]>
Commit | Line | Data |
---|---|---|
3947be19 DH |
1 | /* |
2 | * linux/mm/memory_hotplug.c | |
3 | * | |
4 | * Copyright (C) | |
5 | */ | |
6 | ||
3947be19 DH |
7 | #include <linux/stddef.h> |
8 | #include <linux/mm.h> | |
9 | #include <linux/swap.h> | |
10 | #include <linux/interrupt.h> | |
11 | #include <linux/pagemap.h> | |
3947be19 | 12 | #include <linux/compiler.h> |
b95f1b31 | 13 | #include <linux/export.h> |
3947be19 | 14 | #include <linux/pagevec.h> |
2d1d43f6 | 15 | #include <linux/writeback.h> |
3947be19 DH |
16 | #include <linux/slab.h> |
17 | #include <linux/sysctl.h> | |
18 | #include <linux/cpu.h> | |
19 | #include <linux/memory.h> | |
4b94ffdc | 20 | #include <linux/memremap.h> |
3947be19 DH |
21 | #include <linux/memory_hotplug.h> |
22 | #include <linux/highmem.h> | |
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> |
f784a3f1 | 35 | #include <linux/bootmem.h> |
698b1b30 | 36 | #include <linux/compaction.h> |
3947be19 DH |
37 | |
38 | #include <asm/tlbflush.h> | |
39 | ||
1e5ad9a3 AB |
40 | #include "internal.h" |
41 | ||
9d0ad8ca DK |
42 | /* |
43 | * online_page_callback contains pointer to current page onlining function. | |
44 | * Initially it is generic_online_page(). If it is required it could be | |
45 | * changed by calling set_online_page_callback() for callback registration | |
46 | * and restore_online_page_callback() for generic callback restore. | |
47 | */ | |
48 | ||
49 | static void generic_online_page(struct page *page); | |
50 | ||
51 | static online_page_callback_t online_page_callback = generic_online_page; | |
bfc8c901 | 52 | static DEFINE_MUTEX(online_page_callback_lock); |
9d0ad8ca | 53 | |
bfc8c901 VD |
54 | /* The same as the cpu_hotplug lock, but for memory hotplug. */ |
55 | static struct { | |
56 | struct task_struct *active_writer; | |
57 | struct mutex lock; /* Synchronizes accesses to refcount, */ | |
58 | /* | |
59 | * Also blocks the new readers during | |
60 | * an ongoing mem hotplug operation. | |
61 | */ | |
62 | int refcount; | |
63 | ||
64 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
65 | struct lockdep_map dep_map; | |
66 | #endif | |
67 | } mem_hotplug = { | |
68 | .active_writer = NULL, | |
69 | .lock = __MUTEX_INITIALIZER(mem_hotplug.lock), | |
70 | .refcount = 0, | |
71 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
72 | .dep_map = {.name = "mem_hotplug.lock" }, | |
73 | #endif | |
74 | }; | |
75 | ||
76 | /* Lockdep annotations for get/put_online_mems() and mem_hotplug_begin/end() */ | |
77 | #define memhp_lock_acquire_read() lock_map_acquire_read(&mem_hotplug.dep_map) | |
78 | #define memhp_lock_acquire() lock_map_acquire(&mem_hotplug.dep_map) | |
79 | #define memhp_lock_release() lock_map_release(&mem_hotplug.dep_map) | |
80 | ||
8604d9e5 | 81 | #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE |
31bc3858 | 82 | bool memhp_auto_online; |
8604d9e5 VK |
83 | #else |
84 | bool memhp_auto_online = true; | |
85 | #endif | |
31bc3858 VK |
86 | EXPORT_SYMBOL_GPL(memhp_auto_online); |
87 | ||
86dd995d VK |
88 | static int __init setup_memhp_default_state(char *str) |
89 | { | |
90 | if (!strcmp(str, "online")) | |
91 | memhp_auto_online = true; | |
92 | else if (!strcmp(str, "offline")) | |
93 | memhp_auto_online = false; | |
94 | ||
95 | return 1; | |
96 | } | |
97 | __setup("memhp_default_state=", setup_memhp_default_state); | |
98 | ||
bfc8c901 VD |
99 | void get_online_mems(void) |
100 | { | |
101 | might_sleep(); | |
102 | if (mem_hotplug.active_writer == current) | |
103 | return; | |
104 | memhp_lock_acquire_read(); | |
105 | mutex_lock(&mem_hotplug.lock); | |
106 | mem_hotplug.refcount++; | |
107 | mutex_unlock(&mem_hotplug.lock); | |
108 | ||
109 | } | |
20d6c96b | 110 | |
bfc8c901 | 111 | void put_online_mems(void) |
20d6c96b | 112 | { |
bfc8c901 VD |
113 | if (mem_hotplug.active_writer == current) |
114 | return; | |
115 | mutex_lock(&mem_hotplug.lock); | |
116 | ||
117 | if (WARN_ON(!mem_hotplug.refcount)) | |
118 | mem_hotplug.refcount++; /* try to fix things up */ | |
119 | ||
120 | if (!--mem_hotplug.refcount && unlikely(mem_hotplug.active_writer)) | |
121 | wake_up_process(mem_hotplug.active_writer); | |
122 | mutex_unlock(&mem_hotplug.lock); | |
123 | memhp_lock_release(); | |
124 | ||
20d6c96b KM |
125 | } |
126 | ||
30467e0b | 127 | void mem_hotplug_begin(void) |
20d6c96b | 128 | { |
3fc21924 DW |
129 | assert_held_device_hotplug(); |
130 | ||
bfc8c901 VD |
131 | mem_hotplug.active_writer = current; |
132 | ||
133 | memhp_lock_acquire(); | |
134 | for (;;) { | |
135 | mutex_lock(&mem_hotplug.lock); | |
136 | if (likely(!mem_hotplug.refcount)) | |
137 | break; | |
138 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
139 | mutex_unlock(&mem_hotplug.lock); | |
140 | schedule(); | |
141 | } | |
20d6c96b KM |
142 | } |
143 | ||
30467e0b | 144 | void mem_hotplug_done(void) |
bfc8c901 VD |
145 | { |
146 | mem_hotplug.active_writer = NULL; | |
147 | mutex_unlock(&mem_hotplug.lock); | |
148 | memhp_lock_release(); | |
149 | } | |
20d6c96b | 150 | |
45e0b78b KM |
151 | /* add this memory to iomem resource */ |
152 | static struct resource *register_memory_resource(u64 start, u64 size) | |
153 | { | |
154 | struct resource *res; | |
155 | res = kzalloc(sizeof(struct resource), GFP_KERNEL); | |
6f754ba4 VK |
156 | if (!res) |
157 | return ERR_PTR(-ENOMEM); | |
45e0b78b KM |
158 | |
159 | res->name = "System RAM"; | |
160 | res->start = start; | |
161 | res->end = start + size - 1; | |
782b8664 | 162 | res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY; |
45e0b78b | 163 | if (request_resource(&iomem_resource, res) < 0) { |
4996eed8 | 164 | pr_debug("System RAM resource %pR cannot be added\n", res); |
45e0b78b | 165 | kfree(res); |
6f754ba4 | 166 | return ERR_PTR(-EEXIST); |
45e0b78b KM |
167 | } |
168 | return res; | |
169 | } | |
170 | ||
171 | static void release_memory_resource(struct resource *res) | |
172 | { | |
173 | if (!res) | |
174 | return; | |
175 | release_resource(res); | |
176 | kfree(res); | |
177 | return; | |
178 | } | |
179 | ||
53947027 | 180 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
46723bfa YI |
181 | void get_page_bootmem(unsigned long info, struct page *page, |
182 | unsigned long type) | |
04753278 | 183 | { |
ddffe98d | 184 | page->freelist = (void *)type; |
04753278 YG |
185 | SetPagePrivate(page); |
186 | set_page_private(page, info); | |
fe896d18 | 187 | page_ref_inc(page); |
04753278 YG |
188 | } |
189 | ||
170a5a7e | 190 | void put_page_bootmem(struct page *page) |
04753278 | 191 | { |
5f24ce5f | 192 | unsigned long type; |
04753278 | 193 | |
ddffe98d | 194 | type = (unsigned long) page->freelist; |
5f24ce5f AA |
195 | BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || |
196 | type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); | |
04753278 | 197 | |
fe896d18 | 198 | if (page_ref_dec_return(page) == 1) { |
ddffe98d | 199 | page->freelist = NULL; |
04753278 YG |
200 | ClearPagePrivate(page); |
201 | set_page_private(page, 0); | |
5f24ce5f | 202 | INIT_LIST_HEAD(&page->lru); |
170a5a7e | 203 | free_reserved_page(page); |
04753278 | 204 | } |
04753278 YG |
205 | } |
206 | ||
46723bfa YI |
207 | #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE |
208 | #ifndef CONFIG_SPARSEMEM_VMEMMAP | |
d92bc318 | 209 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
04753278 YG |
210 | { |
211 | unsigned long *usemap, mapsize, section_nr, i; | |
212 | struct mem_section *ms; | |
213 | struct page *page, *memmap; | |
214 | ||
04753278 YG |
215 | section_nr = pfn_to_section_nr(start_pfn); |
216 | ms = __nr_to_section(section_nr); | |
217 | ||
218 | /* Get section's memmap address */ | |
219 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
220 | ||
221 | /* | |
222 | * Get page for the memmap's phys address | |
223 | * XXX: need more consideration for sparse_vmemmap... | |
224 | */ | |
225 | page = virt_to_page(memmap); | |
226 | mapsize = sizeof(struct page) * PAGES_PER_SECTION; | |
227 | mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; | |
228 | ||
229 | /* remember memmap's page */ | |
230 | for (i = 0; i < mapsize; i++, page++) | |
231 | get_page_bootmem(section_nr, page, SECTION_INFO); | |
232 | ||
233 | usemap = __nr_to_section(section_nr)->pageblock_flags; | |
234 | page = virt_to_page(usemap); | |
235 | ||
236 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
237 | ||
238 | for (i = 0; i < mapsize; i++, page++) | |
af370fb8 | 239 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
04753278 YG |
240 | |
241 | } | |
46723bfa YI |
242 | #else /* CONFIG_SPARSEMEM_VMEMMAP */ |
243 | static void register_page_bootmem_info_section(unsigned long start_pfn) | |
244 | { | |
245 | unsigned long *usemap, mapsize, section_nr, i; | |
246 | struct mem_section *ms; | |
247 | struct page *page, *memmap; | |
248 | ||
249 | if (!pfn_valid(start_pfn)) | |
250 | return; | |
251 | ||
252 | section_nr = pfn_to_section_nr(start_pfn); | |
253 | ms = __nr_to_section(section_nr); | |
254 | ||
255 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
256 | ||
257 | register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION); | |
258 | ||
259 | usemap = __nr_to_section(section_nr)->pageblock_flags; | |
260 | page = virt_to_page(usemap); | |
261 | ||
262 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
263 | ||
264 | for (i = 0; i < mapsize; i++, page++) | |
265 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); | |
266 | } | |
267 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ | |
04753278 | 268 | |
7ded384a | 269 | void __init register_page_bootmem_info_node(struct pglist_data *pgdat) |
04753278 YG |
270 | { |
271 | unsigned long i, pfn, end_pfn, nr_pages; | |
272 | int node = pgdat->node_id; | |
273 | struct page *page; | |
04753278 YG |
274 | |
275 | nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; | |
276 | page = virt_to_page(pgdat); | |
277 | ||
278 | for (i = 0; i < nr_pages; i++, page++) | |
279 | get_page_bootmem(node, page, NODE_INFO); | |
280 | ||
04753278 | 281 | pfn = pgdat->node_start_pfn; |
c1f19495 | 282 | end_pfn = pgdat_end_pfn(pgdat); |
04753278 | 283 | |
7e9f5eb0 | 284 | /* register section info */ |
f14851af | 285 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
286 | /* | |
287 | * Some platforms can assign the same pfn to multiple nodes - on | |
288 | * node0 as well as nodeN. To avoid registering a pfn against | |
289 | * multiple nodes we check that this pfn does not already | |
7e9f5eb0 | 290 | * reside in some other nodes. |
f14851af | 291 | */ |
f65e91df | 292 | if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node)) |
f14851af | 293 | register_page_bootmem_info_section(pfn); |
294 | } | |
04753278 | 295 | } |
46723bfa | 296 | #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */ |
04753278 | 297 | |
f2765404 FF |
298 | static void __meminit grow_zone_span(struct zone *zone, unsigned long start_pfn, |
299 | unsigned long end_pfn) | |
76cdd58e HC |
300 | { |
301 | unsigned long old_zone_end_pfn; | |
302 | ||
303 | zone_span_writelock(zone); | |
304 | ||
c33bc315 | 305 | old_zone_end_pfn = zone_end_pfn(zone); |
8080fc03 | 306 | if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn) |
76cdd58e HC |
307 | zone->zone_start_pfn = start_pfn; |
308 | ||
309 | zone->spanned_pages = max(old_zone_end_pfn, end_pfn) - | |
310 | zone->zone_start_pfn; | |
311 | ||
312 | zone_span_writeunlock(zone); | |
313 | } | |
314 | ||
511c2aba LJ |
315 | static void resize_zone(struct zone *zone, unsigned long start_pfn, |
316 | unsigned long end_pfn) | |
317 | { | |
318 | zone_span_writelock(zone); | |
319 | ||
e455a9b9 LJ |
320 | if (end_pfn - start_pfn) { |
321 | zone->zone_start_pfn = start_pfn; | |
322 | zone->spanned_pages = end_pfn - start_pfn; | |
323 | } else { | |
324 | /* | |
325 | * make it consist as free_area_init_core(), | |
326 | * if spanned_pages = 0, then keep start_pfn = 0 | |
327 | */ | |
328 | zone->zone_start_pfn = 0; | |
329 | zone->spanned_pages = 0; | |
330 | } | |
511c2aba LJ |
331 | |
332 | zone_span_writeunlock(zone); | |
333 | } | |
334 | ||
335 | static void fix_zone_id(struct zone *zone, unsigned long start_pfn, | |
336 | unsigned long end_pfn) | |
337 | { | |
338 | enum zone_type zid = zone_idx(zone); | |
339 | int nid = zone->zone_pgdat->node_id; | |
340 | unsigned long pfn; | |
341 | ||
342 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
343 | set_page_links(pfn_to_page(pfn), zid, nid, pfn); | |
344 | } | |
345 | ||
f6bbb78e | 346 | /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or |
9e43aa2b | 347 | * alloc_bootmem_node_nopanic()/memblock_virt_alloc_node_nopanic() */ |
f6bbb78e CS |
348 | static int __ref ensure_zone_is_initialized(struct zone *zone, |
349 | unsigned long start_pfn, unsigned long num_pages) | |
350 | { | |
351 | if (!zone_is_initialized(zone)) | |
b171e409 YB |
352 | return init_currently_empty_zone(zone, start_pfn, num_pages); |
353 | ||
f6bbb78e CS |
354 | return 0; |
355 | } | |
356 | ||
e455a9b9 | 357 | static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2, |
511c2aba LJ |
358 | unsigned long start_pfn, unsigned long end_pfn) |
359 | { | |
e455a9b9 | 360 | int ret; |
511c2aba | 361 | unsigned long flags; |
e455a9b9 LJ |
362 | unsigned long z1_start_pfn; |
363 | ||
64dd1b29 CS |
364 | ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn); |
365 | if (ret) | |
366 | return ret; | |
511c2aba LJ |
367 | |
368 | pgdat_resize_lock(z1->zone_pgdat, &flags); | |
369 | ||
370 | /* can't move pfns which are higher than @z2 */ | |
108bcc96 | 371 | if (end_pfn > zone_end_pfn(z2)) |
511c2aba | 372 | goto out_fail; |
834405c3 | 373 | /* the move out part must be at the left most of @z2 */ |
511c2aba LJ |
374 | if (start_pfn > z2->zone_start_pfn) |
375 | goto out_fail; | |
376 | /* must included/overlap */ | |
377 | if (end_pfn <= z2->zone_start_pfn) | |
378 | goto out_fail; | |
379 | ||
e455a9b9 | 380 | /* use start_pfn for z1's start_pfn if z1 is empty */ |
8080fc03 | 381 | if (!zone_is_empty(z1)) |
e455a9b9 LJ |
382 | z1_start_pfn = z1->zone_start_pfn; |
383 | else | |
384 | z1_start_pfn = start_pfn; | |
385 | ||
386 | resize_zone(z1, z1_start_pfn, end_pfn); | |
108bcc96 | 387 | resize_zone(z2, end_pfn, zone_end_pfn(z2)); |
511c2aba LJ |
388 | |
389 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
390 | ||
391 | fix_zone_id(z1, start_pfn, end_pfn); | |
392 | ||
393 | return 0; | |
394 | out_fail: | |
395 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
396 | return -1; | |
397 | } | |
398 | ||
e455a9b9 | 399 | static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2, |
511c2aba LJ |
400 | unsigned long start_pfn, unsigned long end_pfn) |
401 | { | |
e455a9b9 | 402 | int ret; |
511c2aba | 403 | unsigned long flags; |
e455a9b9 LJ |
404 | unsigned long z2_end_pfn; |
405 | ||
64dd1b29 CS |
406 | ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn); |
407 | if (ret) | |
408 | return ret; | |
511c2aba LJ |
409 | |
410 | pgdat_resize_lock(z1->zone_pgdat, &flags); | |
411 | ||
412 | /* can't move pfns which are lower than @z1 */ | |
413 | if (z1->zone_start_pfn > start_pfn) | |
414 | goto out_fail; | |
415 | /* the move out part mast at the right most of @z1 */ | |
108bcc96 | 416 | if (zone_end_pfn(z1) > end_pfn) |
511c2aba LJ |
417 | goto out_fail; |
418 | /* must included/overlap */ | |
108bcc96 | 419 | if (start_pfn >= zone_end_pfn(z1)) |
511c2aba LJ |
420 | goto out_fail; |
421 | ||
e455a9b9 | 422 | /* use end_pfn for z2's end_pfn if z2 is empty */ |
8080fc03 | 423 | if (!zone_is_empty(z2)) |
108bcc96 | 424 | z2_end_pfn = zone_end_pfn(z2); |
e455a9b9 LJ |
425 | else |
426 | z2_end_pfn = end_pfn; | |
427 | ||
511c2aba | 428 | resize_zone(z1, z1->zone_start_pfn, start_pfn); |
e455a9b9 | 429 | resize_zone(z2, start_pfn, z2_end_pfn); |
511c2aba LJ |
430 | |
431 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
432 | ||
433 | fix_zone_id(z2, start_pfn, end_pfn); | |
434 | ||
435 | return 0; | |
436 | out_fail: | |
437 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
438 | return -1; | |
439 | } | |
440 | ||
e51e6c8f RA |
441 | static struct zone * __meminit move_pfn_range(int zone_shift, |
442 | unsigned long start_pfn, unsigned long end_pfn) | |
443 | { | |
444 | struct zone *zone = page_zone(pfn_to_page(start_pfn)); | |
445 | int ret = 0; | |
446 | ||
447 | if (zone_shift < 0) | |
448 | ret = move_pfn_range_left(zone + zone_shift, zone, | |
449 | start_pfn, end_pfn); | |
450 | else if (zone_shift) | |
451 | ret = move_pfn_range_right(zone, zone + zone_shift, | |
452 | start_pfn, end_pfn); | |
453 | ||
454 | if (ret) | |
455 | return NULL; | |
456 | ||
457 | return zone + zone_shift; | |
458 | } | |
459 | ||
f2765404 FF |
460 | static void __meminit grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn, |
461 | unsigned long end_pfn) | |
76cdd58e | 462 | { |
83285c72 | 463 | unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat); |
76cdd58e | 464 | |
712cd386 | 465 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) |
76cdd58e HC |
466 | pgdat->node_start_pfn = start_pfn; |
467 | ||
468 | pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) - | |
469 | pgdat->node_start_pfn; | |
470 | } | |
471 | ||
31168481 | 472 | static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn) |
3947be19 DH |
473 | { |
474 | struct pglist_data *pgdat = zone->zone_pgdat; | |
475 | int nr_pages = PAGES_PER_SECTION; | |
476 | int nid = pgdat->node_id; | |
477 | int zone_type; | |
e298ff75 | 478 | unsigned long flags, pfn; |
64dd1b29 | 479 | int ret; |
3947be19 DH |
480 | |
481 | zone_type = zone - pgdat->node_zones; | |
64dd1b29 CS |
482 | ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages); |
483 | if (ret) | |
484 | return ret; | |
76cdd58e | 485 | |
76cdd58e HC |
486 | pgdat_resize_lock(zone->zone_pgdat, &flags); |
487 | grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages); | |
488 | grow_pgdat_span(zone->zone_pgdat, phys_start_pfn, | |
489 | phys_start_pfn + nr_pages); | |
490 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
a2f3aa02 DH |
491 | memmap_init_zone(nr_pages, nid, zone_type, |
492 | phys_start_pfn, MEMMAP_HOTPLUG); | |
e298ff75 MG |
493 | |
494 | /* online_page_range is called later and expects pages reserved */ | |
495 | for (pfn = phys_start_pfn; pfn < phys_start_pfn + nr_pages; pfn++) { | |
496 | if (!pfn_valid(pfn)) | |
497 | continue; | |
498 | ||
499 | SetPageReserved(pfn_to_page(pfn)); | |
500 | } | |
718127cc | 501 | return 0; |
3947be19 DH |
502 | } |
503 | ||
c04fc586 GH |
504 | static int __meminit __add_section(int nid, struct zone *zone, |
505 | unsigned long phys_start_pfn) | |
3947be19 | 506 | { |
3947be19 DH |
507 | int ret; |
508 | ||
ebd15302 KH |
509 | if (pfn_valid(phys_start_pfn)) |
510 | return -EEXIST; | |
511 | ||
85b35fea | 512 | ret = sparse_add_one_section(zone, phys_start_pfn); |
3947be19 DH |
513 | |
514 | if (ret < 0) | |
515 | return ret; | |
516 | ||
718127cc YG |
517 | ret = __add_zone(zone, phys_start_pfn); |
518 | ||
519 | if (ret < 0) | |
520 | return ret; | |
521 | ||
c04fc586 | 522 | return register_new_memory(nid, __pfn_to_section(phys_start_pfn)); |
3947be19 DH |
523 | } |
524 | ||
4edd7cef DR |
525 | /* |
526 | * Reasonably generic function for adding memory. It is | |
527 | * expected that archs that support memory hotplug will | |
528 | * call this function after deciding the zone to which to | |
529 | * add the new pages. | |
530 | */ | |
531 | int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn, | |
532 | unsigned long nr_pages) | |
533 | { | |
534 | unsigned long i; | |
535 | int err = 0; | |
536 | int start_sec, end_sec; | |
4b94ffdc DW |
537 | struct vmem_altmap *altmap; |
538 | ||
7cf91a98 JK |
539 | clear_zone_contiguous(zone); |
540 | ||
4edd7cef DR |
541 | /* during initialize mem_map, align hot-added range to section */ |
542 | start_sec = pfn_to_section_nr(phys_start_pfn); | |
543 | end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1); | |
544 | ||
4b94ffdc DW |
545 | altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn)); |
546 | if (altmap) { | |
547 | /* | |
548 | * Validate altmap is within bounds of the total request | |
549 | */ | |
550 | if (altmap->base_pfn != phys_start_pfn | |
551 | || vmem_altmap_offset(altmap) > nr_pages) { | |
552 | pr_warn_once("memory add fail, invalid altmap\n"); | |
7cf91a98 JK |
553 | err = -EINVAL; |
554 | goto out; | |
4b94ffdc DW |
555 | } |
556 | altmap->alloc = 0; | |
557 | } | |
558 | ||
4edd7cef | 559 | for (i = start_sec; i <= end_sec; i++) { |
19c07d5e | 560 | err = __add_section(nid, zone, section_nr_to_pfn(i)); |
4edd7cef DR |
561 | |
562 | /* | |
563 | * EEXIST is finally dealt with by ioresource collision | |
564 | * check. see add_memory() => register_memory_resource() | |
565 | * Warning will be printed if there is collision. | |
566 | */ | |
567 | if (err && (err != -EEXIST)) | |
568 | break; | |
569 | err = 0; | |
570 | } | |
c435a390 | 571 | vmemmap_populate_print_last(); |
7cf91a98 JK |
572 | out: |
573 | set_zone_contiguous(zone); | |
4edd7cef DR |
574 | return err; |
575 | } | |
576 | EXPORT_SYMBOL_GPL(__add_pages); | |
577 | ||
578 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
815121d2 YI |
579 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ |
580 | static int find_smallest_section_pfn(int nid, struct zone *zone, | |
581 | unsigned long start_pfn, | |
582 | unsigned long end_pfn) | |
583 | { | |
584 | struct mem_section *ms; | |
585 | ||
586 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) { | |
587 | ms = __pfn_to_section(start_pfn); | |
588 | ||
589 | if (unlikely(!valid_section(ms))) | |
590 | continue; | |
591 | ||
592 | if (unlikely(pfn_to_nid(start_pfn) != nid)) | |
593 | continue; | |
594 | ||
595 | if (zone && zone != page_zone(pfn_to_page(start_pfn))) | |
596 | continue; | |
597 | ||
598 | return start_pfn; | |
599 | } | |
600 | ||
601 | return 0; | |
602 | } | |
603 | ||
604 | /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ | |
605 | static int find_biggest_section_pfn(int nid, struct zone *zone, | |
606 | unsigned long start_pfn, | |
607 | unsigned long end_pfn) | |
608 | { | |
609 | struct mem_section *ms; | |
610 | unsigned long pfn; | |
611 | ||
612 | /* pfn is the end pfn of a memory section. */ | |
613 | pfn = end_pfn - 1; | |
614 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) { | |
615 | ms = __pfn_to_section(pfn); | |
616 | ||
617 | if (unlikely(!valid_section(ms))) | |
618 | continue; | |
619 | ||
620 | if (unlikely(pfn_to_nid(pfn) != nid)) | |
621 | continue; | |
622 | ||
623 | if (zone && zone != page_zone(pfn_to_page(pfn))) | |
624 | continue; | |
625 | ||
626 | return pfn; | |
627 | } | |
628 | ||
629 | return 0; | |
630 | } | |
631 | ||
632 | static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, | |
633 | unsigned long end_pfn) | |
634 | { | |
c33bc315 XQ |
635 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
636 | unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */ | |
637 | unsigned long zone_end_pfn = z; | |
815121d2 YI |
638 | unsigned long pfn; |
639 | struct mem_section *ms; | |
640 | int nid = zone_to_nid(zone); | |
641 | ||
642 | zone_span_writelock(zone); | |
643 | if (zone_start_pfn == start_pfn) { | |
644 | /* | |
645 | * If the section is smallest section in the zone, it need | |
646 | * shrink zone->zone_start_pfn and zone->zone_spanned_pages. | |
647 | * In this case, we find second smallest valid mem_section | |
648 | * for shrinking zone. | |
649 | */ | |
650 | pfn = find_smallest_section_pfn(nid, zone, end_pfn, | |
651 | zone_end_pfn); | |
652 | if (pfn) { | |
653 | zone->zone_start_pfn = pfn; | |
654 | zone->spanned_pages = zone_end_pfn - pfn; | |
655 | } | |
656 | } else if (zone_end_pfn == end_pfn) { | |
657 | /* | |
658 | * If the section is biggest section in the zone, it need | |
659 | * shrink zone->spanned_pages. | |
660 | * In this case, we find second biggest valid mem_section for | |
661 | * shrinking zone. | |
662 | */ | |
663 | pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn, | |
664 | start_pfn); | |
665 | if (pfn) | |
666 | zone->spanned_pages = pfn - zone_start_pfn + 1; | |
667 | } | |
668 | ||
669 | /* | |
670 | * The section is not biggest or smallest mem_section in the zone, it | |
671 | * only creates a hole in the zone. So in this case, we need not | |
672 | * change the zone. But perhaps, the zone has only hole data. Thus | |
673 | * it check the zone has only hole or not. | |
674 | */ | |
675 | pfn = zone_start_pfn; | |
676 | for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) { | |
677 | ms = __pfn_to_section(pfn); | |
678 | ||
679 | if (unlikely(!valid_section(ms))) | |
680 | continue; | |
681 | ||
682 | if (page_zone(pfn_to_page(pfn)) != zone) | |
683 | continue; | |
684 | ||
685 | /* If the section is current section, it continues the loop */ | |
686 | if (start_pfn == pfn) | |
687 | continue; | |
688 | ||
689 | /* If we find valid section, we have nothing to do */ | |
690 | zone_span_writeunlock(zone); | |
691 | return; | |
692 | } | |
693 | ||
694 | /* The zone has no valid section */ | |
695 | zone->zone_start_pfn = 0; | |
696 | zone->spanned_pages = 0; | |
697 | zone_span_writeunlock(zone); | |
698 | } | |
699 | ||
700 | static void shrink_pgdat_span(struct pglist_data *pgdat, | |
701 | unsigned long start_pfn, unsigned long end_pfn) | |
702 | { | |
83285c72 XQ |
703 | unsigned long pgdat_start_pfn = pgdat->node_start_pfn; |
704 | unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */ | |
705 | unsigned long pgdat_end_pfn = p; | |
815121d2 YI |
706 | unsigned long pfn; |
707 | struct mem_section *ms; | |
708 | int nid = pgdat->node_id; | |
709 | ||
710 | if (pgdat_start_pfn == start_pfn) { | |
711 | /* | |
712 | * If the section is smallest section in the pgdat, it need | |
713 | * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages. | |
714 | * In this case, we find second smallest valid mem_section | |
715 | * for shrinking zone. | |
716 | */ | |
717 | pfn = find_smallest_section_pfn(nid, NULL, end_pfn, | |
718 | pgdat_end_pfn); | |
719 | if (pfn) { | |
720 | pgdat->node_start_pfn = pfn; | |
721 | pgdat->node_spanned_pages = pgdat_end_pfn - pfn; | |
722 | } | |
723 | } else if (pgdat_end_pfn == end_pfn) { | |
724 | /* | |
725 | * If the section is biggest section in the pgdat, it need | |
726 | * shrink pgdat->node_spanned_pages. | |
727 | * In this case, we find second biggest valid mem_section for | |
728 | * shrinking zone. | |
729 | */ | |
730 | pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn, | |
731 | start_pfn); | |
732 | if (pfn) | |
733 | pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1; | |
734 | } | |
735 | ||
736 | /* | |
737 | * If the section is not biggest or smallest mem_section in the pgdat, | |
738 | * it only creates a hole in the pgdat. So in this case, we need not | |
739 | * change the pgdat. | |
740 | * But perhaps, the pgdat has only hole data. Thus it check the pgdat | |
741 | * has only hole or not. | |
742 | */ | |
743 | pfn = pgdat_start_pfn; | |
744 | for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) { | |
745 | ms = __pfn_to_section(pfn); | |
746 | ||
747 | if (unlikely(!valid_section(ms))) | |
748 | continue; | |
749 | ||
750 | if (pfn_to_nid(pfn) != nid) | |
751 | continue; | |
752 | ||
753 | /* If the section is current section, it continues the loop */ | |
754 | if (start_pfn == pfn) | |
755 | continue; | |
756 | ||
757 | /* If we find valid section, we have nothing to do */ | |
758 | return; | |
759 | } | |
760 | ||
761 | /* The pgdat has no valid section */ | |
762 | pgdat->node_start_pfn = 0; | |
763 | pgdat->node_spanned_pages = 0; | |
764 | } | |
765 | ||
766 | static void __remove_zone(struct zone *zone, unsigned long start_pfn) | |
767 | { | |
768 | struct pglist_data *pgdat = zone->zone_pgdat; | |
769 | int nr_pages = PAGES_PER_SECTION; | |
770 | int zone_type; | |
771 | unsigned long flags; | |
772 | ||
773 | zone_type = zone - pgdat->node_zones; | |
774 | ||
775 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
776 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); | |
777 | shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages); | |
778 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
779 | } | |
780 | ||
4b94ffdc DW |
781 | static int __remove_section(struct zone *zone, struct mem_section *ms, |
782 | unsigned long map_offset) | |
ea01ea93 | 783 | { |
815121d2 YI |
784 | unsigned long start_pfn; |
785 | int scn_nr; | |
ea01ea93 BP |
786 | int ret = -EINVAL; |
787 | ||
788 | if (!valid_section(ms)) | |
789 | return ret; | |
790 | ||
791 | ret = unregister_memory_section(ms); | |
792 | if (ret) | |
793 | return ret; | |
794 | ||
815121d2 YI |
795 | scn_nr = __section_nr(ms); |
796 | start_pfn = section_nr_to_pfn(scn_nr); | |
797 | __remove_zone(zone, start_pfn); | |
798 | ||
4b94ffdc | 799 | sparse_remove_one_section(zone, ms, map_offset); |
ea01ea93 BP |
800 | return 0; |
801 | } | |
802 | ||
ea01ea93 BP |
803 | /** |
804 | * __remove_pages() - remove sections of pages from a zone | |
805 | * @zone: zone from which pages need to be removed | |
806 | * @phys_start_pfn: starting pageframe (must be aligned to start of a section) | |
807 | * @nr_pages: number of pages to remove (must be multiple of section size) | |
808 | * | |
809 | * Generic helper function to remove section mappings and sysfs entries | |
810 | * for the section of the memory we are removing. Caller needs to make | |
811 | * sure that pages are marked reserved and zones are adjust properly by | |
812 | * calling offline_pages(). | |
813 | */ | |
814 | int __remove_pages(struct zone *zone, unsigned long phys_start_pfn, | |
815 | unsigned long nr_pages) | |
816 | { | |
fe74ebb1 | 817 | unsigned long i; |
4b94ffdc DW |
818 | unsigned long map_offset = 0; |
819 | int sections_to_remove, ret = 0; | |
820 | ||
821 | /* In the ZONE_DEVICE case device driver owns the memory region */ | |
822 | if (is_dev_zone(zone)) { | |
823 | struct page *page = pfn_to_page(phys_start_pfn); | |
824 | struct vmem_altmap *altmap; | |
825 | ||
826 | altmap = to_vmem_altmap((unsigned long) page); | |
827 | if (altmap) | |
828 | map_offset = vmem_altmap_offset(altmap); | |
829 | } else { | |
830 | resource_size_t start, size; | |
831 | ||
832 | start = phys_start_pfn << PAGE_SHIFT; | |
833 | size = nr_pages * PAGE_SIZE; | |
834 | ||
835 | ret = release_mem_region_adjustable(&iomem_resource, start, | |
836 | size); | |
837 | if (ret) { | |
838 | resource_size_t endres = start + size - 1; | |
839 | ||
840 | pr_warn("Unable to release resource <%pa-%pa> (%d)\n", | |
841 | &start, &endres, ret); | |
842 | } | |
843 | } | |
ea01ea93 | 844 | |
7cf91a98 JK |
845 | clear_zone_contiguous(zone); |
846 | ||
ea01ea93 BP |
847 | /* |
848 | * We can only remove entire sections | |
849 | */ | |
850 | BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK); | |
851 | BUG_ON(nr_pages % PAGES_PER_SECTION); | |
852 | ||
ea01ea93 BP |
853 | sections_to_remove = nr_pages / PAGES_PER_SECTION; |
854 | for (i = 0; i < sections_to_remove; i++) { | |
855 | unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; | |
4b94ffdc DW |
856 | |
857 | ret = __remove_section(zone, __pfn_to_section(pfn), map_offset); | |
858 | map_offset = 0; | |
ea01ea93 BP |
859 | if (ret) |
860 | break; | |
861 | } | |
7cf91a98 JK |
862 | |
863 | set_zone_contiguous(zone); | |
864 | ||
ea01ea93 BP |
865 | return ret; |
866 | } | |
4edd7cef | 867 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
ea01ea93 | 868 | |
9d0ad8ca DK |
869 | int set_online_page_callback(online_page_callback_t callback) |
870 | { | |
871 | int rc = -EINVAL; | |
872 | ||
bfc8c901 VD |
873 | get_online_mems(); |
874 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
875 | |
876 | if (online_page_callback == generic_online_page) { | |
877 | online_page_callback = callback; | |
878 | rc = 0; | |
879 | } | |
880 | ||
bfc8c901 VD |
881 | mutex_unlock(&online_page_callback_lock); |
882 | put_online_mems(); | |
9d0ad8ca DK |
883 | |
884 | return rc; | |
885 | } | |
886 | EXPORT_SYMBOL_GPL(set_online_page_callback); | |
887 | ||
888 | int restore_online_page_callback(online_page_callback_t callback) | |
889 | { | |
890 | int rc = -EINVAL; | |
891 | ||
bfc8c901 VD |
892 | get_online_mems(); |
893 | mutex_lock(&online_page_callback_lock); | |
9d0ad8ca DK |
894 | |
895 | if (online_page_callback == callback) { | |
896 | online_page_callback = generic_online_page; | |
897 | rc = 0; | |
898 | } | |
899 | ||
bfc8c901 VD |
900 | mutex_unlock(&online_page_callback_lock); |
901 | put_online_mems(); | |
9d0ad8ca DK |
902 | |
903 | return rc; | |
904 | } | |
905 | EXPORT_SYMBOL_GPL(restore_online_page_callback); | |
906 | ||
907 | void __online_page_set_limits(struct page *page) | |
180c06ef | 908 | { |
9d0ad8ca DK |
909 | } |
910 | EXPORT_SYMBOL_GPL(__online_page_set_limits); | |
911 | ||
912 | void __online_page_increment_counters(struct page *page) | |
913 | { | |
3dcc0571 | 914 | adjust_managed_page_count(page, 1); |
9d0ad8ca DK |
915 | } |
916 | EXPORT_SYMBOL_GPL(__online_page_increment_counters); | |
180c06ef | 917 | |
9d0ad8ca DK |
918 | void __online_page_free(struct page *page) |
919 | { | |
3dcc0571 | 920 | __free_reserved_page(page); |
180c06ef | 921 | } |
9d0ad8ca DK |
922 | EXPORT_SYMBOL_GPL(__online_page_free); |
923 | ||
924 | static void generic_online_page(struct page *page) | |
925 | { | |
926 | __online_page_set_limits(page); | |
927 | __online_page_increment_counters(page); | |
928 | __online_page_free(page); | |
929 | } | |
180c06ef | 930 | |
75884fb1 KH |
931 | static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, |
932 | void *arg) | |
3947be19 DH |
933 | { |
934 | unsigned long i; | |
75884fb1 KH |
935 | unsigned long onlined_pages = *(unsigned long *)arg; |
936 | struct page *page; | |
937 | if (PageReserved(pfn_to_page(start_pfn))) | |
938 | for (i = 0; i < nr_pages; i++) { | |
939 | page = pfn_to_page(start_pfn + i); | |
9d0ad8ca | 940 | (*online_page_callback)(page); |
75884fb1 KH |
941 | onlined_pages++; |
942 | } | |
943 | *(unsigned long *)arg = onlined_pages; | |
944 | return 0; | |
945 | } | |
946 | ||
09285af7 | 947 | #ifdef CONFIG_MOVABLE_NODE |
79a4dcef TC |
948 | /* |
949 | * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have | |
950 | * normal memory. | |
951 | */ | |
09285af7 LJ |
952 | static bool can_online_high_movable(struct zone *zone) |
953 | { | |
954 | return true; | |
955 | } | |
79a4dcef | 956 | #else /* CONFIG_MOVABLE_NODE */ |
74d42d8f LJ |
957 | /* ensure every online node has NORMAL memory */ |
958 | static bool can_online_high_movable(struct zone *zone) | |
959 | { | |
960 | return node_state(zone_to_nid(zone), N_NORMAL_MEMORY); | |
961 | } | |
79a4dcef | 962 | #endif /* CONFIG_MOVABLE_NODE */ |
74d42d8f | 963 | |
d9713679 LJ |
964 | /* check which state of node_states will be changed when online memory */ |
965 | static void node_states_check_changes_online(unsigned long nr_pages, | |
966 | struct zone *zone, struct memory_notify *arg) | |
967 | { | |
968 | int nid = zone_to_nid(zone); | |
969 | enum zone_type zone_last = ZONE_NORMAL; | |
970 | ||
971 | /* | |
6715ddf9 LJ |
972 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
973 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
974 | * set zone_last to ZONE_NORMAL. | |
d9713679 | 975 | * |
6715ddf9 LJ |
976 | * If we don't have HIGHMEM nor movable node, |
977 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
978 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
d9713679 | 979 | */ |
6715ddf9 | 980 | if (N_MEMORY == N_NORMAL_MEMORY) |
d9713679 LJ |
981 | zone_last = ZONE_MOVABLE; |
982 | ||
983 | /* | |
984 | * if the memory to be online is in a zone of 0...zone_last, and | |
985 | * the zones of 0...zone_last don't have memory before online, we will | |
986 | * need to set the node to node_states[N_NORMAL_MEMORY] after | |
987 | * the memory is online. | |
988 | */ | |
989 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY)) | |
990 | arg->status_change_nid_normal = nid; | |
991 | else | |
992 | arg->status_change_nid_normal = -1; | |
993 | ||
6715ddf9 LJ |
994 | #ifdef CONFIG_HIGHMEM |
995 | /* | |
996 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
997 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
998 | * set zone_last to ZONE_HIGHMEM. | |
999 | * | |
1000 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
1001 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
1002 | * set zone_last to ZONE_MOVABLE. | |
1003 | */ | |
1004 | zone_last = ZONE_HIGHMEM; | |
1005 | if (N_MEMORY == N_HIGH_MEMORY) | |
1006 | zone_last = ZONE_MOVABLE; | |
1007 | ||
1008 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY)) | |
1009 | arg->status_change_nid_high = nid; | |
1010 | else | |
1011 | arg->status_change_nid_high = -1; | |
1012 | #else | |
1013 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
1014 | #endif | |
1015 | ||
d9713679 LJ |
1016 | /* |
1017 | * if the node don't have memory befor online, we will need to | |
6715ddf9 | 1018 | * set the node to node_states[N_MEMORY] after the memory |
d9713679 LJ |
1019 | * is online. |
1020 | */ | |
6715ddf9 | 1021 | if (!node_state(nid, N_MEMORY)) |
d9713679 LJ |
1022 | arg->status_change_nid = nid; |
1023 | else | |
1024 | arg->status_change_nid = -1; | |
1025 | } | |
1026 | ||
1027 | static void node_states_set_node(int node, struct memory_notify *arg) | |
1028 | { | |
1029 | if (arg->status_change_nid_normal >= 0) | |
1030 | node_set_state(node, N_NORMAL_MEMORY); | |
1031 | ||
6715ddf9 LJ |
1032 | if (arg->status_change_nid_high >= 0) |
1033 | node_set_state(node, N_HIGH_MEMORY); | |
1034 | ||
1035 | node_set_state(node, N_MEMORY); | |
d9713679 LJ |
1036 | } |
1037 | ||
8a1f780e YI |
1038 | bool zone_can_shift(unsigned long pfn, unsigned long nr_pages, |
1039 | enum zone_type target, int *zone_shift) | |
df429ac0 RA |
1040 | { |
1041 | struct zone *zone = page_zone(pfn_to_page(pfn)); | |
1042 | enum zone_type idx = zone_idx(zone); | |
1043 | int i; | |
1044 | ||
8a1f780e YI |
1045 | *zone_shift = 0; |
1046 | ||
df429ac0 RA |
1047 | if (idx < target) { |
1048 | /* pages must be at end of current zone */ | |
1049 | if (pfn + nr_pages != zone_end_pfn(zone)) | |
8a1f780e | 1050 | return false; |
df429ac0 RA |
1051 | |
1052 | /* no zones in use between current zone and target */ | |
1053 | for (i = idx + 1; i < target; i++) | |
1054 | if (zone_is_initialized(zone - idx + i)) | |
8a1f780e | 1055 | return false; |
df429ac0 RA |
1056 | } |
1057 | ||
1058 | if (target < idx) { | |
1059 | /* pages must be at beginning of current zone */ | |
1060 | if (pfn != zone->zone_start_pfn) | |
8a1f780e | 1061 | return false; |
df429ac0 RA |
1062 | |
1063 | /* no zones in use between current zone and target */ | |
1064 | for (i = target + 1; i < idx; i++) | |
1065 | if (zone_is_initialized(zone - idx + i)) | |
8a1f780e | 1066 | return false; |
df429ac0 RA |
1067 | } |
1068 | ||
8a1f780e YI |
1069 | *zone_shift = target - idx; |
1070 | return true; | |
df429ac0 | 1071 | } |
75884fb1 | 1072 | |
30467e0b | 1073 | /* Must be protected by mem_hotplug_begin() */ |
511c2aba | 1074 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type) |
75884fb1 | 1075 | { |
aa47228a | 1076 | unsigned long flags; |
3947be19 DH |
1077 | unsigned long onlined_pages = 0; |
1078 | struct zone *zone; | |
6811378e | 1079 | int need_zonelists_rebuild = 0; |
7b78d335 YG |
1080 | int nid; |
1081 | int ret; | |
1082 | struct memory_notify arg; | |
e51e6c8f | 1083 | int zone_shift = 0; |
7b78d335 | 1084 | |
d9713679 LJ |
1085 | /* |
1086 | * This doesn't need a lock to do pfn_to_page(). | |
1087 | * The section can't be removed here because of the | |
1088 | * memory_block->state_mutex. | |
1089 | */ | |
1090 | zone = page_zone(pfn_to_page(pfn)); | |
1091 | ||
4f7c6b49 TC |
1092 | if ((zone_idx(zone) > ZONE_NORMAL || |
1093 | online_type == MMOP_ONLINE_MOVABLE) && | |
bfc8c901 | 1094 | !can_online_high_movable(zone)) |
30467e0b | 1095 | return -EINVAL; |
74d42d8f | 1096 | |
8a1f780e YI |
1097 | if (online_type == MMOP_ONLINE_KERNEL) { |
1098 | if (!zone_can_shift(pfn, nr_pages, ZONE_NORMAL, &zone_shift)) | |
1099 | return -EINVAL; | |
1100 | } else if (online_type == MMOP_ONLINE_MOVABLE) { | |
1101 | if (!zone_can_shift(pfn, nr_pages, ZONE_MOVABLE, &zone_shift)) | |
1102 | return -EINVAL; | |
1103 | } | |
511c2aba | 1104 | |
e51e6c8f RA |
1105 | zone = move_pfn_range(zone_shift, pfn, pfn + nr_pages); |
1106 | if (!zone) | |
1107 | return -EINVAL; | |
511c2aba | 1108 | |
7b78d335 YG |
1109 | arg.start_pfn = pfn; |
1110 | arg.nr_pages = nr_pages; | |
d9713679 | 1111 | node_states_check_changes_online(nr_pages, zone, &arg); |
7b78d335 | 1112 | |
e888ca35 | 1113 | nid = zone_to_nid(zone); |
3947be19 | 1114 | |
7b78d335 YG |
1115 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
1116 | ret = notifier_to_errno(ret); | |
e33e33b4 CY |
1117 | if (ret) |
1118 | goto failed_addition; | |
1119 | ||
6811378e YG |
1120 | /* |
1121 | * If this zone is not populated, then it is not in zonelist. | |
1122 | * This means the page allocator ignores this zone. | |
1123 | * So, zonelist must be updated after online. | |
1124 | */ | |
4eaf3f64 | 1125 | mutex_lock(&zonelists_mutex); |
6dcd73d7 | 1126 | if (!populated_zone(zone)) { |
6811378e | 1127 | need_zonelists_rebuild = 1; |
6dcd73d7 WC |
1128 | build_all_zonelists(NULL, zone); |
1129 | } | |
6811378e | 1130 | |
908eedc6 | 1131 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, |
75884fb1 | 1132 | online_pages_range); |
fd8a4221 | 1133 | if (ret) { |
6dcd73d7 WC |
1134 | if (need_zonelists_rebuild) |
1135 | zone_pcp_reset(zone); | |
4eaf3f64 | 1136 | mutex_unlock(&zonelists_mutex); |
e33e33b4 | 1137 | goto failed_addition; |
fd8a4221 GL |
1138 | } |
1139 | ||
3947be19 | 1140 | zone->present_pages += onlined_pages; |
aa47228a CS |
1141 | |
1142 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
f2937be5 | 1143 | zone->zone_pgdat->node_present_pages += onlined_pages; |
aa47228a CS |
1144 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
1145 | ||
08dff7b7 | 1146 | if (onlined_pages) { |
e888ca35 | 1147 | node_states_set_node(nid, &arg); |
08dff7b7 | 1148 | if (need_zonelists_rebuild) |
6dcd73d7 | 1149 | build_all_zonelists(NULL, NULL); |
08dff7b7 JL |
1150 | else |
1151 | zone_pcp_update(zone); | |
1152 | } | |
3947be19 | 1153 | |
4eaf3f64 | 1154 | mutex_unlock(&zonelists_mutex); |
1b79acc9 KM |
1155 | |
1156 | init_per_zone_wmark_min(); | |
1157 | ||
698b1b30 | 1158 | if (onlined_pages) { |
e888ca35 | 1159 | kswapd_run(nid); |
698b1b30 VB |
1160 | kcompactd_run(nid); |
1161 | } | |
61b13993 | 1162 | |
1f522509 | 1163 | vm_total_pages = nr_free_pagecache_pages(); |
2f7f24ec | 1164 | |
2d1d43f6 | 1165 | writeback_set_ratelimit(); |
7b78d335 YG |
1166 | |
1167 | if (onlined_pages) | |
1168 | memory_notify(MEM_ONLINE, &arg); | |
30467e0b | 1169 | return 0; |
e33e33b4 CY |
1170 | |
1171 | failed_addition: | |
1172 | pr_debug("online_pages [mem %#010llx-%#010llx] failed\n", | |
1173 | (unsigned long long) pfn << PAGE_SHIFT, | |
1174 | (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1); | |
1175 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
1176 | return ret; | |
3947be19 | 1177 | } |
53947027 | 1178 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
bc02af93 | 1179 | |
0bd85420 TC |
1180 | static void reset_node_present_pages(pg_data_t *pgdat) |
1181 | { | |
1182 | struct zone *z; | |
1183 | ||
1184 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) | |
1185 | z->present_pages = 0; | |
1186 | ||
1187 | pgdat->node_present_pages = 0; | |
1188 | } | |
1189 | ||
e1319331 HS |
1190 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
1191 | static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) | |
9af3c2de YG |
1192 | { |
1193 | struct pglist_data *pgdat; | |
1194 | unsigned long zones_size[MAX_NR_ZONES] = {0}; | |
1195 | unsigned long zholes_size[MAX_NR_ZONES] = {0}; | |
c8e861a5 | 1196 | unsigned long start_pfn = PFN_DOWN(start); |
9af3c2de | 1197 | |
a1e565aa TC |
1198 | pgdat = NODE_DATA(nid); |
1199 | if (!pgdat) { | |
1200 | pgdat = arch_alloc_nodedata(nid); | |
1201 | if (!pgdat) | |
1202 | return NULL; | |
9af3c2de | 1203 | |
a1e565aa | 1204 | arch_refresh_nodedata(nid, pgdat); |
b0dc3a34 | 1205 | } else { |
38087d9b | 1206 | /* Reset the nr_zones, order and classzone_idx before reuse */ |
b0dc3a34 | 1207 | pgdat->nr_zones = 0; |
38087d9b MG |
1208 | pgdat->kswapd_order = 0; |
1209 | pgdat->kswapd_classzone_idx = 0; | |
a1e565aa | 1210 | } |
9af3c2de YG |
1211 | |
1212 | /* we can use NODE_DATA(nid) from here */ | |
1213 | ||
1214 | /* init node's zones as empty zones, we don't have any present pages.*/ | |
9109fb7b | 1215 | free_area_init_node(nid, zones_size, start_pfn, zholes_size); |
5830169f | 1216 | pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat); |
9af3c2de | 1217 | |
959ecc48 KH |
1218 | /* |
1219 | * The node we allocated has no zone fallback lists. For avoiding | |
1220 | * to access not-initialized zonelist, build here. | |
1221 | */ | |
f957db4f | 1222 | mutex_lock(&zonelists_mutex); |
9adb62a5 | 1223 | build_all_zonelists(pgdat, NULL); |
f957db4f | 1224 | mutex_unlock(&zonelists_mutex); |
959ecc48 | 1225 | |
f784a3f1 TC |
1226 | /* |
1227 | * zone->managed_pages is set to an approximate value in | |
1228 | * free_area_init_core(), which will cause | |
1229 | * /sys/device/system/node/nodeX/meminfo has wrong data. | |
1230 | * So reset it to 0 before any memory is onlined. | |
1231 | */ | |
1232 | reset_node_managed_pages(pgdat); | |
1233 | ||
0bd85420 TC |
1234 | /* |
1235 | * When memory is hot-added, all the memory is in offline state. So | |
1236 | * clear all zones' present_pages because they will be updated in | |
1237 | * online_pages() and offline_pages(). | |
1238 | */ | |
1239 | reset_node_present_pages(pgdat); | |
1240 | ||
9af3c2de YG |
1241 | return pgdat; |
1242 | } | |
1243 | ||
1244 | static void rollback_node_hotadd(int nid, pg_data_t *pgdat) | |
1245 | { | |
1246 | arch_refresh_nodedata(nid, NULL); | |
5830169f | 1247 | free_percpu(pgdat->per_cpu_nodestats); |
9af3c2de YG |
1248 | arch_free_nodedata(pgdat); |
1249 | return; | |
1250 | } | |
1251 | ||
0a547039 | 1252 | |
01b0f197 TK |
1253 | /** |
1254 | * try_online_node - online a node if offlined | |
1255 | * | |
cf23422b | 1256 | * called by cpu_up() to online a node without onlined memory. |
1257 | */ | |
01b0f197 | 1258 | int try_online_node(int nid) |
cf23422b | 1259 | { |
1260 | pg_data_t *pgdat; | |
1261 | int ret; | |
1262 | ||
01b0f197 TK |
1263 | if (node_online(nid)) |
1264 | return 0; | |
1265 | ||
bfc8c901 | 1266 | mem_hotplug_begin(); |
cf23422b | 1267 | pgdat = hotadd_new_pgdat(nid, 0); |
7553e8f2 | 1268 | if (!pgdat) { |
01b0f197 | 1269 | pr_err("Cannot online node %d due to NULL pgdat\n", nid); |
cf23422b | 1270 | ret = -ENOMEM; |
1271 | goto out; | |
1272 | } | |
1273 | node_set_online(nid); | |
1274 | ret = register_one_node(nid); | |
1275 | BUG_ON(ret); | |
1276 | ||
01b0f197 TK |
1277 | if (pgdat->node_zonelists->_zonerefs->zone == NULL) { |
1278 | mutex_lock(&zonelists_mutex); | |
1279 | build_all_zonelists(NULL, NULL); | |
1280 | mutex_unlock(&zonelists_mutex); | |
1281 | } | |
1282 | ||
cf23422b | 1283 | out: |
bfc8c901 | 1284 | mem_hotplug_done(); |
cf23422b | 1285 | return ret; |
1286 | } | |
1287 | ||
27356f54 TK |
1288 | static int check_hotplug_memory_range(u64 start, u64 size) |
1289 | { | |
c8e861a5 | 1290 | u64 start_pfn = PFN_DOWN(start); |
27356f54 TK |
1291 | u64 nr_pages = size >> PAGE_SHIFT; |
1292 | ||
1293 | /* Memory range must be aligned with section */ | |
1294 | if ((start_pfn & ~PAGE_SECTION_MASK) || | |
1295 | (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) { | |
1296 | pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n", | |
1297 | (unsigned long long)start, | |
1298 | (unsigned long long)size); | |
1299 | return -EINVAL; | |
1300 | } | |
1301 | ||
1302 | return 0; | |
1303 | } | |
1304 | ||
63264400 WN |
1305 | /* |
1306 | * If movable zone has already been setup, newly added memory should be check. | |
1307 | * If its address is higher than movable zone, it should be added as movable. | |
1308 | * Without this check, movable zone may overlap with other zone. | |
1309 | */ | |
1310 | static int should_add_memory_movable(int nid, u64 start, u64 size) | |
1311 | { | |
1312 | unsigned long start_pfn = start >> PAGE_SHIFT; | |
1313 | pg_data_t *pgdat = NODE_DATA(nid); | |
1314 | struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE; | |
1315 | ||
1316 | if (zone_is_empty(movable_zone)) | |
1317 | return 0; | |
1318 | ||
1319 | if (movable_zone->zone_start_pfn <= start_pfn) | |
1320 | return 1; | |
1321 | ||
1322 | return 0; | |
1323 | } | |
1324 | ||
033fbae9 DW |
1325 | int zone_for_memory(int nid, u64 start, u64 size, int zone_default, |
1326 | bool for_device) | |
63264400 | 1327 | { |
033fbae9 DW |
1328 | #ifdef CONFIG_ZONE_DEVICE |
1329 | if (for_device) | |
1330 | return ZONE_DEVICE; | |
1331 | #endif | |
63264400 WN |
1332 | if (should_add_memory_movable(nid, start, size)) |
1333 | return ZONE_MOVABLE; | |
1334 | ||
1335 | return zone_default; | |
1336 | } | |
1337 | ||
31bc3858 VK |
1338 | static int online_memory_block(struct memory_block *mem, void *arg) |
1339 | { | |
1340 | return memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE); | |
1341 | } | |
1342 | ||
31168481 | 1343 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
31bc3858 | 1344 | int __ref add_memory_resource(int nid, struct resource *res, bool online) |
bc02af93 | 1345 | { |
62cedb9f | 1346 | u64 start, size; |
9af3c2de | 1347 | pg_data_t *pgdat = NULL; |
a1e565aa TC |
1348 | bool new_pgdat; |
1349 | bool new_node; | |
bc02af93 YG |
1350 | int ret; |
1351 | ||
62cedb9f DV |
1352 | start = res->start; |
1353 | size = resource_size(res); | |
1354 | ||
27356f54 TK |
1355 | ret = check_hotplug_memory_range(start, size); |
1356 | if (ret) | |
1357 | return ret; | |
1358 | ||
a1e565aa TC |
1359 | { /* Stupid hack to suppress address-never-null warning */ |
1360 | void *p = NODE_DATA(nid); | |
1361 | new_pgdat = !p; | |
1362 | } | |
ac13c462 | 1363 | |
bfc8c901 | 1364 | mem_hotplug_begin(); |
ac13c462 | 1365 | |
7f36e3e5 TC |
1366 | /* |
1367 | * Add new range to memblock so that when hotadd_new_pgdat() is called | |
1368 | * to allocate new pgdat, get_pfn_range_for_nid() will be able to find | |
1369 | * this new range and calculate total pages correctly. The range will | |
1370 | * be removed at hot-remove time. | |
1371 | */ | |
1372 | memblock_add_node(start, size, nid); | |
1373 | ||
a1e565aa TC |
1374 | new_node = !node_online(nid); |
1375 | if (new_node) { | |
9af3c2de | 1376 | pgdat = hotadd_new_pgdat(nid, start); |
6ad696d2 | 1377 | ret = -ENOMEM; |
9af3c2de | 1378 | if (!pgdat) |
41b9e2d7 | 1379 | goto error; |
9af3c2de YG |
1380 | } |
1381 | ||
bc02af93 | 1382 | /* call arch's memory hotadd */ |
033fbae9 | 1383 | ret = arch_add_memory(nid, start, size, false); |
bc02af93 | 1384 | |
9af3c2de YG |
1385 | if (ret < 0) |
1386 | goto error; | |
1387 | ||
0fc44159 | 1388 | /* we online node here. we can't roll back from here. */ |
9af3c2de YG |
1389 | node_set_online(nid); |
1390 | ||
a1e565aa | 1391 | if (new_node) { |
0fc44159 YG |
1392 | ret = register_one_node(nid); |
1393 | /* | |
1394 | * If sysfs file of new node can't create, cpu on the node | |
1395 | * can't be hot-added. There is no rollback way now. | |
1396 | * So, check by BUG_ON() to catch it reluctantly.. | |
1397 | */ | |
1398 | BUG_ON(ret); | |
1399 | } | |
1400 | ||
d96ae530 AM |
1401 | /* create new memmap entry */ |
1402 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
1403 | ||
31bc3858 VK |
1404 | /* online pages if requested */ |
1405 | if (online) | |
1406 | walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), | |
1407 | NULL, online_memory_block); | |
1408 | ||
6ad696d2 AK |
1409 | goto out; |
1410 | ||
9af3c2de YG |
1411 | error: |
1412 | /* rollback pgdat allocation and others */ | |
1413 | if (new_pgdat) | |
1414 | rollback_node_hotadd(nid, pgdat); | |
7f36e3e5 | 1415 | memblock_remove(start, size); |
9af3c2de | 1416 | |
6ad696d2 | 1417 | out: |
bfc8c901 | 1418 | mem_hotplug_done(); |
bc02af93 YG |
1419 | return ret; |
1420 | } | |
62cedb9f DV |
1421 | EXPORT_SYMBOL_GPL(add_memory_resource); |
1422 | ||
1423 | int __ref add_memory(int nid, u64 start, u64 size) | |
1424 | { | |
1425 | struct resource *res; | |
1426 | int ret; | |
1427 | ||
1428 | res = register_memory_resource(start, size); | |
6f754ba4 VK |
1429 | if (IS_ERR(res)) |
1430 | return PTR_ERR(res); | |
62cedb9f | 1431 | |
31bc3858 | 1432 | ret = add_memory_resource(nid, res, memhp_auto_online); |
62cedb9f DV |
1433 | if (ret < 0) |
1434 | release_memory_resource(res); | |
1435 | return ret; | |
1436 | } | |
bc02af93 | 1437 | EXPORT_SYMBOL_GPL(add_memory); |
0c0e6195 KH |
1438 | |
1439 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5c755e9f BP |
1440 | /* |
1441 | * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy | |
1442 | * set and the size of the free page is given by page_order(). Using this, | |
1443 | * the function determines if the pageblock contains only free pages. | |
1444 | * Due to buddy contraints, a free page at least the size of a pageblock will | |
1445 | * be located at the start of the pageblock | |
1446 | */ | |
1447 | static inline int pageblock_free(struct page *page) | |
1448 | { | |
1449 | return PageBuddy(page) && page_order(page) >= pageblock_order; | |
1450 | } | |
1451 | ||
1452 | /* Return the start of the next active pageblock after a given page */ | |
1453 | static struct page *next_active_pageblock(struct page *page) | |
1454 | { | |
5c755e9f BP |
1455 | /* Ensure the starting page is pageblock-aligned */ |
1456 | BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1)); | |
1457 | ||
5c755e9f | 1458 | /* If the entire pageblock is free, move to the end of free page */ |
0dcc48c1 KH |
1459 | if (pageblock_free(page)) { |
1460 | int order; | |
1461 | /* be careful. we don't have locks, page_order can be changed.*/ | |
1462 | order = page_order(page); | |
1463 | if ((order < MAX_ORDER) && (order >= pageblock_order)) | |
1464 | return page + (1 << order); | |
1465 | } | |
5c755e9f | 1466 | |
0dcc48c1 | 1467 | return page + pageblock_nr_pages; |
5c755e9f BP |
1468 | } |
1469 | ||
1470 | /* Checks if this range of memory is likely to be hot-removable. */ | |
c98940f6 | 1471 | bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) |
5c755e9f | 1472 | { |
5c755e9f BP |
1473 | struct page *page = pfn_to_page(start_pfn); |
1474 | struct page *end_page = page + nr_pages; | |
1475 | ||
1476 | /* Check the starting page of each pageblock within the range */ | |
1477 | for (; page < end_page; page = next_active_pageblock(page)) { | |
49ac8255 | 1478 | if (!is_pageblock_removable_nolock(page)) |
c98940f6 | 1479 | return false; |
49ac8255 | 1480 | cond_resched(); |
5c755e9f BP |
1481 | } |
1482 | ||
1483 | /* All pageblocks in the memory block are likely to be hot-removable */ | |
c98940f6 | 1484 | return true; |
5c755e9f BP |
1485 | } |
1486 | ||
0c0e6195 | 1487 | /* |
deb88a2a | 1488 | * Confirm all pages in a range [start, end) belong to the same zone. |
a96dfddb | 1489 | * When true, return its valid [start, end). |
0c0e6195 | 1490 | */ |
a96dfddb TK |
1491 | int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn, |
1492 | unsigned long *valid_start, unsigned long *valid_end) | |
0c0e6195 | 1493 | { |
5f0f2887 | 1494 | unsigned long pfn, sec_end_pfn; |
a96dfddb | 1495 | unsigned long start, end; |
0c0e6195 KH |
1496 | struct zone *zone = NULL; |
1497 | struct page *page; | |
1498 | int i; | |
deb88a2a | 1499 | for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1); |
0c0e6195 | 1500 | pfn < end_pfn; |
deb88a2a | 1501 | pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) { |
5f0f2887 AB |
1502 | /* Make sure the memory section is present first */ |
1503 | if (!present_section_nr(pfn_to_section_nr(pfn))) | |
0c0e6195 | 1504 | continue; |
5f0f2887 AB |
1505 | for (; pfn < sec_end_pfn && pfn < end_pfn; |
1506 | pfn += MAX_ORDER_NR_PAGES) { | |
1507 | i = 0; | |
1508 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ | |
1509 | while ((i < MAX_ORDER_NR_PAGES) && | |
1510 | !pfn_valid_within(pfn + i)) | |
1511 | i++; | |
d6d8c8a4 | 1512 | if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn) |
5f0f2887 AB |
1513 | continue; |
1514 | page = pfn_to_page(pfn + i); | |
1515 | if (zone && page_zone(page) != zone) | |
1516 | return 0; | |
a96dfddb TK |
1517 | if (!zone) |
1518 | start = pfn + i; | |
5f0f2887 | 1519 | zone = page_zone(page); |
a96dfddb | 1520 | end = pfn + MAX_ORDER_NR_PAGES; |
5f0f2887 | 1521 | } |
0c0e6195 | 1522 | } |
deb88a2a | 1523 | |
a96dfddb TK |
1524 | if (zone) { |
1525 | *valid_start = start; | |
d6d8c8a4 | 1526 | *valid_end = min(end, end_pfn); |
deb88a2a | 1527 | return 1; |
a96dfddb | 1528 | } else { |
deb88a2a | 1529 | return 0; |
a96dfddb | 1530 | } |
0c0e6195 KH |
1531 | } |
1532 | ||
1533 | /* | |
0efadf48 YX |
1534 | * Scan pfn range [start,end) to find movable/migratable pages (LRU pages, |
1535 | * non-lru movable pages and hugepages). We scan pfn because it's much | |
1536 | * easier than scanning over linked list. This function returns the pfn | |
1537 | * of the first found movable page if it's found, otherwise 0. | |
0c0e6195 | 1538 | */ |
c8721bbb | 1539 | static unsigned long scan_movable_pages(unsigned long start, unsigned long end) |
0c0e6195 KH |
1540 | { |
1541 | unsigned long pfn; | |
1542 | struct page *page; | |
1543 | for (pfn = start; pfn < end; pfn++) { | |
1544 | if (pfn_valid(pfn)) { | |
1545 | page = pfn_to_page(pfn); | |
1546 | if (PageLRU(page)) | |
1547 | return pfn; | |
0efadf48 YX |
1548 | if (__PageMovable(page)) |
1549 | return pfn; | |
c8721bbb | 1550 | if (PageHuge(page)) { |
7e1f049e | 1551 | if (page_huge_active(page)) |
c8721bbb NH |
1552 | return pfn; |
1553 | else | |
1554 | pfn = round_up(pfn + 1, | |
1555 | 1 << compound_order(page)) - 1; | |
1556 | } | |
0c0e6195 KH |
1557 | } |
1558 | } | |
1559 | return 0; | |
1560 | } | |
1561 | ||
394e31d2 XQ |
1562 | static struct page *new_node_page(struct page *page, unsigned long private, |
1563 | int **result) | |
1564 | { | |
1565 | gfp_t gfp_mask = GFP_USER | __GFP_MOVABLE; | |
1566 | int nid = page_to_nid(page); | |
231e97e2 LZ |
1567 | nodemask_t nmask = node_states[N_MEMORY]; |
1568 | struct page *new_page = NULL; | |
394e31d2 XQ |
1569 | |
1570 | /* | |
1571 | * TODO: allocate a destination hugepage from a nearest neighbor node, | |
1572 | * accordance with memory policy of the user process if possible. For | |
1573 | * now as a simple work-around, we use the next node for destination. | |
1574 | */ | |
1575 | if (PageHuge(page)) | |
1576 | return alloc_huge_page_node(page_hstate(compound_head(page)), | |
1577 | next_node_in(nid, nmask)); | |
1578 | ||
231e97e2 | 1579 | node_clear(nid, nmask); |
9bb627be | 1580 | |
394e31d2 XQ |
1581 | if (PageHighMem(page) |
1582 | || (zone_idx(page_zone(page)) == ZONE_MOVABLE)) | |
1583 | gfp_mask |= __GFP_HIGHMEM; | |
1584 | ||
231e97e2 LZ |
1585 | if (!nodes_empty(nmask)) |
1586 | new_page = __alloc_pages_nodemask(gfp_mask, 0, | |
394e31d2 XQ |
1587 | node_zonelist(nid, gfp_mask), &nmask); |
1588 | if (!new_page) | |
1589 | new_page = __alloc_pages(gfp_mask, 0, | |
1590 | node_zonelist(nid, gfp_mask)); | |
1591 | ||
1592 | return new_page; | |
1593 | } | |
1594 | ||
0c0e6195 KH |
1595 | #define NR_OFFLINE_AT_ONCE_PAGES (256) |
1596 | static int | |
1597 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
1598 | { | |
1599 | unsigned long pfn; | |
1600 | struct page *page; | |
1601 | int move_pages = NR_OFFLINE_AT_ONCE_PAGES; | |
1602 | int not_managed = 0; | |
1603 | int ret = 0; | |
1604 | LIST_HEAD(source); | |
1605 | ||
1606 | for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) { | |
1607 | if (!pfn_valid(pfn)) | |
1608 | continue; | |
1609 | page = pfn_to_page(pfn); | |
c8721bbb NH |
1610 | |
1611 | if (PageHuge(page)) { | |
1612 | struct page *head = compound_head(page); | |
1613 | pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1; | |
1614 | if (compound_order(head) > PFN_SECTION_SHIFT) { | |
1615 | ret = -EBUSY; | |
1616 | break; | |
1617 | } | |
1618 | if (isolate_huge_page(page, &source)) | |
1619 | move_pages -= 1 << compound_order(head); | |
1620 | continue; | |
1621 | } | |
1622 | ||
700c2a46 | 1623 | if (!get_page_unless_zero(page)) |
0c0e6195 KH |
1624 | continue; |
1625 | /* | |
0efadf48 YX |
1626 | * We can skip free pages. And we can deal with pages on |
1627 | * LRU and non-lru movable pages. | |
0c0e6195 | 1628 | */ |
0efadf48 YX |
1629 | if (PageLRU(page)) |
1630 | ret = isolate_lru_page(page); | |
1631 | else | |
1632 | ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE); | |
0c0e6195 | 1633 | if (!ret) { /* Success */ |
700c2a46 | 1634 | put_page(page); |
62695a84 | 1635 | list_add_tail(&page->lru, &source); |
0c0e6195 | 1636 | move_pages--; |
0efadf48 YX |
1637 | if (!__PageMovable(page)) |
1638 | inc_node_page_state(page, NR_ISOLATED_ANON + | |
1639 | page_is_file_cache(page)); | |
6d9c285a | 1640 | |
0c0e6195 | 1641 | } else { |
0c0e6195 | 1642 | #ifdef CONFIG_DEBUG_VM |
0efadf48 YX |
1643 | pr_alert("failed to isolate pfn %lx\n", pfn); |
1644 | dump_page(page, "isolation failed"); | |
0c0e6195 | 1645 | #endif |
700c2a46 | 1646 | put_page(page); |
25985edc | 1647 | /* Because we don't have big zone->lock. we should |
809c4449 BL |
1648 | check this again here. */ |
1649 | if (page_count(page)) { | |
1650 | not_managed++; | |
f3ab2636 | 1651 | ret = -EBUSY; |
809c4449 BL |
1652 | break; |
1653 | } | |
0c0e6195 KH |
1654 | } |
1655 | } | |
f3ab2636 BL |
1656 | if (!list_empty(&source)) { |
1657 | if (not_managed) { | |
c8721bbb | 1658 | putback_movable_pages(&source); |
f3ab2636 BL |
1659 | goto out; |
1660 | } | |
74c08f98 | 1661 | |
394e31d2 XQ |
1662 | /* Allocate a new page from the nearest neighbor node */ |
1663 | ret = migrate_pages(&source, new_node_page, NULL, 0, | |
9c620e2b | 1664 | MIGRATE_SYNC, MR_MEMORY_HOTPLUG); |
f3ab2636 | 1665 | if (ret) |
c8721bbb | 1666 | putback_movable_pages(&source); |
0c0e6195 | 1667 | } |
0c0e6195 KH |
1668 | out: |
1669 | return ret; | |
1670 | } | |
1671 | ||
1672 | /* | |
1673 | * remove from free_area[] and mark all as Reserved. | |
1674 | */ | |
1675 | static int | |
1676 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, | |
1677 | void *data) | |
1678 | { | |
1679 | __offline_isolated_pages(start, start + nr_pages); | |
1680 | return 0; | |
1681 | } | |
1682 | ||
1683 | static void | |
1684 | offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
1685 | { | |
908eedc6 | 1686 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL, |
0c0e6195 KH |
1687 | offline_isolated_pages_cb); |
1688 | } | |
1689 | ||
1690 | /* | |
1691 | * Check all pages in range, recoreded as memory resource, are isolated. | |
1692 | */ | |
1693 | static int | |
1694 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, | |
1695 | void *data) | |
1696 | { | |
1697 | int ret; | |
1698 | long offlined = *(long *)data; | |
b023f468 | 1699 | ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true); |
0c0e6195 KH |
1700 | offlined = nr_pages; |
1701 | if (!ret) | |
1702 | *(long *)data += offlined; | |
1703 | return ret; | |
1704 | } | |
1705 | ||
1706 | static long | |
1707 | check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn) | |
1708 | { | |
1709 | long offlined = 0; | |
1710 | int ret; | |
1711 | ||
908eedc6 | 1712 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined, |
0c0e6195 KH |
1713 | check_pages_isolated_cb); |
1714 | if (ret < 0) | |
1715 | offlined = (long)ret; | |
1716 | return offlined; | |
1717 | } | |
1718 | ||
09285af7 | 1719 | #ifdef CONFIG_MOVABLE_NODE |
79a4dcef TC |
1720 | /* |
1721 | * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have | |
1722 | * normal memory. | |
1723 | */ | |
09285af7 LJ |
1724 | static bool can_offline_normal(struct zone *zone, unsigned long nr_pages) |
1725 | { | |
1726 | return true; | |
1727 | } | |
79a4dcef | 1728 | #else /* CONFIG_MOVABLE_NODE */ |
74d42d8f LJ |
1729 | /* ensure the node has NORMAL memory if it is still online */ |
1730 | static bool can_offline_normal(struct zone *zone, unsigned long nr_pages) | |
1731 | { | |
1732 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1733 | unsigned long present_pages = 0; | |
1734 | enum zone_type zt; | |
1735 | ||
1736 | for (zt = 0; zt <= ZONE_NORMAL; zt++) | |
1737 | present_pages += pgdat->node_zones[zt].present_pages; | |
1738 | ||
1739 | if (present_pages > nr_pages) | |
1740 | return true; | |
1741 | ||
1742 | present_pages = 0; | |
1743 | for (; zt <= ZONE_MOVABLE; zt++) | |
1744 | present_pages += pgdat->node_zones[zt].present_pages; | |
1745 | ||
1746 | /* | |
1747 | * we can't offline the last normal memory until all | |
1748 | * higher memory is offlined. | |
1749 | */ | |
1750 | return present_pages == 0; | |
1751 | } | |
79a4dcef | 1752 | #endif /* CONFIG_MOVABLE_NODE */ |
74d42d8f | 1753 | |
c5320926 TC |
1754 | static int __init cmdline_parse_movable_node(char *p) |
1755 | { | |
1756 | #ifdef CONFIG_MOVABLE_NODE | |
55ac590c | 1757 | movable_node_enabled = true; |
c5320926 TC |
1758 | #else |
1759 | pr_warn("movable_node option not supported\n"); | |
1760 | #endif | |
1761 | return 0; | |
1762 | } | |
1763 | early_param("movable_node", cmdline_parse_movable_node); | |
1764 | ||
d9713679 LJ |
1765 | /* check which state of node_states will be changed when offline memory */ |
1766 | static void node_states_check_changes_offline(unsigned long nr_pages, | |
1767 | struct zone *zone, struct memory_notify *arg) | |
1768 | { | |
1769 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1770 | unsigned long present_pages = 0; | |
1771 | enum zone_type zt, zone_last = ZONE_NORMAL; | |
1772 | ||
1773 | /* | |
6715ddf9 LJ |
1774 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
1775 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
1776 | * set zone_last to ZONE_NORMAL. | |
d9713679 | 1777 | * |
6715ddf9 LJ |
1778 | * If we don't have HIGHMEM nor movable node, |
1779 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
1780 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
d9713679 | 1781 | */ |
6715ddf9 | 1782 | if (N_MEMORY == N_NORMAL_MEMORY) |
d9713679 LJ |
1783 | zone_last = ZONE_MOVABLE; |
1784 | ||
1785 | /* | |
1786 | * check whether node_states[N_NORMAL_MEMORY] will be changed. | |
1787 | * If the memory to be offline is in a zone of 0...zone_last, | |
1788 | * and it is the last present memory, 0...zone_last will | |
1789 | * become empty after offline , thus we can determind we will | |
1790 | * need to clear the node from node_states[N_NORMAL_MEMORY]. | |
1791 | */ | |
1792 | for (zt = 0; zt <= zone_last; zt++) | |
1793 | present_pages += pgdat->node_zones[zt].present_pages; | |
1794 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1795 | arg->status_change_nid_normal = zone_to_nid(zone); | |
1796 | else | |
1797 | arg->status_change_nid_normal = -1; | |
1798 | ||
6715ddf9 LJ |
1799 | #ifdef CONFIG_HIGHMEM |
1800 | /* | |
1801 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
1802 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
1803 | * set zone_last to ZONE_HIGHMEM. | |
1804 | * | |
1805 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
1806 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
1807 | * set zone_last to ZONE_MOVABLE. | |
1808 | */ | |
1809 | zone_last = ZONE_HIGHMEM; | |
1810 | if (N_MEMORY == N_HIGH_MEMORY) | |
1811 | zone_last = ZONE_MOVABLE; | |
1812 | ||
1813 | for (; zt <= zone_last; zt++) | |
1814 | present_pages += pgdat->node_zones[zt].present_pages; | |
1815 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1816 | arg->status_change_nid_high = zone_to_nid(zone); | |
1817 | else | |
1818 | arg->status_change_nid_high = -1; | |
1819 | #else | |
1820 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
1821 | #endif | |
1822 | ||
d9713679 LJ |
1823 | /* |
1824 | * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE | |
1825 | */ | |
1826 | zone_last = ZONE_MOVABLE; | |
1827 | ||
1828 | /* | |
1829 | * check whether node_states[N_HIGH_MEMORY] will be changed | |
1830 | * If we try to offline the last present @nr_pages from the node, | |
1831 | * we can determind we will need to clear the node from | |
1832 | * node_states[N_HIGH_MEMORY]. | |
1833 | */ | |
1834 | for (; zt <= zone_last; zt++) | |
1835 | present_pages += pgdat->node_zones[zt].present_pages; | |
1836 | if (nr_pages >= present_pages) | |
1837 | arg->status_change_nid = zone_to_nid(zone); | |
1838 | else | |
1839 | arg->status_change_nid = -1; | |
1840 | } | |
1841 | ||
1842 | static void node_states_clear_node(int node, struct memory_notify *arg) | |
1843 | { | |
1844 | if (arg->status_change_nid_normal >= 0) | |
1845 | node_clear_state(node, N_NORMAL_MEMORY); | |
1846 | ||
6715ddf9 LJ |
1847 | if ((N_MEMORY != N_NORMAL_MEMORY) && |
1848 | (arg->status_change_nid_high >= 0)) | |
d9713679 | 1849 | node_clear_state(node, N_HIGH_MEMORY); |
6715ddf9 LJ |
1850 | |
1851 | if ((N_MEMORY != N_HIGH_MEMORY) && | |
1852 | (arg->status_change_nid >= 0)) | |
1853 | node_clear_state(node, N_MEMORY); | |
d9713679 LJ |
1854 | } |
1855 | ||
a16cee10 | 1856 | static int __ref __offline_pages(unsigned long start_pfn, |
0c0e6195 KH |
1857 | unsigned long end_pfn, unsigned long timeout) |
1858 | { | |
1859 | unsigned long pfn, nr_pages, expire; | |
1860 | long offlined_pages; | |
7b78d335 | 1861 | int ret, drain, retry_max, node; |
d702909f | 1862 | unsigned long flags; |
a96dfddb | 1863 | unsigned long valid_start, valid_end; |
0c0e6195 | 1864 | struct zone *zone; |
7b78d335 | 1865 | struct memory_notify arg; |
0c0e6195 | 1866 | |
0c0e6195 KH |
1867 | /* at least, alignment against pageblock is necessary */ |
1868 | if (!IS_ALIGNED(start_pfn, pageblock_nr_pages)) | |
1869 | return -EINVAL; | |
1870 | if (!IS_ALIGNED(end_pfn, pageblock_nr_pages)) | |
1871 | return -EINVAL; | |
1872 | /* This makes hotplug much easier...and readable. | |
1873 | we assume this for now. .*/ | |
a96dfddb | 1874 | if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, &valid_end)) |
0c0e6195 | 1875 | return -EINVAL; |
7b78d335 | 1876 | |
a96dfddb | 1877 | zone = page_zone(pfn_to_page(valid_start)); |
7b78d335 YG |
1878 | node = zone_to_nid(zone); |
1879 | nr_pages = end_pfn - start_pfn; | |
1880 | ||
74d42d8f | 1881 | if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages)) |
30467e0b | 1882 | return -EINVAL; |
74d42d8f | 1883 | |
0c0e6195 | 1884 | /* set above range as isolated */ |
b023f468 WC |
1885 | ret = start_isolate_page_range(start_pfn, end_pfn, |
1886 | MIGRATE_MOVABLE, true); | |
0c0e6195 | 1887 | if (ret) |
30467e0b | 1888 | return ret; |
7b78d335 YG |
1889 | |
1890 | arg.start_pfn = start_pfn; | |
1891 | arg.nr_pages = nr_pages; | |
d9713679 | 1892 | node_states_check_changes_offline(nr_pages, zone, &arg); |
7b78d335 YG |
1893 | |
1894 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
1895 | ret = notifier_to_errno(ret); | |
1896 | if (ret) | |
1897 | goto failed_removal; | |
1898 | ||
0c0e6195 KH |
1899 | pfn = start_pfn; |
1900 | expire = jiffies + timeout; | |
1901 | drain = 0; | |
1902 | retry_max = 5; | |
1903 | repeat: | |
1904 | /* start memory hot removal */ | |
1905 | ret = -EAGAIN; | |
1906 | if (time_after(jiffies, expire)) | |
1907 | goto failed_removal; | |
1908 | ret = -EINTR; | |
1909 | if (signal_pending(current)) | |
1910 | goto failed_removal; | |
1911 | ret = 0; | |
1912 | if (drain) { | |
1913 | lru_add_drain_all(); | |
0c0e6195 | 1914 | cond_resched(); |
c0554329 | 1915 | drain_all_pages(zone); |
0c0e6195 KH |
1916 | } |
1917 | ||
c8721bbb NH |
1918 | pfn = scan_movable_pages(start_pfn, end_pfn); |
1919 | if (pfn) { /* We have movable pages */ | |
0c0e6195 KH |
1920 | ret = do_migrate_range(pfn, end_pfn); |
1921 | if (!ret) { | |
1922 | drain = 1; | |
1923 | goto repeat; | |
1924 | } else { | |
1925 | if (ret < 0) | |
1926 | if (--retry_max == 0) | |
1927 | goto failed_removal; | |
1928 | yield(); | |
1929 | drain = 1; | |
1930 | goto repeat; | |
1931 | } | |
1932 | } | |
b3834be5 | 1933 | /* drain all zone's lru pagevec, this is asynchronous... */ |
0c0e6195 | 1934 | lru_add_drain_all(); |
0c0e6195 | 1935 | yield(); |
b3834be5 | 1936 | /* drain pcp pages, this is synchronous. */ |
c0554329 | 1937 | drain_all_pages(zone); |
c8721bbb NH |
1938 | /* |
1939 | * dissolve free hugepages in the memory block before doing offlining | |
1940 | * actually in order to make hugetlbfs's object counting consistent. | |
1941 | */ | |
082d5b6b GS |
1942 | ret = dissolve_free_huge_pages(start_pfn, end_pfn); |
1943 | if (ret) | |
1944 | goto failed_removal; | |
0c0e6195 KH |
1945 | /* check again */ |
1946 | offlined_pages = check_pages_isolated(start_pfn, end_pfn); | |
1947 | if (offlined_pages < 0) { | |
1948 | ret = -EBUSY; | |
1949 | goto failed_removal; | |
1950 | } | |
e33e33b4 | 1951 | pr_info("Offlined Pages %ld\n", offlined_pages); |
b3834be5 | 1952 | /* Ok, all of our target is isolated. |
0c0e6195 KH |
1953 | We cannot do rollback at this point. */ |
1954 | offline_isolated_pages(start_pfn, end_pfn); | |
dbc0e4ce | 1955 | /* reset pagetype flags and makes migrate type to be MOVABLE */ |
0815f3d8 | 1956 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
0c0e6195 | 1957 | /* removal success */ |
3dcc0571 | 1958 | adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages); |
0c0e6195 | 1959 | zone->present_pages -= offlined_pages; |
d702909f CS |
1960 | |
1961 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
0c0e6195 | 1962 | zone->zone_pgdat->node_present_pages -= offlined_pages; |
d702909f | 1963 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
7b78d335 | 1964 | |
1b79acc9 KM |
1965 | init_per_zone_wmark_min(); |
1966 | ||
1e8537ba | 1967 | if (!populated_zone(zone)) { |
340175b7 | 1968 | zone_pcp_reset(zone); |
1e8537ba XQ |
1969 | mutex_lock(&zonelists_mutex); |
1970 | build_all_zonelists(NULL, NULL); | |
1971 | mutex_unlock(&zonelists_mutex); | |
1972 | } else | |
1973 | zone_pcp_update(zone); | |
340175b7 | 1974 | |
d9713679 | 1975 | node_states_clear_node(node, &arg); |
698b1b30 | 1976 | if (arg.status_change_nid >= 0) { |
8fe23e05 | 1977 | kswapd_stop(node); |
698b1b30 VB |
1978 | kcompactd_stop(node); |
1979 | } | |
bce7394a | 1980 | |
0c0e6195 KH |
1981 | vm_total_pages = nr_free_pagecache_pages(); |
1982 | writeback_set_ratelimit(); | |
7b78d335 YG |
1983 | |
1984 | memory_notify(MEM_OFFLINE, &arg); | |
0c0e6195 KH |
1985 | return 0; |
1986 | ||
1987 | failed_removal: | |
e33e33b4 CY |
1988 | pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n", |
1989 | (unsigned long long) start_pfn << PAGE_SHIFT, | |
1990 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1); | |
7b78d335 | 1991 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
0c0e6195 | 1992 | /* pushback to free area */ |
0815f3d8 | 1993 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
0c0e6195 KH |
1994 | return ret; |
1995 | } | |
71088785 | 1996 | |
30467e0b | 1997 | /* Must be protected by mem_hotplug_begin() */ |
a16cee10 WC |
1998 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
1999 | { | |
2000 | return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ); | |
2001 | } | |
e2ff3940 | 2002 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
a16cee10 | 2003 | |
bbc76be6 WC |
2004 | /** |
2005 | * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn) | |
2006 | * @start_pfn: start pfn of the memory range | |
e05c4bbf | 2007 | * @end_pfn: end pfn of the memory range |
bbc76be6 WC |
2008 | * @arg: argument passed to func |
2009 | * @func: callback for each memory section walked | |
2010 | * | |
2011 | * This function walks through all present mem sections in range | |
2012 | * [start_pfn, end_pfn) and call func on each mem section. | |
2013 | * | |
2014 | * Returns the return value of func. | |
2015 | */ | |
e2ff3940 | 2016 | int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn, |
bbc76be6 | 2017 | void *arg, int (*func)(struct memory_block *, void *)) |
71088785 | 2018 | { |
e90bdb7f WC |
2019 | struct memory_block *mem = NULL; |
2020 | struct mem_section *section; | |
e90bdb7f WC |
2021 | unsigned long pfn, section_nr; |
2022 | int ret; | |
e90bdb7f WC |
2023 | |
2024 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
2025 | section_nr = pfn_to_section_nr(pfn); | |
2026 | if (!present_section_nr(section_nr)) | |
2027 | continue; | |
2028 | ||
2029 | section = __nr_to_section(section_nr); | |
2030 | /* same memblock? */ | |
2031 | if (mem) | |
2032 | if ((section_nr >= mem->start_section_nr) && | |
2033 | (section_nr <= mem->end_section_nr)) | |
2034 | continue; | |
2035 | ||
2036 | mem = find_memory_block_hinted(section, mem); | |
2037 | if (!mem) | |
2038 | continue; | |
2039 | ||
bbc76be6 | 2040 | ret = func(mem, arg); |
e90bdb7f | 2041 | if (ret) { |
bbc76be6 WC |
2042 | kobject_put(&mem->dev.kobj); |
2043 | return ret; | |
e90bdb7f WC |
2044 | } |
2045 | } | |
2046 | ||
2047 | if (mem) | |
2048 | kobject_put(&mem->dev.kobj); | |
2049 | ||
bbc76be6 WC |
2050 | return 0; |
2051 | } | |
2052 | ||
e2ff3940 | 2053 | #ifdef CONFIG_MEMORY_HOTREMOVE |
d6de9d53 | 2054 | static int check_memblock_offlined_cb(struct memory_block *mem, void *arg) |
bbc76be6 WC |
2055 | { |
2056 | int ret = !is_memblock_offlined(mem); | |
2057 | ||
349daa0f RD |
2058 | if (unlikely(ret)) { |
2059 | phys_addr_t beginpa, endpa; | |
2060 | ||
2061 | beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)); | |
2062 | endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1; | |
756a025f | 2063 | pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n", |
349daa0f RD |
2064 | &beginpa, &endpa); |
2065 | } | |
bbc76be6 WC |
2066 | |
2067 | return ret; | |
2068 | } | |
2069 | ||
0f1cfe9d | 2070 | static int check_cpu_on_node(pg_data_t *pgdat) |
60a5a19e | 2071 | { |
60a5a19e TC |
2072 | int cpu; |
2073 | ||
2074 | for_each_present_cpu(cpu) { | |
2075 | if (cpu_to_node(cpu) == pgdat->node_id) | |
2076 | /* | |
2077 | * the cpu on this node isn't removed, and we can't | |
2078 | * offline this node. | |
2079 | */ | |
2080 | return -EBUSY; | |
2081 | } | |
2082 | ||
2083 | return 0; | |
2084 | } | |
2085 | ||
0f1cfe9d | 2086 | static void unmap_cpu_on_node(pg_data_t *pgdat) |
e13fe869 WC |
2087 | { |
2088 | #ifdef CONFIG_ACPI_NUMA | |
e13fe869 WC |
2089 | int cpu; |
2090 | ||
2091 | for_each_possible_cpu(cpu) | |
2092 | if (cpu_to_node(cpu) == pgdat->node_id) | |
2093 | numa_clear_node(cpu); | |
2094 | #endif | |
2095 | } | |
2096 | ||
0f1cfe9d | 2097 | static int check_and_unmap_cpu_on_node(pg_data_t *pgdat) |
e13fe869 | 2098 | { |
0f1cfe9d | 2099 | int ret; |
e13fe869 | 2100 | |
0f1cfe9d | 2101 | ret = check_cpu_on_node(pgdat); |
e13fe869 WC |
2102 | if (ret) |
2103 | return ret; | |
2104 | ||
2105 | /* | |
2106 | * the node will be offlined when we come here, so we can clear | |
2107 | * the cpu_to_node() now. | |
2108 | */ | |
2109 | ||
0f1cfe9d | 2110 | unmap_cpu_on_node(pgdat); |
e13fe869 WC |
2111 | return 0; |
2112 | } | |
2113 | ||
0f1cfe9d TK |
2114 | /** |
2115 | * try_offline_node | |
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 | { |
d822b86a WC |
2124 | pg_data_t *pgdat = NODE_DATA(nid); |
2125 | unsigned long start_pfn = pgdat->node_start_pfn; | |
2126 | unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages; | |
60a5a19e TC |
2127 | unsigned long pfn; |
2128 | ||
2129 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
2130 | unsigned long section_nr = pfn_to_section_nr(pfn); | |
2131 | ||
2132 | if (!present_section_nr(section_nr)) | |
2133 | continue; | |
2134 | ||
2135 | if (pfn_to_nid(pfn) != nid) | |
2136 | continue; | |
2137 | ||
2138 | /* | |
2139 | * some memory sections of this node are not removed, and we | |
2140 | * can't offline node now. | |
2141 | */ | |
2142 | return; | |
2143 | } | |
2144 | ||
0f1cfe9d | 2145 | if (check_and_unmap_cpu_on_node(pgdat)) |
60a5a19e TC |
2146 | return; |
2147 | ||
2148 | /* | |
2149 | * all memory/cpu of this node are removed, we can offline this | |
2150 | * node now. | |
2151 | */ | |
2152 | node_set_offline(nid); | |
2153 | unregister_one_node(nid); | |
2154 | } | |
90b30cdc | 2155 | EXPORT_SYMBOL(try_offline_node); |
60a5a19e | 2156 | |
0f1cfe9d TK |
2157 | /** |
2158 | * remove_memory | |
2159 | * | |
2160 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
2161 | * and online/offline operations before this call, as required by | |
2162 | * try_offline_node(). | |
2163 | */ | |
242831eb | 2164 | void __ref remove_memory(int nid, u64 start, u64 size) |
bbc76be6 | 2165 | { |
242831eb | 2166 | int ret; |
993c1aad | 2167 | |
27356f54 TK |
2168 | BUG_ON(check_hotplug_memory_range(start, size)); |
2169 | ||
bfc8c901 | 2170 | mem_hotplug_begin(); |
6677e3ea YI |
2171 | |
2172 | /* | |
242831eb RW |
2173 | * All memory blocks must be offlined before removing memory. Check |
2174 | * whether all memory blocks in question are offline and trigger a BUG() | |
2175 | * if this is not the case. | |
6677e3ea | 2176 | */ |
242831eb | 2177 | ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL, |
d6de9d53 | 2178 | check_memblock_offlined_cb); |
bfc8c901 | 2179 | if (ret) |
242831eb | 2180 | BUG(); |
6677e3ea | 2181 | |
46c66c4b YI |
2182 | /* remove memmap entry */ |
2183 | firmware_map_remove(start, start + size, "System RAM"); | |
f9126ab9 XQ |
2184 | memblock_free(start, size); |
2185 | memblock_remove(start, size); | |
46c66c4b | 2186 | |
24d335ca WC |
2187 | arch_remove_memory(start, size); |
2188 | ||
60a5a19e TC |
2189 | try_offline_node(nid); |
2190 | ||
bfc8c901 | 2191 | mem_hotplug_done(); |
71088785 | 2192 | } |
71088785 | 2193 | EXPORT_SYMBOL_GPL(remove_memory); |
aba6efc4 | 2194 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |