages; | |
1115 | ||
1116 | /* | |
1117 | * It might be that the vmemmap_pages fully span sections. If that is | |
1118 | * the case, mark those sections offline here as otherwise they will be | |
1119 | * left online. | |
1120 | */ | |
1121 | if (nr_pages >= PAGES_PER_SECTION) | |
1122 | offline_mem_sections(pfn, ALIGN_DOWN(end_pfn, PAGES_PER_SECTION)); | |
1123 | ||
1124 | /* | |
1125 | * The pages associated with this vmemmap have been offlined, so | |
1126 | * we can reset its state here. | |
1127 | */ | |
1128 | remove_pfn_range_from_zone(page_zone(pfn_to_page(pfn)), pfn, nr_pages); | |
1129 | kasan_remove_zero_shadow(__va(PFN_PHYS(pfn)), PFN_PHYS(nr_pages)); | |
1130 | } | |
1131 | ||
836809ec DH |
1132 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, |
1133 | struct zone *zone, struct memory_group *group) | |
75884fb1 | 1134 | { |
aa47228a | 1135 | unsigned long flags; |
6811378e | 1136 | int need_zonelists_rebuild = 0; |
a08a2ae3 | 1137 | const int nid = zone_to_nid(zone); |
7b78d335 YG |
1138 | int ret; |
1139 | struct memory_notify arg; | |
d0dc12e8 | 1140 | |
dd8e2f23 OS |
1141 | /* |
1142 | * {on,off}lining is constrained to full memory sections (or more | |
041711ce | 1143 | * precisely to memory blocks from the user space POV). |
dd8e2f23 OS |
1144 | * memmap_on_memory is an exception because it reserves initial part |
1145 | * of the physical memory space for vmemmaps. That space is pageblock | |
1146 | * aligned. | |
1147 | */ | |
ee0913c4 | 1148 | if (WARN_ON_ONCE(!nr_pages || !pageblock_aligned(pfn) || |
dd8e2f23 | 1149 | !IS_ALIGNED(pfn + nr_pages, PAGES_PER_SECTION))) |
4986fac1 DH |
1150 | return -EINVAL; |
1151 | ||
381eab4a DH |
1152 | mem_hotplug_begin(); |
1153 | ||
f1dd2cd1 | 1154 | /* associate pfn range with the zone */ |
b30c5927 | 1155 | move_pfn_range_to_zone(zone, pfn, nr_pages, NULL, MIGRATE_ISOLATE); |
f1dd2cd1 | 1156 | |
7b78d335 YG |
1157 | arg.start_pfn = pfn; |
1158 | arg.nr_pages = nr_pages; | |
d9713679 | 1159 | node_states_check_changes_online(nr_pages, zone, &arg); |
7b78d335 | 1160 | |
7b78d335 YG |
1161 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
1162 | ret = notifier_to_errno(ret); | |
e33e33b4 CY |
1163 | if (ret) |
1164 | goto failed_addition; | |
1165 | ||
b30c5927 DH |
1166 | /* |
1167 | * Fixup the number of isolated pageblocks before marking the sections | |
1168 | * onlining, such that undo_isolate_page_range() works correctly. | |
1169 | */ | |
1170 | spin_lock_irqsave(&zone->lock, flags); | |
1171 | zone->nr_isolate_pageblock += nr_pages / pageblock_nr_pages; | |
1172 | spin_unlock_irqrestore(&zone->lock, flags); | |
1173 | ||
6811378e YG |
1174 | /* |
1175 | * If this zone is not populated, then it is not in zonelist. | |
1176 | * This means the page allocator ignores this zone. | |
1177 | * So, zonelist must be updated after online. | |
1178 | */ | |
6dcd73d7 | 1179 | if (!populated_zone(zone)) { |
6811378e | 1180 | need_zonelists_rebuild = 1; |
72675e13 | 1181 | setup_zone_pageset(zone); |
6dcd73d7 | 1182 | } |
6811378e | 1183 | |
aac65321 | 1184 | online_pages_range(pfn, nr_pages); |
836809ec | 1185 | adjust_present_page_count(pfn_to_page(pfn), group, nr_pages); |
aa47228a | 1186 | |
b30c5927 DH |
1187 | node_states_set_node(nid, &arg); |
1188 | if (need_zonelists_rebuild) | |
1189 | build_all_zonelists(NULL); | |
b30c5927 DH |
1190 | |
1191 | /* Basic onlining is complete, allow allocation of onlined pages. */ | |
1192 | undo_isolate_page_range(pfn, pfn + nr_pages, MIGRATE_MOVABLE); | |
1193 | ||
93146d98 | 1194 | /* |
b86c5fc4 DH |
1195 | * Freshly onlined pages aren't shuffled (e.g., all pages are placed to |
1196 | * the tail of the freelist when undoing isolation). Shuffle the whole | |
1197 | * zone to make sure the just onlined pages are properly distributed | |
1198 | * across the whole freelist - to create an initial shuffle. | |
93146d98 | 1199 | */ |
e900a918 DW |
1200 | shuffle_zone(zone); |
1201 | ||
b92ca18e | 1202 | /* reinitialise watermarks and update pcp limits */ |
1b79acc9 KM |
1203 | init_per_zone_wmark_min(); |
1204 | ||
ca9a46f8 DH |
1205 | kswapd_run(nid); |
1206 | kcompactd_run(nid); | |
61b13993 | 1207 | |
2d1d43f6 | 1208 | writeback_set_ratelimit(); |
7b78d335 | 1209 | |
ca9a46f8 | 1210 | memory_notify(MEM_ONLINE, &arg); |
381eab4a | 1211 | mem_hotplug_done(); |
30467e0b | 1212 | return 0; |
e33e33b4 CY |
1213 | |
1214 | failed_addition: | |
1215 | pr_debug("online_pages [mem %#010llx-%#010llx] failed\n", | |
1216 | (unsigned long long) pfn << PAGE_SHIFT, | |
1217 | (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1); | |
1218 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
feee6b29 | 1219 | remove_pfn_range_from_zone(zone, pfn, nr_pages); |
381eab4a | 1220 | mem_hotplug_done(); |
e33e33b4 | 1221 | return ret; |
3947be19 | 1222 | } |
bc02af93 | 1223 | |
e1319331 | 1224 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
09f49dca | 1225 | static pg_data_t __ref *hotadd_init_pgdat(int nid) |
9af3c2de YG |
1226 | { |
1227 | struct pglist_data *pgdat; | |
9af3c2de | 1228 | |
09f49dca MH |
1229 | /* |
1230 | * NODE_DATA is preallocated (free_area_init) but its internal | |
1231 | * state is not allocated completely. Add missing pieces. | |
1232 | * Completely offline nodes stay around and they just need | |
1233 | * reintialization. | |
1234 | */ | |
70b5b46a | 1235 | pgdat = NODE_DATA(nid); |
03e85f9d | 1236 | |
9af3c2de | 1237 | /* init node's zones as empty zones, we don't have any present pages.*/ |
70b5b46a | 1238 | free_area_init_core_hotplug(pgdat); |
9af3c2de | 1239 | |
959ecc48 KH |
1240 | /* |
1241 | * The node we allocated has no zone fallback lists. For avoiding | |
1242 | * to access not-initialized zonelist, build here. | |
1243 | */ | |
72675e13 | 1244 | build_all_zonelists(pgdat); |
959ecc48 | 1245 | |
9af3c2de YG |
1246 | return pgdat; |
1247 | } | |
1248 | ||
ba2d2666 MG |
1249 | /* |
1250 | * __try_online_node - online a node if offlined | |
e8b098fc | 1251 | * @nid: the node ID |
b9ff0360 | 1252 | * @set_node_online: Whether we want to online the node |
cf23422b | 1253 | * called by cpu_up() to online a node without onlined memory. |
b9ff0360 OS |
1254 | * |
1255 | * Returns: | |
1256 | * 1 -> a new node has been allocated | |
1257 | * 0 -> the node is already online | |
1258 | * -ENOMEM -> the node could not be allocated | |
cf23422b | 1259 | */ |
c68ab18c | 1260 | static int __try_online_node(int nid, bool set_node_online) |
cf23422b | 1261 | { |
b9ff0360 OS |
1262 | pg_data_t *pgdat; |
1263 | int ret = 1; | |
cf23422b | 1264 | |
01b0f197 TK |
1265 | if (node_online(nid)) |
1266 | return 0; | |
1267 | ||
09f49dca | 1268 | pgdat = hotadd_init_pgdat(nid); |
7553e8f2 | 1269 | if (!pgdat) { |
01b0f197 | 1270 | pr_err("Cannot online node %d due to NULL pgdat\n", nid); |
cf23422b | 1271 | ret = -ENOMEM; |
1272 | goto out; | |
1273 | } | |
b9ff0360 OS |
1274 | |
1275 | if (set_node_online) { | |
1276 | node_set_online(nid); | |
1277 | ret = register_one_node(nid); | |
1278 | BUG_ON(ret); | |
1279 | } | |
cf23422b | 1280 | out: |
b9ff0360 OS |
1281 | return ret; |
1282 | } | |
1283 | ||
1284 | /* | |
1285 | * Users of this function always want to online/register the node | |
1286 | */ | |
1287 | int try_online_node(int nid) | |
1288 | { | |
1289 | int ret; | |
1290 | ||
1291 | mem_hotplug_begin(); | |
c68ab18c | 1292 | ret = __try_online_node(nid, true); |
bfc8c901 | 1293 | mem_hotplug_done(); |
cf23422b | 1294 | return ret; |
1295 | } | |
1296 | ||
27356f54 TK |
1297 | static int check_hotplug_memory_range(u64 start, u64 size) |
1298 | { | |
ba325585 | 1299 | /* memory range must be block size aligned */ |
cec3ebd0 DH |
1300 | if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) || |
1301 | !IS_ALIGNED(size, memory_block_size_bytes())) { | |
ba325585 | 1302 | pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx", |
cec3ebd0 | 1303 | memory_block_size_bytes(), start, size); |
27356f54 TK |
1304 | return -EINVAL; |
1305 | } | |
1306 | ||
1307 | return 0; | |
1308 | } | |
1309 | ||
31bc3858 VK |
1310 | static int online_memory_block(struct memory_block *mem, void *arg) |
1311 | { | |
1adf8b46 | 1312 | mem->online_type = mhp_default_online_type; |
dc18d706 | 1313 | return device_online(&mem->dev); |
31bc3858 VK |
1314 | } |
1315 | ||
85a2b4b0 AK |
1316 | #ifndef arch_supports_memmap_on_memory |
1317 | static inline bool arch_supports_memmap_on_memory(unsigned long vmemmap_size) | |
1318 | { | |
1319 | /* | |
1320 | * As default, we want the vmemmap to span a complete PMD such that we | |
1321 | * can map the vmemmap using a single PMD if supported by the | |
1322 | * architecture. | |
1323 | */ | |
1324 | return IS_ALIGNED(vmemmap_size, PMD_SIZE); | |
1325 | } | |
1326 | #endif | |
1327 | ||
e3c2bfdd | 1328 | static bool mhp_supports_memmap_on_memory(unsigned long size) |
a08a2ae3 | 1329 | { |
85a2b4b0 | 1330 | unsigned long vmemmap_size = memory_block_memmap_size(); |
2d1f649c | 1331 | unsigned long memmap_pages = memory_block_memmap_on_memory_pages(); |
a08a2ae3 OS |
1332 | |
1333 | /* | |
1334 | * Besides having arch support and the feature enabled at runtime, we | |
1335 | * need a few more assumptions to hold true: | |
1336 | * | |
1337 | * a) We span a single memory block: memory onlining/offlinin;g happens | |
1338 | * in memory block granularity. We don't want the vmemmap of online | |
1339 | * memory blocks to reside on offline memory blocks. In the future, | |
1340 | * we might want to support variable-sized memory blocks to make the | |
1341 | * feature more versatile. | |
1342 | * | |
1343 | * b) The vmemmap pages span complete PMDs: We don't want vmemmap code | |
1344 | * to populate memory from the altmap for unrelated parts (i.e., | |
1345 | * other memory blocks) | |
1346 | * | |
1347 | * c) The vmemmap pages (and thereby the pages that will be exposed to | |
1348 | * the buddy) have to cover full pageblocks: memory onlining/offlining | |
1349 | * code requires applicable ranges to be page-aligned, for example, to | |
1350 | * set the migratetypes properly. | |
1351 | * | |
1352 | * TODO: Although we have a check here to make sure that vmemmap pages | |
1353 | * fully populate a PMD, it is not the right place to check for | |
1354 | * this. A much better solution involves improving vmemmap code | |
1355 | * to fallback to base pages when trying to populate vmemmap using | |
1356 | * altmap as an alternative source of memory, and we do not exactly | |
1357 | * populate a single PMD. | |
1358 | */ | |
2d1f649c AK |
1359 | if (!mhp_memmap_on_memory() || size != memory_block_size_bytes()) |
1360 | return false; | |
1361 | ||
1362 | /* | |
1363 | * Make sure the vmemmap allocation is fully contained | |
1364 | * so that we always allocate vmemmap memory from altmap area. | |
1365 | */ | |
1366 | if (!IS_ALIGNED(vmemmap_size, PAGE_SIZE)) | |
1367 | return false; | |
1368 | ||
1369 | /* | |
1370 | * start pfn should be pageblock_nr_pages aligned for correctly | |
1371 | * setting migrate types | |
1372 | */ | |
1373 | if (!pageblock_aligned(memmap_pages)) | |
1374 | return false; | |
1375 | ||
1376 | if (memmap_pages == PHYS_PFN(memory_block_size_bytes())) | |
1377 | /* No effective hotplugged memory doesn't make sense. */ | |
1378 | return false; | |
1379 | ||
1380 | return arch_supports_memmap_on_memory(vmemmap_size); | |
a08a2ae3 OS |
1381 | } |
1382 | ||
8df1d0e4 DH |
1383 | /* |
1384 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
1385 | * and online/offline operations (triggered e.g. by sysfs). | |
1386 | * | |
1387 | * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG | |
1388 | */ | |
b6117199 | 1389 | int __ref add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags) |
bc02af93 | 1390 | { |
d15dfd31 | 1391 | struct mhp_params params = { .pgprot = pgprot_mhp(PAGE_KERNEL) }; |
32befe9e | 1392 | enum memblock_flags memblock_flags = MEMBLOCK_NONE; |
2d1f649c AK |
1393 | struct vmem_altmap mhp_altmap = { |
1394 | .base_pfn = PHYS_PFN(res->start), | |
1395 | .end_pfn = PHYS_PFN(res->end), | |
1396 | }; | |
028fc57a | 1397 | struct memory_group *group = NULL; |
62cedb9f | 1398 | u64 start, size; |
b9ff0360 | 1399 | bool new_node = false; |
bc02af93 YG |
1400 | int ret; |
1401 | ||
62cedb9f DV |
1402 | start = res->start; |
1403 | size = resource_size(res); | |
1404 | ||
27356f54 TK |
1405 | ret = check_hotplug_memory_range(start, size); |
1406 | if (ret) | |
1407 | return ret; | |
1408 | ||
028fc57a DH |
1409 | if (mhp_flags & MHP_NID_IS_MGID) { |
1410 | group = memory_group_find_by_id(nid); | |
1411 | if (!group) | |
1412 | return -EINVAL; | |
1413 | nid = group->nid; | |
1414 | } | |
1415 | ||
fa6d9ec7 VV |
1416 | if (!node_possible(nid)) { |
1417 | WARN(1, "node %d was absent from the node_possible_map\n", nid); | |
1418 | return -EINVAL; | |
1419 | } | |
1420 | ||
bfc8c901 | 1421 | mem_hotplug_begin(); |
ac13c462 | 1422 | |
53d38316 | 1423 | if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) { |
32befe9e DH |
1424 | if (res->flags & IORESOURCE_SYSRAM_DRIVER_MANAGED) |
1425 | memblock_flags = MEMBLOCK_DRIVER_MANAGED; | |
1426 | ret = memblock_add_node(start, size, nid, memblock_flags); | |
53d38316 DH |
1427 | if (ret) |
1428 | goto error_mem_hotplug_end; | |
1429 | } | |
7f36e3e5 | 1430 | |
c68ab18c | 1431 | ret = __try_online_node(nid, false); |
b9ff0360 OS |
1432 | if (ret < 0) |
1433 | goto error; | |
1434 | new_node = ret; | |
9af3c2de | 1435 | |
a08a2ae3 OS |
1436 | /* |
1437 | * Self hosted memmap array | |
1438 | */ | |
1439 | if (mhp_flags & MHP_MEMMAP_ON_MEMORY) { | |
e3c2bfdd | 1440 | if (mhp_supports_memmap_on_memory(size)) { |
2d1f649c | 1441 | mhp_altmap.free = memory_block_memmap_on_memory_pages(); |
1a8c64e1 AK |
1442 | params.altmap = kmalloc(sizeof(struct vmem_altmap), GFP_KERNEL); |
1443 | if (!params.altmap) { | |
1444 | ret = -ENOMEM; | |
1445 | goto error; | |
1446 | } | |
1447 | ||
1448 | memcpy(params.altmap, &mhp_altmap, sizeof(mhp_altmap)); | |
a08a2ae3 | 1449 | } |
e3c2bfdd | 1450 | /* fallback to not using altmap */ |
a08a2ae3 OS |
1451 | } |
1452 | ||
bc02af93 | 1453 | /* call arch's memory hotadd */ |
f5637d3b | 1454 | ret = arch_add_memory(nid, start, size, ¶ms); |
9af3c2de | 1455 | if (ret < 0) |
1a8c64e1 | 1456 | goto error_free; |
9af3c2de | 1457 | |
db051a0d | 1458 | /* create memory block devices after memory was added */ |
1a8c64e1 | 1459 | ret = create_memory_block_devices(start, size, params.altmap, group); |
db051a0d | 1460 | if (ret) { |
65a2aa5f | 1461 | arch_remove_memory(start, size, NULL); |
1a8c64e1 | 1462 | goto error_free; |
db051a0d DH |
1463 | } |
1464 | ||
a1e565aa | 1465 | if (new_node) { |
d5b6f6a3 | 1466 | /* If sysfs file of new node can't be created, cpu on the node |
0fc44159 YG |
1467 | * can't be hot-added. There is no rollback way now. |
1468 | * So, check by BUG_ON() to catch it reluctantly.. | |
d5b6f6a3 | 1469 | * We online node here. We can't roll back from here. |
0fc44159 | 1470 | */ |
d5b6f6a3 OS |
1471 | node_set_online(nid); |
1472 | ret = __register_one_node(nid); | |
0fc44159 YG |
1473 | BUG_ON(ret); |
1474 | } | |
1475 | ||
cc651559 DH |
1476 | register_memory_blocks_under_node(nid, PFN_DOWN(start), |
1477 | PFN_UP(start + size - 1), | |
1478 | MEMINIT_HOTPLUG); | |
d5b6f6a3 | 1479 | |
d96ae530 | 1480 | /* create new memmap entry */ |
7b7b2721 DH |
1481 | if (!strcmp(res->name, "System RAM")) |
1482 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
d96ae530 | 1483 | |
381eab4a DH |
1484 | /* device_online() will take the lock when calling online_pages() */ |
1485 | mem_hotplug_done(); | |
1486 | ||
9ca6551e DH |
1487 | /* |
1488 | * In case we're allowed to merge the resource, flag it and trigger | |
1489 | * merging now that adding succeeded. | |
1490 | */ | |
26011267 | 1491 | if (mhp_flags & MHP_MERGE_RESOURCE) |
9ca6551e DH |
1492 | merge_system_ram_resource(res); |
1493 | ||
31bc3858 | 1494 | /* online pages if requested */ |
1adf8b46 | 1495 | if (mhp_default_online_type != MMOP_OFFLINE) |
fbcf73ce | 1496 | walk_memory_blocks(start, size, NULL, online_memory_block); |
31bc3858 | 1497 | |
381eab4a | 1498 | return ret; |
1a8c64e1 AK |
1499 | error_free: |
1500 | kfree(params.altmap); | |
9af3c2de | 1501 | error: |
52219aea DH |
1502 | if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) |
1503 | memblock_remove(start, size); | |
53d38316 | 1504 | error_mem_hotplug_end: |
bfc8c901 | 1505 | mem_hotplug_done(); |
bc02af93 YG |
1506 | return ret; |
1507 | } | |
62cedb9f | 1508 | |
8df1d0e4 | 1509 | /* requires device_hotplug_lock, see add_memory_resource() */ |
b6117199 | 1510 | int __ref __add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags) |
62cedb9f DV |
1511 | { |
1512 | struct resource *res; | |
1513 | int ret; | |
1514 | ||
7b7b2721 | 1515 | res = register_memory_resource(start, size, "System RAM"); |
6f754ba4 VK |
1516 | if (IS_ERR(res)) |
1517 | return PTR_ERR(res); | |
62cedb9f | 1518 | |
b6117199 | 1519 | ret = add_memory_resource(nid, res, mhp_flags); |
62cedb9f DV |
1520 | if (ret < 0) |
1521 | release_memory_resource(res); | |
1522 | return ret; | |
1523 | } | |
8df1d0e4 | 1524 | |
b6117199 | 1525 | int add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags) |
8df1d0e4 DH |
1526 | { |
1527 | int rc; | |
1528 | ||
1529 | lock_device_hotplug(); | |
b6117199 | 1530 | rc = __add_memory(nid, start, size, mhp_flags); |
8df1d0e4 DH |
1531 | unlock_device_hotplug(); |
1532 | ||
1533 | return rc; | |
1534 | } | |
bc02af93 | 1535 | EXPORT_SYMBOL_GPL(add_memory); |
0c0e6195 | 1536 | |
7b7b2721 DH |
1537 | /* |
1538 | * Add special, driver-managed memory to the system as system RAM. Such | |
1539 | * memory is not exposed via the raw firmware-provided memmap as system | |
1540 | * RAM, instead, it is detected and added by a driver - during cold boot, | |
1541 | * after a reboot, and after kexec. | |
1542 | * | |
1543 | * Reasons why this memory should not be used for the initial memmap of a | |
1544 | * kexec kernel or for placing kexec images: | |
1545 | * - The booting kernel is in charge of determining how this memory will be | |
1546 | * used (e.g., use persistent memory as system RAM) | |
1547 | * - Coordination with a hypervisor is required before this memory | |
1548 | * can be used (e.g., inaccessible parts). | |
1549 | * | |
1550 | * For this memory, no entries in /sys/firmware/memmap ("raw firmware-provided | |
1551 | * memory map") are created. Also, the created memory resource is flagged | |
7cf603d1 | 1552 | * with IORESOURCE_SYSRAM_DRIVER_MANAGED, so in-kernel users can special-case |
7b7b2721 DH |
1553 | * this memory as well (esp., not place kexec images onto it). |
1554 | * | |
1555 | * The resource_name (visible via /proc/iomem) has to have the format | |
1556 | * "System RAM ($DRIVER)". | |
1557 | */ | |
1558 | int add_memory_driver_managed(int nid, u64 start, u64 size, | |
b6117199 | 1559 | const char *resource_name, mhp_t mhp_flags) |
7b7b2721 DH |
1560 | { |
1561 | struct resource *res; | |
1562 | int rc; | |
1563 | ||
1564 | if (!resource_name || | |
1565 | strstr(resource_name, "System RAM (") != resource_name || | |
1566 | resource_name[strlen(resource_name) - 1] != ')') | |
1567 | return -EINVAL; | |
1568 | ||
1569 | lock_device_hotplug(); | |
1570 | ||
1571 | res = register_memory_resource(start, size, resource_name); | |
1572 | if (IS_ERR(res)) { | |
1573 | rc = PTR_ERR(res); | |
1574 | goto out_unlock; | |
1575 | } | |
1576 | ||
b6117199 | 1577 | rc = add_memory_resource(nid, res, mhp_flags); |
7b7b2721 DH |
1578 | if (rc < 0) |
1579 | release_memory_resource(res); | |
1580 | ||
1581 | out_unlock: | |
1582 | unlock_device_hotplug(); | |
1583 | return rc; | |
1584 | } | |
1585 | EXPORT_SYMBOL_GPL(add_memory_driver_managed); | |
1586 | ||
bca3feaa AK |
1587 | /* |
1588 | * Platforms should define arch_get_mappable_range() that provides | |
1589 | * maximum possible addressable physical memory range for which the | |
1590 | * linear mapping could be created. The platform returned address | |
1591 | * range must adhere to these following semantics. | |
1592 | * | |
1593 | * - range.start <= range.end | |
1594 | * - Range includes both end points [range.start..range.end] | |
1595 | * | |
1596 | * There is also a fallback definition provided here, allowing the | |
1597 | * entire possible physical address range in case any platform does | |
1598 | * not define arch_get_mappable_range(). | |
1599 | */ | |
1600 | struct range __weak arch_get_mappable_range(void) | |
1601 | { | |
1602 | struct range mhp_range = { | |
1603 | .start = 0UL, | |
1604 | .end = -1ULL, | |
1605 | }; | |
1606 | return mhp_range; | |
1607 | } | |
1608 | ||
1609 | struct range mhp_get_pluggable_range(bool need_mapping) | |
1610 | { | |
1611 | const u64 max_phys = (1ULL << MAX_PHYSMEM_BITS) - 1; | |
1612 | struct range mhp_range; | |
1613 | ||
1614 | if (need_mapping) { | |
1615 | mhp_range = arch_get_mappable_range(); | |
1616 | if (mhp_range.start > max_phys) { | |
1617 | mhp_range.start = 0; | |
1618 | mhp_range.end = 0; | |
1619 | } | |
1620 | mhp_range.end = min_t(u64, mhp_range.end, max_phys); | |
1621 | } else { | |
1622 | mhp_range.start = 0; | |
1623 | mhp_range.end = max_phys; | |
1624 | } | |
1625 | return mhp_range; | |
1626 | } | |
1627 | EXPORT_SYMBOL_GPL(mhp_get_pluggable_range); | |
1628 | ||
1629 | bool mhp_range_allowed(u64 start, u64 size, bool need_mapping) | |
1630 | { | |
1631 | struct range mhp_range = mhp_get_pluggable_range(need_mapping); | |
1632 | u64 end = start + size; | |
1633 | ||
1634 | if (start < end && start >= mhp_range.start && (end - 1) <= mhp_range.end) | |
1635 | return true; | |
1636 | ||
1637 | pr_warn("Hotplug memory [%#llx-%#llx] exceeds maximum addressable range [%#llx-%#llx]\n", | |
1638 | start, end, mhp_range.start, mhp_range.end); | |
1639 | return false; | |
1640 | } | |
1641 | ||
0c0e6195 | 1642 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 | 1643 | /* |
0efadf48 | 1644 | * Scan pfn range [start,end) to find movable/migratable pages (LRU pages, |
aa218795 DH |
1645 | * non-lru movable pages and hugepages). Will skip over most unmovable |
1646 | * pages (esp., pages that can be skipped when offlining), but bail out on | |
1647 | * definitely unmovable pages. | |
1648 | * | |
1649 | * Returns: | |
1650 | * 0 in case a movable page is found and movable_pfn was updated. | |
1651 | * -ENOENT in case no movable page was found. | |
1652 | * -EBUSY in case a definitely unmovable page was found. | |
0c0e6195 | 1653 | */ |
aa218795 DH |
1654 | static int scan_movable_pages(unsigned long start, unsigned long end, |
1655 | unsigned long *movable_pfn) | |
0c0e6195 KH |
1656 | { |
1657 | unsigned long pfn; | |
eeb0efd0 | 1658 | |
0c0e6195 | 1659 | for (pfn = start; pfn < end; pfn++) { |
eeb0efd0 OS |
1660 | struct page *page, *head; |
1661 | unsigned long skip; | |
1662 | ||
1663 | if (!pfn_valid(pfn)) | |
1664 | continue; | |
1665 | page = pfn_to_page(pfn); | |
1666 | if (PageLRU(page)) | |
aa218795 | 1667 | goto found; |
eeb0efd0 | 1668 | if (__PageMovable(page)) |
aa218795 DH |
1669 | goto found; |
1670 | ||
1671 | /* | |
1672 | * PageOffline() pages that are not marked __PageMovable() and | |
1673 | * have a reference count > 0 (after MEM_GOING_OFFLINE) are | |
1674 | * definitely unmovable. If their reference count would be 0, | |
1675 | * they could at least be skipped when offlining memory. | |
1676 | */ | |
1677 | if (PageOffline(page) && page_count(page)) | |
1678 | return -EBUSY; | |
eeb0efd0 OS |
1679 | |
1680 | if (!PageHuge(page)) | |
1681 | continue; | |
1682 | head = compound_head(page); | |
8f251a3d MK |
1683 | /* |
1684 | * This test is racy as we hold no reference or lock. The | |
1685 | * hugetlb page could have been free'ed and head is no longer | |
1686 | * a hugetlb page before the following check. In such unlikely | |
1687 | * cases false positives and negatives are possible. Calling | |
1688 | * code must deal with these scenarios. | |
1689 | */ | |
1690 | if (HPageMigratable(head)) | |
aa218795 | 1691 | goto found; |
1640a0ef | 1692 | skip = compound_nr(head) - (pfn - page_to_pfn(head)); |
eeb0efd0 | 1693 | pfn += skip - 1; |
0c0e6195 | 1694 | } |
aa218795 DH |
1695 | return -ENOENT; |
1696 | found: | |
1697 | *movable_pfn = pfn; | |
0c0e6195 KH |
1698 | return 0; |
1699 | } | |
1700 | ||
32cf666e | 1701 | static void do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) |
0c0e6195 KH |
1702 | { |
1703 | unsigned long pfn; | |
6c357848 | 1704 | struct page *page, *head; |
0c0e6195 | 1705 | LIST_HEAD(source); |
786dee86 LM |
1706 | static DEFINE_RATELIMIT_STATE(migrate_rs, DEFAULT_RATELIMIT_INTERVAL, |
1707 | DEFAULT_RATELIMIT_BURST); | |
0c0e6195 | 1708 | |
a85009c3 | 1709 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
869f7ee6 | 1710 | struct folio *folio; |
f7f9c00d | 1711 | bool isolated; |
869f7ee6 | 1712 | |
0c0e6195 KH |
1713 | if (!pfn_valid(pfn)) |
1714 | continue; | |
1715 | page = pfn_to_page(pfn); | |
869f7ee6 MWO |
1716 | folio = page_folio(page); |
1717 | head = &folio->page; | |
c8721bbb NH |
1718 | |
1719 | if (PageHuge(page)) { | |
d8c6546b | 1720 | pfn = page_to_pfn(head) + compound_nr(head) - 1; |
6aa3a920 | 1721 | isolate_hugetlb(folio, &source); |
c8721bbb | 1722 | continue; |
94723aaf | 1723 | } else if (PageTransHuge(page)) |
6c357848 | 1724 | pfn = page_to_pfn(head) + thp_nr_pages(page) - 1; |
c8721bbb | 1725 | |
b15c8726 MH |
1726 | /* |
1727 | * HWPoison pages have elevated reference counts so the migration would | |
1728 | * fail on them. It also doesn't make any sense to migrate them in the | |
1729 | * first place. Still try to unmap such a page in case it is still mapped | |
1730 | * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep | |
1731 | * the unmap as the catch all safety net). | |
1732 | */ | |
1733 | if (PageHWPoison(page)) { | |
869f7ee6 MWO |
1734 | if (WARN_ON(folio_test_lru(folio))) |
1735 | folio_isolate_lru(folio); | |
1736 | if (folio_mapped(folio)) | |
1737 | try_to_unmap(folio, TTU_IGNORE_MLOCK); | |
b15c8726 MH |
1738 | continue; |
1739 | } | |
1740 | ||
700c2a46 | 1741 | if (!get_page_unless_zero(page)) |
0c0e6195 KH |
1742 | continue; |
1743 | /* | |
0efadf48 YX |
1744 | * We can skip free pages. And we can deal with pages on |
1745 | * LRU and non-lru movable pages. | |
0c0e6195 | 1746 | */ |
cd775580 | 1747 | if (PageLRU(page)) |
f7f9c00d | 1748 | isolated = isolate_lru_page(page); |
cd775580 BW |
1749 | else |
1750 | isolated = isolate_movable_page(page, ISOLATE_UNEVICTABLE); | |
1751 | if (isolated) { | |
62695a84 | 1752 | list_add_tail(&page->lru, &source); |
0efadf48 YX |
1753 | if (!__PageMovable(page)) |
1754 | inc_node_page_state(page, NR_ISOLATED_ANON + | |
9de4f22a | 1755 | page_is_file_lru(page)); |
6d9c285a | 1756 | |
0c0e6195 | 1757 | } else { |
786dee86 LM |
1758 | if (__ratelimit(&migrate_rs)) { |
1759 | pr_warn("failed to isolate pfn %lx\n", pfn); | |
1760 | dump_page(page, "isolation failed"); | |
1761 | } | |
0c0e6195 | 1762 | } |
1723058e | 1763 | put_page(page); |
0c0e6195 | 1764 | } |
f3ab2636 | 1765 | if (!list_empty(&source)) { |
203e6e5c JK |
1766 | nodemask_t nmask = node_states[N_MEMORY]; |
1767 | struct migration_target_control mtc = { | |
1768 | .nmask = &nmask, | |
1769 | .gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL, | |
1770 | }; | |
32cf666e | 1771 | int ret; |
203e6e5c JK |
1772 | |
1773 | /* | |
1774 | * We have checked that migration range is on a single zone so | |
1775 | * we can use the nid of the first page to all the others. | |
1776 | */ | |
1777 | mtc.nid = page_to_nid(list_first_entry(&source, struct page, lru)); | |
1778 | ||
1779 | /* | |
1780 | * try to allocate from a different node but reuse this node | |
1781 | * if there are no other online nodes to be used (e.g. we are | |
1782 | * offlining a part of the only existing node) | |
1783 | */ | |
1784 | node_clear(mtc.nid, nmask); | |
1785 | if (nodes_empty(nmask)) | |
1786 | node_set(mtc.nid, nmask); | |
1787 | ret = migrate_pages(&source, alloc_migration_target, NULL, | |
5ac95884 | 1788 | (unsigned long)&mtc, MIGRATE_SYNC, MR_MEMORY_HOTPLUG, NULL); |
2932c8b0 MH |
1789 | if (ret) { |
1790 | list_for_each_entry(page, &source, lru) { | |
786dee86 LM |
1791 | if (__ratelimit(&migrate_rs)) { |
1792 | pr_warn("migrating pfn %lx failed ret:%d\n", | |
1793 | page_to_pfn(page), ret); | |
1794 | dump_page(page, "migration failure"); | |
1795 | } | |
2932c8b0 | 1796 | } |
c8721bbb | 1797 | putback_movable_pages(&source); |
2932c8b0 | 1798 | } |
0c0e6195 | 1799 | } |
0c0e6195 KH |
1800 | } |
1801 | ||
c5320926 TC |
1802 | static int __init cmdline_parse_movable_node(char *p) |
1803 | { | |
55ac590c | 1804 | movable_node_enabled = true; |
c5320926 TC |
1805 | return 0; |
1806 | } | |
1807 | early_param("movable_node", cmdline_parse_movable_node); | |
1808 | ||
d9713679 LJ |
1809 | /* check which state of node_states will be changed when offline memory */ |
1810 | static void node_states_check_changes_offline(unsigned long nr_pages, | |
1811 | struct zone *zone, struct memory_notify *arg) | |
1812 | { | |
1813 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1814 | unsigned long present_pages = 0; | |
86b27bea | 1815 | enum zone_type zt; |
d9713679 | 1816 | |
98fa15f3 AK |
1817 | arg->status_change_nid = NUMA_NO_NODE; |
1818 | arg->status_change_nid_normal = NUMA_NO_NODE; | |
d9713679 LJ |
1819 | |
1820 | /* | |
86b27bea OS |
1821 | * Check whether node_states[N_NORMAL_MEMORY] will be changed. |
1822 | * If the memory to be offline is within the range | |
1823 | * [0..ZONE_NORMAL], and it is the last present memory there, | |
1824 | * the zones in that range will become empty after the offlining, | |
1825 | * thus we can determine that we need to clear the node from | |
1826 | * node_states[N_NORMAL_MEMORY]. | |
d9713679 | 1827 | */ |
86b27bea | 1828 | for (zt = 0; zt <= ZONE_NORMAL; zt++) |
d9713679 | 1829 | present_pages += pgdat->node_zones[zt].present_pages; |
86b27bea | 1830 | if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages) |
d9713679 | 1831 | arg->status_change_nid_normal = zone_to_nid(zone); |
d9713679 LJ |
1832 | |
1833 | /* | |
6b740c6c DH |
1834 | * We have accounted the pages from [0..ZONE_NORMAL); ZONE_HIGHMEM |
1835 | * does not apply as we don't support 32bit. | |
86b27bea OS |
1836 | * Here we count the possible pages from ZONE_MOVABLE. |
1837 | * If after having accounted all the pages, we see that the nr_pages | |
1838 | * to be offlined is over or equal to the accounted pages, | |
1839 | * we know that the node will become empty, and so, we can clear | |
1840 | * it for N_MEMORY as well. | |
d9713679 | 1841 | */ |
86b27bea | 1842 | present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages; |
d9713679 | 1843 | |
d9713679 LJ |
1844 | if (nr_pages >= present_pages) |
1845 | arg->status_change_nid = zone_to_nid(zone); | |
d9713679 LJ |
1846 | } |
1847 | ||
1848 | static void node_states_clear_node(int node, struct memory_notify *arg) | |
1849 | { | |
1850 | if (arg->status_change_nid_normal >= 0) | |
1851 | node_clear_state(node, N_NORMAL_MEMORY); | |
1852 | ||
cf01f6f5 | 1853 | if (arg->status_change_nid >= 0) |
6715ddf9 | 1854 | node_clear_state(node, N_MEMORY); |
d9713679 LJ |
1855 | } |
1856 | ||
c5e79ef5 DH |
1857 | static int count_system_ram_pages_cb(unsigned long start_pfn, |
1858 | unsigned long nr_pages, void *data) | |
1859 | { | |
1860 | unsigned long *nr_system_ram_pages = data; | |
1861 | ||
1862 | *nr_system_ram_pages += nr_pages; | |
1863 | return 0; | |
1864 | } | |
1865 | ||
836809ec | 1866 | int __ref offline_pages(unsigned long start_pfn, unsigned long nr_pages, |
395f6081 | 1867 | struct zone *zone, struct memory_group *group) |
0c0e6195 | 1868 | { |
73a11c96 | 1869 | const unsigned long end_pfn = start_pfn + nr_pages; |
0a1a9a00 | 1870 | unsigned long pfn, system_ram_pages = 0; |
395f6081 | 1871 | const int node = zone_to_nid(zone); |
d702909f | 1872 | unsigned long flags; |
7b78d335 | 1873 | struct memory_notify arg; |
79605093 | 1874 | char *reason; |
395f6081 | 1875 | int ret; |
0c0e6195 | 1876 | |
dd8e2f23 OS |
1877 | /* |
1878 | * {on,off}lining is constrained to full memory sections (or more | |
041711ce | 1879 | * precisely to memory blocks from the user space POV). |
dd8e2f23 OS |
1880 | * memmap_on_memory is an exception because it reserves initial part |
1881 | * of the physical memory space for vmemmaps. That space is pageblock | |
1882 | * aligned. | |
1883 | */ | |
ee0913c4 | 1884 | if (WARN_ON_ONCE(!nr_pages || !pageblock_aligned(start_pfn) || |
dd8e2f23 | 1885 | !IS_ALIGNED(start_pfn + nr_pages, PAGES_PER_SECTION))) |
4986fac1 DH |
1886 | return -EINVAL; |
1887 | ||
381eab4a DH |
1888 | mem_hotplug_begin(); |
1889 | ||
c5e79ef5 DH |
1890 | /* |
1891 | * Don't allow to offline memory blocks that contain holes. | |
1892 | * Consequently, memory blocks with holes can never get onlined | |
1893 | * via the hotplug path - online_pages() - as hotplugged memory has | |
1894 | * no holes. This way, we e.g., don't have to worry about marking | |
1895 | * memory holes PG_reserved, don't need pfn_valid() checks, and can | |
1896 | * avoid using walk_system_ram_range() later. | |
1897 | */ | |
73a11c96 | 1898 | walk_system_ram_range(start_pfn, nr_pages, &system_ram_pages, |
c5e79ef5 | 1899 | count_system_ram_pages_cb); |
73a11c96 | 1900 | if (system_ram_pages != nr_pages) { |
c5e79ef5 DH |
1901 | ret = -EINVAL; |
1902 | reason = "memory holes"; | |
1903 | goto failed_removal; | |
1904 | } | |
1905 | ||
395f6081 DH |
1906 | /* |
1907 | * We only support offlining of memory blocks managed by a single zone, | |
1908 | * checked by calling code. This is just a sanity check that we might | |
1909 | * want to remove in the future. | |
1910 | */ | |
1911 | if (WARN_ON_ONCE(page_zone(pfn_to_page(start_pfn)) != zone || | |
1912 | page_zone(pfn_to_page(end_pfn - 1)) != zone)) { | |
79605093 MH |
1913 | ret = -EINVAL; |
1914 | reason = "multizone range"; | |
1915 | goto failed_removal; | |
381eab4a | 1916 | } |
7b78d335 | 1917 | |
ec6e8c7e VB |
1918 | /* |
1919 | * Disable pcplists so that page isolation cannot race with freeing | |
1920 | * in a way that pages from isolated pageblock are left on pcplists. | |
1921 | */ | |
1922 | zone_pcp_disable(zone); | |
d479960e | 1923 | lru_cache_disable(); |
ec6e8c7e | 1924 | |
0c0e6195 | 1925 | /* set above range as isolated */ |
b023f468 | 1926 | ret = start_isolate_page_range(start_pfn, end_pfn, |
d381c547 | 1927 | MIGRATE_MOVABLE, |
b2c9e2fb ZY |
1928 | MEMORY_OFFLINE | REPORT_FAILURE, |
1929 | GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL); | |
3fa0c7c7 | 1930 | if (ret) { |
79605093 | 1931 | reason = "failure to isolate range"; |
ec6e8c7e | 1932 | goto failed_removal_pcplists_disabled; |
381eab4a | 1933 | } |
7b78d335 YG |
1934 | |
1935 | arg.start_pfn = start_pfn; | |
1936 | arg.nr_pages = nr_pages; | |
d9713679 | 1937 | node_states_check_changes_offline(nr_pages, zone, &arg); |
7b78d335 YG |
1938 | |
1939 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
1940 | ret = notifier_to_errno(ret); | |
79605093 MH |
1941 | if (ret) { |
1942 | reason = "notifier failure"; | |
1943 | goto failed_removal_isolated; | |
1944 | } | |
7b78d335 | 1945 | |
bb8965bd | 1946 | do { |
aa218795 DH |
1947 | pfn = start_pfn; |
1948 | do { | |
de7cb03d DH |
1949 | /* |
1950 | * Historically we always checked for any signal and | |
1951 | * can't limit it to fatal signals without eventually | |
1952 | * breaking user space. | |
1953 | */ | |
bb8965bd MH |
1954 | if (signal_pending(current)) { |
1955 | ret = -EINTR; | |
1956 | reason = "signal backoff"; | |
1957 | goto failed_removal_isolated; | |
1958 | } | |
72b39cfc | 1959 | |
bb8965bd | 1960 | cond_resched(); |
bb8965bd | 1961 | |
aa218795 DH |
1962 | ret = scan_movable_pages(pfn, end_pfn, &pfn); |
1963 | if (!ret) { | |
bb8965bd MH |
1964 | /* |
1965 | * TODO: fatal migration failures should bail | |
1966 | * out | |
1967 | */ | |
1968 | do_migrate_range(pfn, end_pfn); | |
1969 | } | |
aa218795 DH |
1970 | } while (!ret); |
1971 | ||
1972 | if (ret != -ENOENT) { | |
1973 | reason = "unmovable page"; | |
1974 | goto failed_removal_isolated; | |
bb8965bd | 1975 | } |
0c0e6195 | 1976 | |
bb8965bd MH |
1977 | /* |
1978 | * Dissolve free hugepages in the memory block before doing | |
1979 | * offlining actually in order to make hugetlbfs's object | |
1980 | * counting consistent. | |
1981 | */ | |
1982 | ret = dissolve_free_huge_pages(start_pfn, end_pfn); | |
1983 | if (ret) { | |
1984 | reason = "failure to dissolve huge pages"; | |
1985 | goto failed_removal_isolated; | |
1986 | } | |
0a1a9a00 | 1987 | |
0a1a9a00 | 1988 | ret = test_pages_isolated(start_pfn, end_pfn, MEMORY_OFFLINE); |
ec6e8c7e | 1989 | |
5557c766 | 1990 | } while (ret); |
72b39cfc | 1991 | |
0a1a9a00 DH |
1992 | /* Mark all sections offline and remove free pages from the buddy. */ |
1993 | __offline_isolated_pages(start_pfn, end_pfn); | |
7c33023a | 1994 | pr_debug("Offlined Pages %ld\n", nr_pages); |
0a1a9a00 | 1995 | |
9b7ea46a | 1996 | /* |
b30c5927 DH |
1997 | * The memory sections are marked offline, and the pageblock flags |
1998 | * effectively stale; nobody should be touching them. Fixup the number | |
1999 | * of isolated pageblocks, memory onlining will properly revert this. | |
9b7ea46a QC |
2000 | */ |
2001 | spin_lock_irqsave(&zone->lock, flags); | |
ea15153c | 2002 | zone->nr_isolate_pageblock -= nr_pages / pageblock_nr_pages; |
9b7ea46a QC |
2003 | spin_unlock_irqrestore(&zone->lock, flags); |
2004 | ||
d479960e | 2005 | lru_cache_enable(); |
ec6e8c7e VB |
2006 | zone_pcp_enable(zone); |
2007 | ||
0c0e6195 | 2008 | /* removal success */ |
0a1a9a00 | 2009 | adjust_managed_page_count(pfn_to_page(start_pfn), -nr_pages); |
836809ec | 2010 | adjust_present_page_count(pfn_to_page(start_pfn), group, -nr_pages); |
7b78d335 | 2011 | |
b92ca18e | 2012 | /* reinitialise watermarks and update pcp limits */ |
1b79acc9 KM |
2013 | init_per_zone_wmark_min(); |
2014 | ||
b7812c86 QZ |
2015 | /* |
2016 | * Make sure to mark the node as memory-less before rebuilding the zone | |
2017 | * list. Otherwise this node would still appear in the fallback lists. | |
2018 | */ | |
2019 | node_states_clear_node(node, &arg); | |
1e8537ba | 2020 | if (!populated_zone(zone)) { |
340175b7 | 2021 | zone_pcp_reset(zone); |
72675e13 | 2022 | build_all_zonelists(NULL); |
b92ca18e | 2023 | } |
340175b7 | 2024 | |
698b1b30 | 2025 | if (arg.status_change_nid >= 0) { |
698b1b30 | 2026 | kcompactd_stop(node); |
b4a0215e | 2027 | kswapd_stop(node); |
698b1b30 | 2028 | } |
bce7394a | 2029 | |
0c0e6195 | 2030 | writeback_set_ratelimit(); |
7b78d335 YG |
2031 | |
2032 | memory_notify(MEM_OFFLINE, &arg); | |
feee6b29 | 2033 | remove_pfn_range_from_zone(zone, start_pfn, nr_pages); |
381eab4a | 2034 | mem_hotplug_done(); |
0c0e6195 KH |
2035 | return 0; |
2036 | ||
79605093 | 2037 | failed_removal_isolated: |
36ba30bc | 2038 | /* pushback to free area */ |
79605093 | 2039 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
c4efe484 | 2040 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
ec6e8c7e | 2041 | failed_removal_pcplists_disabled: |
946746d1 | 2042 | lru_cache_enable(); |
ec6e8c7e | 2043 | zone_pcp_enable(zone); |
0c0e6195 | 2044 | failed_removal: |
79605093 | 2045 | pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n", |
e33e33b4 | 2046 | (unsigned long long) start_pfn << PAGE_SHIFT, |
79605093 MH |
2047 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1, |
2048 | reason); | |
381eab4a | 2049 | mem_hotplug_done(); |
0c0e6195 KH |
2050 | return ret; |
2051 | } | |
71088785 | 2052 | |
d6de9d53 | 2053 | static int check_memblock_offlined_cb(struct memory_block *mem, void *arg) |
bbc76be6 | 2054 | { |
e1c158e4 | 2055 | int *nid = arg; |
bbc76be6 | 2056 | |
e1c158e4 | 2057 | *nid = mem->nid; |
639118d1 | 2058 | if (unlikely(mem->state != MEM_OFFLINE)) { |
349daa0f RD |
2059 | phys_addr_t beginpa, endpa; |
2060 | ||
2061 | beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)); | |
b6c88d3b | 2062 | endpa = beginpa + memory_block_size_bytes() - 1; |
756a025f | 2063 | pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n", |
349daa0f | 2064 | &beginpa, &endpa); |
bbc76be6 | 2065 | |
eca499ab PT |
2066 | return -EBUSY; |
2067 | } | |
2068 | return 0; | |
bbc76be6 WC |
2069 | } |
2070 | ||
1a8c64e1 | 2071 | static int test_has_altmap_cb(struct memory_block *mem, void *arg) |
a08a2ae3 | 2072 | { |
1a8c64e1 | 2073 | struct memory_block **mem_ptr = (struct memory_block **)arg; |
a08a2ae3 | 2074 | /* |
1a8c64e1 AK |
2075 | * return the memblock if we have altmap |
2076 | * and break callback. | |
a08a2ae3 | 2077 | */ |
1a8c64e1 AK |
2078 | if (mem->altmap) { |
2079 | *mem_ptr = mem; | |
2080 | return 1; | |
2081 | } | |
2082 | return 0; | |
a08a2ae3 OS |
2083 | } |
2084 | ||
b27340a5 | 2085 | static int check_cpu_on_node(int nid) |
60a5a19e | 2086 | { |
60a5a19e TC |
2087 | int cpu; |
2088 | ||
2089 | for_each_present_cpu(cpu) { | |
b27340a5 | 2090 | if (cpu_to_node(cpu) == nid) |
60a5a19e TC |
2091 | /* |
2092 | * the cpu on this node isn't removed, and we can't | |
2093 | * offline this node. | |
2094 | */ | |
2095 | return -EBUSY; | |
2096 | } | |
2097 | ||
2098 | return 0; | |
2099 | } | |
2100 | ||
2c91f8fc DH |
2101 | static int check_no_memblock_for_node_cb(struct memory_block *mem, void *arg) |
2102 | { | |
2103 | int nid = *(int *)arg; | |
2104 | ||
2105 | /* | |
2106 | * If a memory block belongs to multiple nodes, the stored nid is not | |
2107 | * reliable. However, such blocks are always online (e.g., cannot get | |
2108 | * offlined) and, therefore, are still spanned by the node. | |
2109 | */ | |
2110 | return mem->nid == nid ? -EEXIST : 0; | |
2111 | } | |
2112 | ||
0f1cfe9d TK |
2113 | /** |
2114 | * try_offline_node | |
e8b098fc | 2115 | * @nid: the node ID |
0f1cfe9d TK |
2116 | * |
2117 | * Offline a node if all memory sections and cpus of the node are removed. | |
2118 | * | |
2119 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
2120 | * and online/offline operations before this call. | |
2121 | */ | |
90b30cdc | 2122 | void try_offline_node(int nid) |
60a5a19e | 2123 | { |
2c91f8fc | 2124 | int rc; |
60a5a19e | 2125 | |
2c91f8fc DH |
2126 | /* |
2127 | * If the node still spans pages (especially ZONE_DEVICE), don't | |
2128 | * offline it. A node spans memory after move_pfn_range_to_zone(), | |
2129 | * e.g., after the memory block was onlined. | |
2130 | */ | |
b27340a5 | 2131 | if (node_spanned_pages(nid)) |
2c91f8fc | 2132 | return; |
60a5a19e | 2133 | |
2c91f8fc DH |
2134 | /* |
2135 | * Especially offline memory blocks might not be spanned by the | |
2136 | * node. They will get spanned by the node once they get onlined. | |
2137 | * However, they link to the node in sysfs and can get onlined later. | |
2138 | */ | |
2139 | rc = for_each_memory_block(&nid, check_no_memblock_for_node_cb); | |
2140 | if (rc) | |
60a5a19e | 2141 | return; |
60a5a19e | 2142 | |
b27340a5 | 2143 | if (check_cpu_on_node(nid)) |
60a5a19e TC |
2144 | return; |
2145 | ||
2146 | /* | |
2147 | * all memory/cpu of this node are removed, we can offline this | |
2148 | * node now. | |
2149 | */ | |
2150 | node_set_offline(nid); | |
2151 | unregister_one_node(nid); | |
2152 | } | |
90b30cdc | 2153 | EXPORT_SYMBOL(try_offline_node); |
60a5a19e | 2154 | |
e1c158e4 | 2155 | static int __ref try_remove_memory(u64 start, u64 size) |
bbc76be6 | 2156 | { |
1a8c64e1 | 2157 | struct memory_block *mem; |
e1c158e4 | 2158 | int rc = 0, nid = NUMA_NO_NODE; |
1a8c64e1 | 2159 | struct vmem_altmap *altmap = NULL; |
993c1aad | 2160 | |
27356f54 TK |
2161 | BUG_ON(check_hotplug_memory_range(start, size)); |
2162 | ||
6677e3ea | 2163 | /* |
242831eb | 2164 | * All memory blocks must be offlined before removing memory. Check |
eca499ab | 2165 | * whether all memory blocks in question are offline and return error |
242831eb | 2166 | * if this is not the case. |
e1c158e4 DH |
2167 | * |
2168 | * While at it, determine the nid. Note that if we'd have mixed nodes, | |
2169 | * we'd only try to offline the last determined one -- which is good | |
2170 | * enough for the cases we care about. | |
6677e3ea | 2171 | */ |
e1c158e4 | 2172 | rc = walk_memory_blocks(start, size, &nid, check_memblock_offlined_cb); |
eca499ab | 2173 | if (rc) |
b4223a51 | 2174 | return rc; |
6677e3ea | 2175 | |
a08a2ae3 OS |
2176 | /* |
2177 | * We only support removing memory added with MHP_MEMMAP_ON_MEMORY in | |
2178 | * the same granularity it was added - a single memory block. | |
2179 | */ | |
6e02c46b | 2180 | if (mhp_memmap_on_memory()) { |
1a8c64e1 AK |
2181 | rc = walk_memory_blocks(start, size, &mem, test_has_altmap_cb); |
2182 | if (rc) { | |
a08a2ae3 OS |
2183 | if (size != memory_block_size_bytes()) { |
2184 | pr_warn("Refuse to remove %#llx - %#llx," | |
2185 | "wrong granularity\n", | |
2186 | start, start + size); | |
2187 | return -EINVAL; | |
2188 | } | |
1a8c64e1 | 2189 | altmap = mem->altmap; |
a08a2ae3 | 2190 | /* |
1a8c64e1 AK |
2191 | * Mark altmap NULL so that we can add a debug |
2192 | * check on memblock free. | |
a08a2ae3 | 2193 | */ |
1a8c64e1 | 2194 | mem->altmap = NULL; |
a08a2ae3 OS |
2195 | } |
2196 | } | |
2197 | ||
46c66c4b YI |
2198 | /* remove memmap entry */ |
2199 | firmware_map_remove(start, start + size, "System RAM"); | |
4c4b7f9b | 2200 | |
f1037ec0 DW |
2201 | /* |
2202 | * Memory block device removal under the device_hotplug_lock is | |
2203 | * a barrier against racing online attempts. | |
2204 | */ | |
4c4b7f9b | 2205 | remove_memory_block_devices(start, size); |
46c66c4b | 2206 | |
f1037ec0 DW |
2207 | mem_hotplug_begin(); |
2208 | ||
65a2aa5f | 2209 | arch_remove_memory(start, size, altmap); |
52219aea | 2210 | |
1a8c64e1 AK |
2211 | /* Verify that all vmemmap pages have actually been freed. */ |
2212 | if (altmap) { | |
2213 | WARN(altmap->alloc, "Altmap not fully unmapped"); | |
2214 | kfree(altmap); | |
2215 | } | |
2216 | ||
52219aea | 2217 | if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) { |
3ecc6834 | 2218 | memblock_phys_free(start, size); |
52219aea DH |
2219 | memblock_remove(start, size); |
2220 | } | |
2221 | ||
cb8e3c8b | 2222 | release_mem_region_adjustable(start, size); |
24d335ca | 2223 | |
e1c158e4 DH |
2224 | if (nid != NUMA_NO_NODE) |
2225 | try_offline_node(nid); | |
60a5a19e | 2226 | |
bfc8c901 | 2227 | mem_hotplug_done(); |
b4223a51 | 2228 | return 0; |
71088785 | 2229 | } |
d15e5926 | 2230 | |
eca499ab | 2231 | /** |
5640c9ca | 2232 | * __remove_memory - Remove memory if every memory block is offline |
eca499ab PT |
2233 | * @start: physical address of the region to remove |
2234 | * @size: size of the region to remove | |
2235 | * | |
2236 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
2237 | * and online/offline operations before this call, as required by | |
2238 | * try_offline_node(). | |
2239 | */ | |
e1c158e4 | 2240 | void __remove_memory(u64 start, u64 size) |
eca499ab PT |
2241 | { |
2242 | ||
2243 | /* | |
29a90db9 | 2244 | * trigger BUG() if some memory is not offlined prior to calling this |
eca499ab PT |
2245 | * function |
2246 | */ | |
e1c158e4 | 2247 | if (try_remove_memory(start, size)) |
eca499ab PT |
2248 | BUG(); |
2249 | } | |
2250 | ||
2251 | /* | |
2252 | * Remove memory if every memory block is offline, otherwise return -EBUSY is | |
2253 | * some memory is not offline | |
2254 | */ | |
e1c158e4 | 2255 | int remove_memory(u64 start, u64 size) |
d15e5926 | 2256 | { |
eca499ab PT |
2257 | int rc; |
2258 | ||
d15e5926 | 2259 | lock_device_hotplug(); |
e1c158e4 | 2260 | rc = try_remove_memory(start, size); |
d15e5926 | 2261 | unlock_device_hotplug(); |
eca499ab PT |
2262 | |
2263 | return rc; | |
d15e5926 | 2264 | } |
71088785 | 2265 | EXPORT_SYMBOL_GPL(remove_memory); |
08b3acd7 | 2266 | |
8dc4bb58 DH |
2267 | static int try_offline_memory_block(struct memory_block *mem, void *arg) |
2268 | { | |
2269 | uint8_t online_type = MMOP_ONLINE_KERNEL; | |
2270 | uint8_t **online_types = arg; | |
2271 | struct page *page; | |
2272 | int rc; | |
2273 | ||
2274 | /* | |
2275 | * Sense the online_type via the zone of the memory block. Offlining | |
2276 | * with multiple zones within one memory block will be rejected | |
2277 | * by offlining code ... so we don't care about that. | |
2278 | */ | |
2279 | page = pfn_to_online_page(section_nr_to_pfn(mem->start_section_nr)); | |
2280 | if (page && zone_idx(page_zone(page)) == ZONE_MOVABLE) | |
2281 | online_type = MMOP_ONLINE_MOVABLE; | |
2282 | ||
2283 | rc = device_offline(&mem->dev); | |
2284 | /* | |
2285 | * Default is MMOP_OFFLINE - change it only if offlining succeeded, | |
2286 | * so try_reonline_memory_block() can do the right thing. | |
2287 | */ | |
2288 | if (!rc) | |
2289 | **online_types = online_type; | |
2290 | ||
2291 | (*online_types)++; | |
2292 | /* Ignore if already offline. */ | |
2293 | return rc < 0 ? rc : 0; | |
2294 | } | |
2295 | ||
2296 | static int try_reonline_memory_block(struct memory_block *mem, void *arg) | |
2297 | { | |
2298 | uint8_t **online_types = arg; | |
2299 | int rc; | |
2300 | ||
2301 | if (**online_types != MMOP_OFFLINE) { | |
2302 | mem->online_type = **online_types; | |
2303 | rc = device_online(&mem->dev); | |
2304 | if (rc < 0) | |
2305 | pr_warn("%s: Failed to re-online memory: %d", | |
2306 | __func__, rc); | |
2307 | } | |
2308 | ||
2309 | /* Continue processing all remaining memory blocks. */ | |
2310 | (*online_types)++; | |
2311 | return 0; | |
2312 | } | |
2313 | ||
08b3acd7 | 2314 | /* |
8dc4bb58 DH |
2315 | * Try to offline and remove memory. Might take a long time to finish in case |
2316 | * memory is still in use. Primarily useful for memory devices that logically | |
2317 | * unplugged all memory (so it's no longer in use) and want to offline + remove | |
2318 | * that memory. | |
08b3acd7 | 2319 | */ |
e1c158e4 | 2320 | int offline_and_remove_memory(u64 start, u64 size) |
08b3acd7 | 2321 | { |
8dc4bb58 DH |
2322 | const unsigned long mb_count = size / memory_block_size_bytes(); |
2323 | uint8_t *online_types, *tmp; | |
2324 | int rc; | |
08b3acd7 DH |
2325 | |
2326 | if (!IS_ALIGNED(start, memory_block_size_bytes()) || | |
8dc4bb58 DH |
2327 | !IS_ALIGNED(size, memory_block_size_bytes()) || !size) |
2328 | return -EINVAL; | |
2329 | ||
2330 | /* | |
2331 | * We'll remember the old online type of each memory block, so we can | |
2332 | * try to revert whatever we did when offlining one memory block fails | |
2333 | * after offlining some others succeeded. | |
2334 | */ | |
2335 | online_types = kmalloc_array(mb_count, sizeof(*online_types), | |
2336 | GFP_KERNEL); | |
2337 | if (!online_types) | |
2338 | return -ENOMEM; | |
2339 | /* | |
2340 | * Initialize all states to MMOP_OFFLINE, so when we abort processing in | |
2341 | * try_offline_memory_block(), we'll skip all unprocessed blocks in | |
2342 | * try_reonline_memory_block(). | |
2343 | */ | |
2344 | memset(online_types, MMOP_OFFLINE, mb_count); | |
08b3acd7 DH |
2345 | |
2346 | lock_device_hotplug(); | |
8dc4bb58 DH |
2347 | |
2348 | tmp = online_types; | |
2349 | rc = walk_memory_blocks(start, size, &tmp, try_offline_memory_block); | |
08b3acd7 DH |
2350 | |
2351 | /* | |
8dc4bb58 | 2352 | * In case we succeeded to offline all memory, remove it. |
08b3acd7 DH |
2353 | * This cannot fail as it cannot get onlined in the meantime. |
2354 | */ | |
2355 | if (!rc) { | |
e1c158e4 | 2356 | rc = try_remove_memory(start, size); |
8dc4bb58 DH |
2357 | if (rc) |
2358 | pr_err("%s: Failed to remove memory: %d", __func__, rc); | |
2359 | } | |
2360 | ||
2361 | /* | |
2362 | * Rollback what we did. While memory onlining might theoretically fail | |
2363 | * (nacked by a notifier), it barely ever happens. | |
2364 | */ | |
2365 | if (rc) { | |
2366 | tmp = online_types; | |
2367 | walk_memory_blocks(start, size, &tmp, | |
2368 | try_reonline_memory_block); | |
08b3acd7 DH |
2369 | } |
2370 | unlock_device_hotplug(); | |
2371 | ||
8dc4bb58 | 2372 | kfree(online_types); |
08b3acd7 DH |
2373 | return rc; |
2374 | } | |
2375 | EXPORT_SYMBOL_GPL(offline_and_remove_memory); | |
aba6efc4 | 2376 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |