]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * linux/mm/memory_hotplug.c | |
3 | * | |
4 | * Copyright (C) | |
5 | */ | |
6 | ||
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> | |
12 | #include <linux/compiler.h> | |
13 | #include <linux/export.h> | |
14 | #include <linux/pagevec.h> | |
15 | #include <linux/writeback.h> | |
16 | #include <linux/slab.h> | |
17 | #include <linux/sysctl.h> | |
18 | #include <linux/cpu.h> | |
19 | #include <linux/memory.h> | |
20 | #include <linux/memremap.h> | |
21 | #include <linux/memory_hotplug.h> | |
22 | #include <linux/highmem.h> | |
23 | #include <linux/vmalloc.h> | |
24 | #include <linux/ioport.h> | |
25 | #include <linux/delay.h> | |
26 | #include <linux/migrate.h> | |
27 | #include <linux/page-isolation.h> | |
28 | #include <linux/pfn.h> | |
29 | #include <linux/suspend.h> | |
30 | #include <linux/mm_inline.h> | |
31 | #include <linux/firmware-map.h> | |
32 | #include <linux/stop_machine.h> | |
33 | #include <linux/hugetlb.h> | |
34 | #include <linux/memblock.h> | |
35 | #include <linux/bootmem.h> | |
36 | #include <linux/compaction.h> | |
37 | ||
38 | #include <asm/tlbflush.h> | |
39 | ||
40 | #include "internal.h" | |
41 | ||
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; | |
52 | static DEFINE_MUTEX(online_page_callback_lock); | |
53 | ||
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 | ||
81 | #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE | |
82 | bool memhp_auto_online; | |
83 | #else | |
84 | bool memhp_auto_online = true; | |
85 | #endif | |
86 | EXPORT_SYMBOL_GPL(memhp_auto_online); | |
87 | ||
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 | ||
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 | } | |
110 | ||
111 | void put_online_mems(void) | |
112 | { | |
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 | ||
125 | } | |
126 | ||
127 | void mem_hotplug_begin(void) | |
128 | { | |
129 | mem_hotplug.active_writer = current; | |
130 | ||
131 | memhp_lock_acquire(); | |
132 | for (;;) { | |
133 | mutex_lock(&mem_hotplug.lock); | |
134 | if (likely(!mem_hotplug.refcount)) | |
135 | break; | |
136 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
137 | mutex_unlock(&mem_hotplug.lock); | |
138 | schedule(); | |
139 | } | |
140 | } | |
141 | ||
142 | void mem_hotplug_done(void) | |
143 | { | |
144 | mem_hotplug.active_writer = NULL; | |
145 | mutex_unlock(&mem_hotplug.lock); | |
146 | memhp_lock_release(); | |
147 | } | |
148 | ||
149 | /* add this memory to iomem resource */ | |
150 | static struct resource *register_memory_resource(u64 start, u64 size) | |
151 | { | |
152 | struct resource *res; | |
153 | res = kzalloc(sizeof(struct resource), GFP_KERNEL); | |
154 | if (!res) | |
155 | return ERR_PTR(-ENOMEM); | |
156 | ||
157 | res->name = "System RAM"; | |
158 | res->start = start; | |
159 | res->end = start + size - 1; | |
160 | res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY; | |
161 | if (request_resource(&iomem_resource, res) < 0) { | |
162 | pr_debug("System RAM resource %pR cannot be added\n", res); | |
163 | kfree(res); | |
164 | return ERR_PTR(-EEXIST); | |
165 | } | |
166 | return res; | |
167 | } | |
168 | ||
169 | static void release_memory_resource(struct resource *res) | |
170 | { | |
171 | if (!res) | |
172 | return; | |
173 | release_resource(res); | |
174 | kfree(res); | |
175 | return; | |
176 | } | |
177 | ||
178 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE | |
179 | void get_page_bootmem(unsigned long info, struct page *page, | |
180 | unsigned long type) | |
181 | { | |
182 | page->lru.next = (struct list_head *) type; | |
183 | SetPagePrivate(page); | |
184 | set_page_private(page, info); | |
185 | page_ref_inc(page); | |
186 | } | |
187 | ||
188 | void put_page_bootmem(struct page *page) | |
189 | { | |
190 | unsigned long type; | |
191 | ||
192 | type = (unsigned long) page->lru.next; | |
193 | BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || | |
194 | type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); | |
195 | ||
196 | if (page_ref_dec_return(page) == 1) { | |
197 | ClearPagePrivate(page); | |
198 | set_page_private(page, 0); | |
199 | INIT_LIST_HEAD(&page->lru); | |
200 | free_reserved_page(page); | |
201 | } | |
202 | } | |
203 | ||
204 | #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE | |
205 | #ifndef CONFIG_SPARSEMEM_VMEMMAP | |
206 | static void register_page_bootmem_info_section(unsigned long start_pfn) | |
207 | { | |
208 | unsigned long *usemap, mapsize, section_nr, i; | |
209 | struct mem_section *ms; | |
210 | struct page *page, *memmap; | |
211 | ||
212 | section_nr = pfn_to_section_nr(start_pfn); | |
213 | ms = __nr_to_section(section_nr); | |
214 | ||
215 | /* Get section's memmap address */ | |
216 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
217 | ||
218 | /* | |
219 | * Get page for the memmap's phys address | |
220 | * XXX: need more consideration for sparse_vmemmap... | |
221 | */ | |
222 | page = virt_to_page(memmap); | |
223 | mapsize = sizeof(struct page) * PAGES_PER_SECTION; | |
224 | mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; | |
225 | ||
226 | /* remember memmap's page */ | |
227 | for (i = 0; i < mapsize; i++, page++) | |
228 | get_page_bootmem(section_nr, page, SECTION_INFO); | |
229 | ||
230 | usemap = __nr_to_section(section_nr)->pageblock_flags; | |
231 | page = virt_to_page(usemap); | |
232 | ||
233 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
234 | ||
235 | for (i = 0; i < mapsize; i++, page++) | |
236 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); | |
237 | ||
238 | } | |
239 | #else /* CONFIG_SPARSEMEM_VMEMMAP */ | |
240 | static void register_page_bootmem_info_section(unsigned long start_pfn) | |
241 | { | |
242 | unsigned long *usemap, mapsize, section_nr, i; | |
243 | struct mem_section *ms; | |
244 | struct page *page, *memmap; | |
245 | ||
246 | if (!pfn_valid(start_pfn)) | |
247 | return; | |
248 | ||
249 | section_nr = pfn_to_section_nr(start_pfn); | |
250 | ms = __nr_to_section(section_nr); | |
251 | ||
252 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); | |
253 | ||
254 | register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION); | |
255 | ||
256 | usemap = __nr_to_section(section_nr)->pageblock_flags; | |
257 | page = virt_to_page(usemap); | |
258 | ||
259 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; | |
260 | ||
261 | for (i = 0; i < mapsize; i++, page++) | |
262 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); | |
263 | } | |
264 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ | |
265 | ||
266 | void __init register_page_bootmem_info_node(struct pglist_data *pgdat) | |
267 | { | |
268 | unsigned long i, pfn, end_pfn, nr_pages; | |
269 | int node = pgdat->node_id; | |
270 | struct page *page; | |
271 | struct zone *zone; | |
272 | ||
273 | nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; | |
274 | page = virt_to_page(pgdat); | |
275 | ||
276 | for (i = 0; i < nr_pages; i++, page++) | |
277 | get_page_bootmem(node, page, NODE_INFO); | |
278 | ||
279 | zone = &pgdat->node_zones[0]; | |
280 | for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) { | |
281 | if (zone_is_initialized(zone)) { | |
282 | nr_pages = zone->wait_table_hash_nr_entries | |
283 | * sizeof(wait_queue_head_t); | |
284 | nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT; | |
285 | page = virt_to_page(zone->wait_table); | |
286 | ||
287 | for (i = 0; i < nr_pages; i++, page++) | |
288 | get_page_bootmem(node, page, NODE_INFO); | |
289 | } | |
290 | } | |
291 | ||
292 | pfn = pgdat->node_start_pfn; | |
293 | end_pfn = pgdat_end_pfn(pgdat); | |
294 | ||
295 | /* register section info */ | |
296 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
297 | /* | |
298 | * Some platforms can assign the same pfn to multiple nodes - on | |
299 | * node0 as well as nodeN. To avoid registering a pfn against | |
300 | * multiple nodes we check that this pfn does not already | |
301 | * reside in some other nodes. | |
302 | */ | |
303 | if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node)) | |
304 | register_page_bootmem_info_section(pfn); | |
305 | } | |
306 | } | |
307 | #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */ | |
308 | ||
309 | static void __meminit grow_zone_span(struct zone *zone, unsigned long start_pfn, | |
310 | unsigned long end_pfn) | |
311 | { | |
312 | unsigned long old_zone_end_pfn; | |
313 | ||
314 | zone_span_writelock(zone); | |
315 | ||
316 | old_zone_end_pfn = zone_end_pfn(zone); | |
317 | if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn) | |
318 | zone->zone_start_pfn = start_pfn; | |
319 | ||
320 | zone->spanned_pages = max(old_zone_end_pfn, end_pfn) - | |
321 | zone->zone_start_pfn; | |
322 | ||
323 | zone_span_writeunlock(zone); | |
324 | } | |
325 | ||
326 | static void resize_zone(struct zone *zone, unsigned long start_pfn, | |
327 | unsigned long end_pfn) | |
328 | { | |
329 | zone_span_writelock(zone); | |
330 | ||
331 | if (end_pfn - start_pfn) { | |
332 | zone->zone_start_pfn = start_pfn; | |
333 | zone->spanned_pages = end_pfn - start_pfn; | |
334 | } else { | |
335 | /* | |
336 | * make it consist as free_area_init_core(), | |
337 | * if spanned_pages = 0, then keep start_pfn = 0 | |
338 | */ | |
339 | zone->zone_start_pfn = 0; | |
340 | zone->spanned_pages = 0; | |
341 | } | |
342 | ||
343 | zone_span_writeunlock(zone); | |
344 | } | |
345 | ||
346 | static void fix_zone_id(struct zone *zone, unsigned long start_pfn, | |
347 | unsigned long end_pfn) | |
348 | { | |
349 | enum zone_type zid = zone_idx(zone); | |
350 | int nid = zone->zone_pgdat->node_id; | |
351 | unsigned long pfn; | |
352 | ||
353 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
354 | set_page_links(pfn_to_page(pfn), zid, nid, pfn); | |
355 | } | |
356 | ||
357 | /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or | |
358 | * alloc_bootmem_node_nopanic()/memblock_virt_alloc_node_nopanic() */ | |
359 | static int __ref ensure_zone_is_initialized(struct zone *zone, | |
360 | unsigned long start_pfn, unsigned long num_pages) | |
361 | { | |
362 | if (!zone_is_initialized(zone)) | |
363 | return init_currently_empty_zone(zone, start_pfn, num_pages); | |
364 | ||
365 | return 0; | |
366 | } | |
367 | ||
368 | static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2, | |
369 | unsigned long start_pfn, unsigned long end_pfn) | |
370 | { | |
371 | int ret; | |
372 | unsigned long flags; | |
373 | unsigned long z1_start_pfn; | |
374 | ||
375 | ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn); | |
376 | if (ret) | |
377 | return ret; | |
378 | ||
379 | pgdat_resize_lock(z1->zone_pgdat, &flags); | |
380 | ||
381 | /* can't move pfns which are higher than @z2 */ | |
382 | if (end_pfn > zone_end_pfn(z2)) | |
383 | goto out_fail; | |
384 | /* the move out part must be at the left most of @z2 */ | |
385 | if (start_pfn > z2->zone_start_pfn) | |
386 | goto out_fail; | |
387 | /* must included/overlap */ | |
388 | if (end_pfn <= z2->zone_start_pfn) | |
389 | goto out_fail; | |
390 | ||
391 | /* use start_pfn for z1's start_pfn if z1 is empty */ | |
392 | if (!zone_is_empty(z1)) | |
393 | z1_start_pfn = z1->zone_start_pfn; | |
394 | else | |
395 | z1_start_pfn = start_pfn; | |
396 | ||
397 | resize_zone(z1, z1_start_pfn, end_pfn); | |
398 | resize_zone(z2, end_pfn, zone_end_pfn(z2)); | |
399 | ||
400 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
401 | ||
402 | fix_zone_id(z1, start_pfn, end_pfn); | |
403 | ||
404 | return 0; | |
405 | out_fail: | |
406 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
407 | return -1; | |
408 | } | |
409 | ||
410 | static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2, | |
411 | unsigned long start_pfn, unsigned long end_pfn) | |
412 | { | |
413 | int ret; | |
414 | unsigned long flags; | |
415 | unsigned long z2_end_pfn; | |
416 | ||
417 | ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn); | |
418 | if (ret) | |
419 | return ret; | |
420 | ||
421 | pgdat_resize_lock(z1->zone_pgdat, &flags); | |
422 | ||
423 | /* can't move pfns which are lower than @z1 */ | |
424 | if (z1->zone_start_pfn > start_pfn) | |
425 | goto out_fail; | |
426 | /* the move out part mast at the right most of @z1 */ | |
427 | if (zone_end_pfn(z1) > end_pfn) | |
428 | goto out_fail; | |
429 | /* must included/overlap */ | |
430 | if (start_pfn >= zone_end_pfn(z1)) | |
431 | goto out_fail; | |
432 | ||
433 | /* use end_pfn for z2's end_pfn if z2 is empty */ | |
434 | if (!zone_is_empty(z2)) | |
435 | z2_end_pfn = zone_end_pfn(z2); | |
436 | else | |
437 | z2_end_pfn = end_pfn; | |
438 | ||
439 | resize_zone(z1, z1->zone_start_pfn, start_pfn); | |
440 | resize_zone(z2, start_pfn, z2_end_pfn); | |
441 | ||
442 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
443 | ||
444 | fix_zone_id(z2, start_pfn, end_pfn); | |
445 | ||
446 | return 0; | |
447 | out_fail: | |
448 | pgdat_resize_unlock(z1->zone_pgdat, &flags); | |
449 | return -1; | |
450 | } | |
451 | ||
452 | static struct zone * __meminit move_pfn_range(int zone_shift, | |
453 | unsigned long start_pfn, unsigned long end_pfn) | |
454 | { | |
455 | struct zone *zone = page_zone(pfn_to_page(start_pfn)); | |
456 | int ret = 0; | |
457 | ||
458 | if (zone_shift < 0) | |
459 | ret = move_pfn_range_left(zone + zone_shift, zone, | |
460 | start_pfn, end_pfn); | |
461 | else if (zone_shift) | |
462 | ret = move_pfn_range_right(zone, zone + zone_shift, | |
463 | start_pfn, end_pfn); | |
464 | ||
465 | if (ret) | |
466 | return NULL; | |
467 | ||
468 | return zone + zone_shift; | |
469 | } | |
470 | ||
471 | static void __meminit grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn, | |
472 | unsigned long end_pfn) | |
473 | { | |
474 | unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat); | |
475 | ||
476 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) | |
477 | pgdat->node_start_pfn = start_pfn; | |
478 | ||
479 | pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) - | |
480 | pgdat->node_start_pfn; | |
481 | } | |
482 | ||
483 | static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn) | |
484 | { | |
485 | struct pglist_data *pgdat = zone->zone_pgdat; | |
486 | int nr_pages = PAGES_PER_SECTION; | |
487 | int nid = pgdat->node_id; | |
488 | int zone_type; | |
489 | unsigned long flags, pfn; | |
490 | int ret; | |
491 | ||
492 | zone_type = zone - pgdat->node_zones; | |
493 | ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages); | |
494 | if (ret) | |
495 | return ret; | |
496 | ||
497 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
498 | grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages); | |
499 | grow_pgdat_span(zone->zone_pgdat, phys_start_pfn, | |
500 | phys_start_pfn + nr_pages); | |
501 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
502 | memmap_init_zone(nr_pages, nid, zone_type, | |
503 | phys_start_pfn, MEMMAP_HOTPLUG); | |
504 | ||
505 | /* online_page_range is called later and expects pages reserved */ | |
506 | for (pfn = phys_start_pfn; pfn < phys_start_pfn + nr_pages; pfn++) { | |
507 | if (!pfn_valid(pfn)) | |
508 | continue; | |
509 | ||
510 | SetPageReserved(pfn_to_page(pfn)); | |
511 | } | |
512 | return 0; | |
513 | } | |
514 | ||
515 | static int __meminit __add_section(int nid, struct zone *zone, | |
516 | unsigned long phys_start_pfn) | |
517 | { | |
518 | int ret; | |
519 | ||
520 | if (pfn_valid(phys_start_pfn)) | |
521 | return -EEXIST; | |
522 | ||
523 | ret = sparse_add_one_section(zone, phys_start_pfn); | |
524 | ||
525 | if (ret < 0) | |
526 | return ret; | |
527 | ||
528 | ret = __add_zone(zone, phys_start_pfn); | |
529 | ||
530 | if (ret < 0) | |
531 | return ret; | |
532 | ||
533 | return register_new_memory(nid, __pfn_to_section(phys_start_pfn)); | |
534 | } | |
535 | ||
536 | /* | |
537 | * Reasonably generic function for adding memory. It is | |
538 | * expected that archs that support memory hotplug will | |
539 | * call this function after deciding the zone to which to | |
540 | * add the new pages. | |
541 | */ | |
542 | int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn, | |
543 | unsigned long nr_pages) | |
544 | { | |
545 | unsigned long i; | |
546 | int err = 0; | |
547 | int start_sec, end_sec; | |
548 | struct vmem_altmap *altmap; | |
549 | ||
550 | clear_zone_contiguous(zone); | |
551 | ||
552 | /* during initialize mem_map, align hot-added range to section */ | |
553 | start_sec = pfn_to_section_nr(phys_start_pfn); | |
554 | end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1); | |
555 | ||
556 | altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn)); | |
557 | if (altmap) { | |
558 | /* | |
559 | * Validate altmap is within bounds of the total request | |
560 | */ | |
561 | if (altmap->base_pfn != phys_start_pfn | |
562 | || vmem_altmap_offset(altmap) > nr_pages) { | |
563 | pr_warn_once("memory add fail, invalid altmap\n"); | |
564 | err = -EINVAL; | |
565 | goto out; | |
566 | } | |
567 | altmap->alloc = 0; | |
568 | } | |
569 | ||
570 | for (i = start_sec; i <= end_sec; i++) { | |
571 | err = __add_section(nid, zone, section_nr_to_pfn(i)); | |
572 | ||
573 | /* | |
574 | * EEXIST is finally dealt with by ioresource collision | |
575 | * check. see add_memory() => register_memory_resource() | |
576 | * Warning will be printed if there is collision. | |
577 | */ | |
578 | if (err && (err != -EEXIST)) | |
579 | break; | |
580 | err = 0; | |
581 | } | |
582 | vmemmap_populate_print_last(); | |
583 | out: | |
584 | set_zone_contiguous(zone); | |
585 | return err; | |
586 | } | |
587 | EXPORT_SYMBOL_GPL(__add_pages); | |
588 | ||
589 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
590 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ | |
591 | static int find_smallest_section_pfn(int nid, struct zone *zone, | |
592 | unsigned long start_pfn, | |
593 | unsigned long end_pfn) | |
594 | { | |
595 | struct mem_section *ms; | |
596 | ||
597 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) { | |
598 | ms = __pfn_to_section(start_pfn); | |
599 | ||
600 | if (unlikely(!valid_section(ms))) | |
601 | continue; | |
602 | ||
603 | if (unlikely(pfn_to_nid(start_pfn) != nid)) | |
604 | continue; | |
605 | ||
606 | if (zone && zone != page_zone(pfn_to_page(start_pfn))) | |
607 | continue; | |
608 | ||
609 | return start_pfn; | |
610 | } | |
611 | ||
612 | return 0; | |
613 | } | |
614 | ||
615 | /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ | |
616 | static int find_biggest_section_pfn(int nid, struct zone *zone, | |
617 | unsigned long start_pfn, | |
618 | unsigned long end_pfn) | |
619 | { | |
620 | struct mem_section *ms; | |
621 | unsigned long pfn; | |
622 | ||
623 | /* pfn is the end pfn of a memory section. */ | |
624 | pfn = end_pfn - 1; | |
625 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) { | |
626 | ms = __pfn_to_section(pfn); | |
627 | ||
628 | if (unlikely(!valid_section(ms))) | |
629 | continue; | |
630 | ||
631 | if (unlikely(pfn_to_nid(pfn) != nid)) | |
632 | continue; | |
633 | ||
634 | if (zone && zone != page_zone(pfn_to_page(pfn))) | |
635 | continue; | |
636 | ||
637 | return pfn; | |
638 | } | |
639 | ||
640 | return 0; | |
641 | } | |
642 | ||
643 | static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, | |
644 | unsigned long end_pfn) | |
645 | { | |
646 | unsigned long zone_start_pfn = zone->zone_start_pfn; | |
647 | unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */ | |
648 | unsigned long zone_end_pfn = z; | |
649 | unsigned long pfn; | |
650 | struct mem_section *ms; | |
651 | int nid = zone_to_nid(zone); | |
652 | ||
653 | zone_span_writelock(zone); | |
654 | if (zone_start_pfn == start_pfn) { | |
655 | /* | |
656 | * If the section is smallest section in the zone, it need | |
657 | * shrink zone->zone_start_pfn and zone->zone_spanned_pages. | |
658 | * In this case, we find second smallest valid mem_section | |
659 | * for shrinking zone. | |
660 | */ | |
661 | pfn = find_smallest_section_pfn(nid, zone, end_pfn, | |
662 | zone_end_pfn); | |
663 | if (pfn) { | |
664 | zone->zone_start_pfn = pfn; | |
665 | zone->spanned_pages = zone_end_pfn - pfn; | |
666 | } | |
667 | } else if (zone_end_pfn == end_pfn) { | |
668 | /* | |
669 | * If the section is biggest section in the zone, it need | |
670 | * shrink zone->spanned_pages. | |
671 | * In this case, we find second biggest valid mem_section for | |
672 | * shrinking zone. | |
673 | */ | |
674 | pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn, | |
675 | start_pfn); | |
676 | if (pfn) | |
677 | zone->spanned_pages = pfn - zone_start_pfn + 1; | |
678 | } | |
679 | ||
680 | /* | |
681 | * The section is not biggest or smallest mem_section in the zone, it | |
682 | * only creates a hole in the zone. So in this case, we need not | |
683 | * change the zone. But perhaps, the zone has only hole data. Thus | |
684 | * it check the zone has only hole or not. | |
685 | */ | |
686 | pfn = zone_start_pfn; | |
687 | for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) { | |
688 | ms = __pfn_to_section(pfn); | |
689 | ||
690 | if (unlikely(!valid_section(ms))) | |
691 | continue; | |
692 | ||
693 | if (page_zone(pfn_to_page(pfn)) != zone) | |
694 | continue; | |
695 | ||
696 | /* If the section is current section, it continues the loop */ | |
697 | if (start_pfn == pfn) | |
698 | continue; | |
699 | ||
700 | /* If we find valid section, we have nothing to do */ | |
701 | zone_span_writeunlock(zone); | |
702 | return; | |
703 | } | |
704 | ||
705 | /* The zone has no valid section */ | |
706 | zone->zone_start_pfn = 0; | |
707 | zone->spanned_pages = 0; | |
708 | zone_span_writeunlock(zone); | |
709 | } | |
710 | ||
711 | static void shrink_pgdat_span(struct pglist_data *pgdat, | |
712 | unsigned long start_pfn, unsigned long end_pfn) | |
713 | { | |
714 | unsigned long pgdat_start_pfn = pgdat->node_start_pfn; | |
715 | unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */ | |
716 | unsigned long pgdat_end_pfn = p; | |
717 | unsigned long pfn; | |
718 | struct mem_section *ms; | |
719 | int nid = pgdat->node_id; | |
720 | ||
721 | if (pgdat_start_pfn == start_pfn) { | |
722 | /* | |
723 | * If the section is smallest section in the pgdat, it need | |
724 | * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages. | |
725 | * In this case, we find second smallest valid mem_section | |
726 | * for shrinking zone. | |
727 | */ | |
728 | pfn = find_smallest_section_pfn(nid, NULL, end_pfn, | |
729 | pgdat_end_pfn); | |
730 | if (pfn) { | |
731 | pgdat->node_start_pfn = pfn; | |
732 | pgdat->node_spanned_pages = pgdat_end_pfn - pfn; | |
733 | } | |
734 | } else if (pgdat_end_pfn == end_pfn) { | |
735 | /* | |
736 | * If the section is biggest section in the pgdat, it need | |
737 | * shrink pgdat->node_spanned_pages. | |
738 | * In this case, we find second biggest valid mem_section for | |
739 | * shrinking zone. | |
740 | */ | |
741 | pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn, | |
742 | start_pfn); | |
743 | if (pfn) | |
744 | pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1; | |
745 | } | |
746 | ||
747 | /* | |
748 | * If the section is not biggest or smallest mem_section in the pgdat, | |
749 | * it only creates a hole in the pgdat. So in this case, we need not | |
750 | * change the pgdat. | |
751 | * But perhaps, the pgdat has only hole data. Thus it check the pgdat | |
752 | * has only hole or not. | |
753 | */ | |
754 | pfn = pgdat_start_pfn; | |
755 | for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) { | |
756 | ms = __pfn_to_section(pfn); | |
757 | ||
758 | if (unlikely(!valid_section(ms))) | |
759 | continue; | |
760 | ||
761 | if (pfn_to_nid(pfn) != nid) | |
762 | continue; | |
763 | ||
764 | /* If the section is current section, it continues the loop */ | |
765 | if (start_pfn == pfn) | |
766 | continue; | |
767 | ||
768 | /* If we find valid section, we have nothing to do */ | |
769 | return; | |
770 | } | |
771 | ||
772 | /* The pgdat has no valid section */ | |
773 | pgdat->node_start_pfn = 0; | |
774 | pgdat->node_spanned_pages = 0; | |
775 | } | |
776 | ||
777 | static void __remove_zone(struct zone *zone, unsigned long start_pfn) | |
778 | { | |
779 | struct pglist_data *pgdat = zone->zone_pgdat; | |
780 | int nr_pages = PAGES_PER_SECTION; | |
781 | int zone_type; | |
782 | unsigned long flags; | |
783 | ||
784 | zone_type = zone - pgdat->node_zones; | |
785 | ||
786 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
787 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); | |
788 | shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages); | |
789 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
790 | } | |
791 | ||
792 | static int __remove_section(struct zone *zone, struct mem_section *ms, | |
793 | unsigned long map_offset) | |
794 | { | |
795 | unsigned long start_pfn; | |
796 | int scn_nr; | |
797 | int ret = -EINVAL; | |
798 | ||
799 | if (!valid_section(ms)) | |
800 | return ret; | |
801 | ||
802 | ret = unregister_memory_section(ms); | |
803 | if (ret) | |
804 | return ret; | |
805 | ||
806 | scn_nr = __section_nr(ms); | |
807 | start_pfn = section_nr_to_pfn(scn_nr); | |
808 | __remove_zone(zone, start_pfn); | |
809 | ||
810 | sparse_remove_one_section(zone, ms, map_offset); | |
811 | return 0; | |
812 | } | |
813 | ||
814 | /** | |
815 | * __remove_pages() - remove sections of pages from a zone | |
816 | * @zone: zone from which pages need to be removed | |
817 | * @phys_start_pfn: starting pageframe (must be aligned to start of a section) | |
818 | * @nr_pages: number of pages to remove (must be multiple of section size) | |
819 | * | |
820 | * Generic helper function to remove section mappings and sysfs entries | |
821 | * for the section of the memory we are removing. Caller needs to make | |
822 | * sure that pages are marked reserved and zones are adjust properly by | |
823 | * calling offline_pages(). | |
824 | */ | |
825 | int __remove_pages(struct zone *zone, unsigned long phys_start_pfn, | |
826 | unsigned long nr_pages) | |
827 | { | |
828 | unsigned long i; | |
829 | unsigned long map_offset = 0; | |
830 | int sections_to_remove, ret = 0; | |
831 | ||
832 | /* In the ZONE_DEVICE case device driver owns the memory region */ | |
833 | if (is_dev_zone(zone)) { | |
834 | struct page *page = pfn_to_page(phys_start_pfn); | |
835 | struct vmem_altmap *altmap; | |
836 | ||
837 | altmap = to_vmem_altmap((unsigned long) page); | |
838 | if (altmap) | |
839 | map_offset = vmem_altmap_offset(altmap); | |
840 | } else { | |
841 | resource_size_t start, size; | |
842 | ||
843 | start = phys_start_pfn << PAGE_SHIFT; | |
844 | size = nr_pages * PAGE_SIZE; | |
845 | ||
846 | ret = release_mem_region_adjustable(&iomem_resource, start, | |
847 | size); | |
848 | if (ret) { | |
849 | resource_size_t endres = start + size - 1; | |
850 | ||
851 | pr_warn("Unable to release resource <%pa-%pa> (%d)\n", | |
852 | &start, &endres, ret); | |
853 | } | |
854 | } | |
855 | ||
856 | clear_zone_contiguous(zone); | |
857 | ||
858 | /* | |
859 | * We can only remove entire sections | |
860 | */ | |
861 | BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK); | |
862 | BUG_ON(nr_pages % PAGES_PER_SECTION); | |
863 | ||
864 | sections_to_remove = nr_pages / PAGES_PER_SECTION; | |
865 | for (i = 0; i < sections_to_remove; i++) { | |
866 | unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; | |
867 | ||
868 | ret = __remove_section(zone, __pfn_to_section(pfn), map_offset); | |
869 | map_offset = 0; | |
870 | if (ret) | |
871 | break; | |
872 | } | |
873 | ||
874 | set_zone_contiguous(zone); | |
875 | ||
876 | return ret; | |
877 | } | |
878 | EXPORT_SYMBOL_GPL(__remove_pages); | |
879 | #endif /* CONFIG_MEMORY_HOTREMOVE */ | |
880 | ||
881 | int set_online_page_callback(online_page_callback_t callback) | |
882 | { | |
883 | int rc = -EINVAL; | |
884 | ||
885 | get_online_mems(); | |
886 | mutex_lock(&online_page_callback_lock); | |
887 | ||
888 | if (online_page_callback == generic_online_page) { | |
889 | online_page_callback = callback; | |
890 | rc = 0; | |
891 | } | |
892 | ||
893 | mutex_unlock(&online_page_callback_lock); | |
894 | put_online_mems(); | |
895 | ||
896 | return rc; | |
897 | } | |
898 | EXPORT_SYMBOL_GPL(set_online_page_callback); | |
899 | ||
900 | int restore_online_page_callback(online_page_callback_t callback) | |
901 | { | |
902 | int rc = -EINVAL; | |
903 | ||
904 | get_online_mems(); | |
905 | mutex_lock(&online_page_callback_lock); | |
906 | ||
907 | if (online_page_callback == callback) { | |
908 | online_page_callback = generic_online_page; | |
909 | rc = 0; | |
910 | } | |
911 | ||
912 | mutex_unlock(&online_page_callback_lock); | |
913 | put_online_mems(); | |
914 | ||
915 | return rc; | |
916 | } | |
917 | EXPORT_SYMBOL_GPL(restore_online_page_callback); | |
918 | ||
919 | void __online_page_set_limits(struct page *page) | |
920 | { | |
921 | } | |
922 | EXPORT_SYMBOL_GPL(__online_page_set_limits); | |
923 | ||
924 | void __online_page_increment_counters(struct page *page) | |
925 | { | |
926 | adjust_managed_page_count(page, 1); | |
927 | } | |
928 | EXPORT_SYMBOL_GPL(__online_page_increment_counters); | |
929 | ||
930 | void __online_page_free(struct page *page) | |
931 | { | |
932 | __free_reserved_page(page); | |
933 | } | |
934 | EXPORT_SYMBOL_GPL(__online_page_free); | |
935 | ||
936 | static void generic_online_page(struct page *page) | |
937 | { | |
938 | __online_page_set_limits(page); | |
939 | __online_page_increment_counters(page); | |
940 | __online_page_free(page); | |
941 | } | |
942 | ||
943 | static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, | |
944 | void *arg) | |
945 | { | |
946 | unsigned long i; | |
947 | unsigned long onlined_pages = *(unsigned long *)arg; | |
948 | struct page *page; | |
949 | if (PageReserved(pfn_to_page(start_pfn))) | |
950 | for (i = 0; i < nr_pages; i++) { | |
951 | page = pfn_to_page(start_pfn + i); | |
952 | (*online_page_callback)(page); | |
953 | onlined_pages++; | |
954 | } | |
955 | *(unsigned long *)arg = onlined_pages; | |
956 | return 0; | |
957 | } | |
958 | ||
959 | #ifdef CONFIG_MOVABLE_NODE | |
960 | /* | |
961 | * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have | |
962 | * normal memory. | |
963 | */ | |
964 | static bool can_online_high_movable(struct zone *zone) | |
965 | { | |
966 | return true; | |
967 | } | |
968 | #else /* CONFIG_MOVABLE_NODE */ | |
969 | /* ensure every online node has NORMAL memory */ | |
970 | static bool can_online_high_movable(struct zone *zone) | |
971 | { | |
972 | return node_state(zone_to_nid(zone), N_NORMAL_MEMORY); | |
973 | } | |
974 | #endif /* CONFIG_MOVABLE_NODE */ | |
975 | ||
976 | /* check which state of node_states will be changed when online memory */ | |
977 | static void node_states_check_changes_online(unsigned long nr_pages, | |
978 | struct zone *zone, struct memory_notify *arg) | |
979 | { | |
980 | int nid = zone_to_nid(zone); | |
981 | enum zone_type zone_last = ZONE_NORMAL; | |
982 | ||
983 | /* | |
984 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] | |
985 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
986 | * set zone_last to ZONE_NORMAL. | |
987 | * | |
988 | * If we don't have HIGHMEM nor movable node, | |
989 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
990 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
991 | */ | |
992 | if (N_MEMORY == N_NORMAL_MEMORY) | |
993 | zone_last = ZONE_MOVABLE; | |
994 | ||
995 | /* | |
996 | * if the memory to be online is in a zone of 0...zone_last, and | |
997 | * the zones of 0...zone_last don't have memory before online, we will | |
998 | * need to set the node to node_states[N_NORMAL_MEMORY] after | |
999 | * the memory is online. | |
1000 | */ | |
1001 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY)) | |
1002 | arg->status_change_nid_normal = nid; | |
1003 | else | |
1004 | arg->status_change_nid_normal = -1; | |
1005 | ||
1006 | #ifdef CONFIG_HIGHMEM | |
1007 | /* | |
1008 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
1009 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
1010 | * set zone_last to ZONE_HIGHMEM. | |
1011 | * | |
1012 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
1013 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
1014 | * set zone_last to ZONE_MOVABLE. | |
1015 | */ | |
1016 | zone_last = ZONE_HIGHMEM; | |
1017 | if (N_MEMORY == N_HIGH_MEMORY) | |
1018 | zone_last = ZONE_MOVABLE; | |
1019 | ||
1020 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY)) | |
1021 | arg->status_change_nid_high = nid; | |
1022 | else | |
1023 | arg->status_change_nid_high = -1; | |
1024 | #else | |
1025 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
1026 | #endif | |
1027 | ||
1028 | /* | |
1029 | * if the node don't have memory befor online, we will need to | |
1030 | * set the node to node_states[N_MEMORY] after the memory | |
1031 | * is online. | |
1032 | */ | |
1033 | if (!node_state(nid, N_MEMORY)) | |
1034 | arg->status_change_nid = nid; | |
1035 | else | |
1036 | arg->status_change_nid = -1; | |
1037 | } | |
1038 | ||
1039 | static void node_states_set_node(int node, struct memory_notify *arg) | |
1040 | { | |
1041 | if (arg->status_change_nid_normal >= 0) | |
1042 | node_set_state(node, N_NORMAL_MEMORY); | |
1043 | ||
1044 | if (arg->status_change_nid_high >= 0) | |
1045 | node_set_state(node, N_HIGH_MEMORY); | |
1046 | ||
1047 | node_set_state(node, N_MEMORY); | |
1048 | } | |
1049 | ||
1050 | int zone_can_shift(unsigned long pfn, unsigned long nr_pages, | |
1051 | enum zone_type target) | |
1052 | { | |
1053 | struct zone *zone = page_zone(pfn_to_page(pfn)); | |
1054 | enum zone_type idx = zone_idx(zone); | |
1055 | int i; | |
1056 | ||
1057 | if (idx < target) { | |
1058 | /* pages must be at end of current zone */ | |
1059 | if (pfn + nr_pages != zone_end_pfn(zone)) | |
1060 | return 0; | |
1061 | ||
1062 | /* no zones in use between current zone and target */ | |
1063 | for (i = idx + 1; i < target; i++) | |
1064 | if (zone_is_initialized(zone - idx + i)) | |
1065 | return 0; | |
1066 | } | |
1067 | ||
1068 | if (target < idx) { | |
1069 | /* pages must be at beginning of current zone */ | |
1070 | if (pfn != zone->zone_start_pfn) | |
1071 | return 0; | |
1072 | ||
1073 | /* no zones in use between current zone and target */ | |
1074 | for (i = target + 1; i < idx; i++) | |
1075 | if (zone_is_initialized(zone - idx + i)) | |
1076 | return 0; | |
1077 | } | |
1078 | ||
1079 | return target - idx; | |
1080 | } | |
1081 | ||
1082 | /* Must be protected by mem_hotplug_begin() */ | |
1083 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type) | |
1084 | { | |
1085 | unsigned long flags; | |
1086 | unsigned long onlined_pages = 0; | |
1087 | struct zone *zone; | |
1088 | int need_zonelists_rebuild = 0; | |
1089 | int nid; | |
1090 | int ret; | |
1091 | struct memory_notify arg; | |
1092 | int zone_shift = 0; | |
1093 | ||
1094 | /* | |
1095 | * This doesn't need a lock to do pfn_to_page(). | |
1096 | * The section can't be removed here because of the | |
1097 | * memory_block->state_mutex. | |
1098 | */ | |
1099 | zone = page_zone(pfn_to_page(pfn)); | |
1100 | ||
1101 | if ((zone_idx(zone) > ZONE_NORMAL || | |
1102 | online_type == MMOP_ONLINE_MOVABLE) && | |
1103 | !can_online_high_movable(zone)) | |
1104 | return -EINVAL; | |
1105 | ||
1106 | if (online_type == MMOP_ONLINE_KERNEL) | |
1107 | zone_shift = zone_can_shift(pfn, nr_pages, ZONE_NORMAL); | |
1108 | else if (online_type == MMOP_ONLINE_MOVABLE) | |
1109 | zone_shift = zone_can_shift(pfn, nr_pages, ZONE_MOVABLE); | |
1110 | ||
1111 | zone = move_pfn_range(zone_shift, pfn, pfn + nr_pages); | |
1112 | if (!zone) | |
1113 | return -EINVAL; | |
1114 | ||
1115 | arg.start_pfn = pfn; | |
1116 | arg.nr_pages = nr_pages; | |
1117 | node_states_check_changes_online(nr_pages, zone, &arg); | |
1118 | ||
1119 | nid = zone_to_nid(zone); | |
1120 | ||
1121 | ret = memory_notify(MEM_GOING_ONLINE, &arg); | |
1122 | ret = notifier_to_errno(ret); | |
1123 | if (ret) | |
1124 | goto failed_addition; | |
1125 | ||
1126 | /* | |
1127 | * If this zone is not populated, then it is not in zonelist. | |
1128 | * This means the page allocator ignores this zone. | |
1129 | * So, zonelist must be updated after online. | |
1130 | */ | |
1131 | mutex_lock(&zonelists_mutex); | |
1132 | if (!populated_zone(zone)) { | |
1133 | need_zonelists_rebuild = 1; | |
1134 | build_all_zonelists(NULL, zone); | |
1135 | } | |
1136 | ||
1137 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, | |
1138 | online_pages_range); | |
1139 | if (ret) { | |
1140 | if (need_zonelists_rebuild) | |
1141 | zone_pcp_reset(zone); | |
1142 | mutex_unlock(&zonelists_mutex); | |
1143 | goto failed_addition; | |
1144 | } | |
1145 | ||
1146 | zone->present_pages += onlined_pages; | |
1147 | ||
1148 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
1149 | zone->zone_pgdat->node_present_pages += onlined_pages; | |
1150 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
1151 | ||
1152 | if (onlined_pages) { | |
1153 | node_states_set_node(nid, &arg); | |
1154 | if (need_zonelists_rebuild) | |
1155 | build_all_zonelists(NULL, NULL); | |
1156 | else | |
1157 | zone_pcp_update(zone); | |
1158 | } | |
1159 | ||
1160 | mutex_unlock(&zonelists_mutex); | |
1161 | ||
1162 | init_per_zone_wmark_min(); | |
1163 | ||
1164 | if (onlined_pages) { | |
1165 | kswapd_run(nid); | |
1166 | kcompactd_run(nid); | |
1167 | } | |
1168 | ||
1169 | vm_total_pages = nr_free_pagecache_pages(); | |
1170 | ||
1171 | writeback_set_ratelimit(); | |
1172 | ||
1173 | if (onlined_pages) | |
1174 | memory_notify(MEM_ONLINE, &arg); | |
1175 | return 0; | |
1176 | ||
1177 | failed_addition: | |
1178 | pr_debug("online_pages [mem %#010llx-%#010llx] failed\n", | |
1179 | (unsigned long long) pfn << PAGE_SHIFT, | |
1180 | (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1); | |
1181 | memory_notify(MEM_CANCEL_ONLINE, &arg); | |
1182 | return ret; | |
1183 | } | |
1184 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ | |
1185 | ||
1186 | static void reset_node_present_pages(pg_data_t *pgdat) | |
1187 | { | |
1188 | struct zone *z; | |
1189 | ||
1190 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) | |
1191 | z->present_pages = 0; | |
1192 | ||
1193 | pgdat->node_present_pages = 0; | |
1194 | } | |
1195 | ||
1196 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ | |
1197 | static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) | |
1198 | { | |
1199 | struct pglist_data *pgdat; | |
1200 | unsigned long zones_size[MAX_NR_ZONES] = {0}; | |
1201 | unsigned long zholes_size[MAX_NR_ZONES] = {0}; | |
1202 | unsigned long start_pfn = PFN_DOWN(start); | |
1203 | ||
1204 | pgdat = NODE_DATA(nid); | |
1205 | if (!pgdat) { | |
1206 | pgdat = arch_alloc_nodedata(nid); | |
1207 | if (!pgdat) | |
1208 | return NULL; | |
1209 | ||
1210 | arch_refresh_nodedata(nid, pgdat); | |
1211 | } else { | |
1212 | /* Reset the nr_zones, order and classzone_idx before reuse */ | |
1213 | pgdat->nr_zones = 0; | |
1214 | pgdat->kswapd_order = 0; | |
1215 | pgdat->kswapd_classzone_idx = 0; | |
1216 | } | |
1217 | ||
1218 | /* we can use NODE_DATA(nid) from here */ | |
1219 | ||
1220 | /* init node's zones as empty zones, we don't have any present pages.*/ | |
1221 | free_area_init_node(nid, zones_size, start_pfn, zholes_size); | |
1222 | pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat); | |
1223 | ||
1224 | /* | |
1225 | * The node we allocated has no zone fallback lists. For avoiding | |
1226 | * to access not-initialized zonelist, build here. | |
1227 | */ | |
1228 | mutex_lock(&zonelists_mutex); | |
1229 | build_all_zonelists(pgdat, NULL); | |
1230 | mutex_unlock(&zonelists_mutex); | |
1231 | ||
1232 | /* | |
1233 | * zone->managed_pages is set to an approximate value in | |
1234 | * free_area_init_core(), which will cause | |
1235 | * /sys/device/system/node/nodeX/meminfo has wrong data. | |
1236 | * So reset it to 0 before any memory is onlined. | |
1237 | */ | |
1238 | reset_node_managed_pages(pgdat); | |
1239 | ||
1240 | /* | |
1241 | * When memory is hot-added, all the memory is in offline state. So | |
1242 | * clear all zones' present_pages because they will be updated in | |
1243 | * online_pages() and offline_pages(). | |
1244 | */ | |
1245 | reset_node_present_pages(pgdat); | |
1246 | ||
1247 | return pgdat; | |
1248 | } | |
1249 | ||
1250 | static void rollback_node_hotadd(int nid, pg_data_t *pgdat) | |
1251 | { | |
1252 | arch_refresh_nodedata(nid, NULL); | |
1253 | free_percpu(pgdat->per_cpu_nodestats); | |
1254 | arch_free_nodedata(pgdat); | |
1255 | return; | |
1256 | } | |
1257 | ||
1258 | ||
1259 | /** | |
1260 | * try_online_node - online a node if offlined | |
1261 | * | |
1262 | * called by cpu_up() to online a node without onlined memory. | |
1263 | */ | |
1264 | int try_online_node(int nid) | |
1265 | { | |
1266 | pg_data_t *pgdat; | |
1267 | int ret; | |
1268 | ||
1269 | if (node_online(nid)) | |
1270 | return 0; | |
1271 | ||
1272 | mem_hotplug_begin(); | |
1273 | pgdat = hotadd_new_pgdat(nid, 0); | |
1274 | if (!pgdat) { | |
1275 | pr_err("Cannot online node %d due to NULL pgdat\n", nid); | |
1276 | ret = -ENOMEM; | |
1277 | goto out; | |
1278 | } | |
1279 | node_set_online(nid); | |
1280 | ret = register_one_node(nid); | |
1281 | BUG_ON(ret); | |
1282 | ||
1283 | if (pgdat->node_zonelists->_zonerefs->zone == NULL) { | |
1284 | mutex_lock(&zonelists_mutex); | |
1285 | build_all_zonelists(NULL, NULL); | |
1286 | mutex_unlock(&zonelists_mutex); | |
1287 | } | |
1288 | ||
1289 | out: | |
1290 | mem_hotplug_done(); | |
1291 | return ret; | |
1292 | } | |
1293 | ||
1294 | static int check_hotplug_memory_range(u64 start, u64 size) | |
1295 | { | |
1296 | u64 start_pfn = PFN_DOWN(start); | |
1297 | u64 nr_pages = size >> PAGE_SHIFT; | |
1298 | ||
1299 | /* Memory range must be aligned with section */ | |
1300 | if ((start_pfn & ~PAGE_SECTION_MASK) || | |
1301 | (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) { | |
1302 | pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n", | |
1303 | (unsigned long long)start, | |
1304 | (unsigned long long)size); | |
1305 | return -EINVAL; | |
1306 | } | |
1307 | ||
1308 | return 0; | |
1309 | } | |
1310 | ||
1311 | /* | |
1312 | * If movable zone has already been setup, newly added memory should be check. | |
1313 | * If its address is higher than movable zone, it should be added as movable. | |
1314 | * Without this check, movable zone may overlap with other zone. | |
1315 | */ | |
1316 | static int should_add_memory_movable(int nid, u64 start, u64 size) | |
1317 | { | |
1318 | unsigned long start_pfn = start >> PAGE_SHIFT; | |
1319 | pg_data_t *pgdat = NODE_DATA(nid); | |
1320 | struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE; | |
1321 | ||
1322 | if (zone_is_empty(movable_zone)) | |
1323 | return 0; | |
1324 | ||
1325 | if (movable_zone->zone_start_pfn <= start_pfn) | |
1326 | return 1; | |
1327 | ||
1328 | return 0; | |
1329 | } | |
1330 | ||
1331 | int zone_for_memory(int nid, u64 start, u64 size, int zone_default, | |
1332 | bool for_device) | |
1333 | { | |
1334 | #ifdef CONFIG_ZONE_DEVICE | |
1335 | if (for_device) | |
1336 | return ZONE_DEVICE; | |
1337 | #endif | |
1338 | if (should_add_memory_movable(nid, start, size)) | |
1339 | return ZONE_MOVABLE; | |
1340 | ||
1341 | return zone_default; | |
1342 | } | |
1343 | ||
1344 | static int online_memory_block(struct memory_block *mem, void *arg) | |
1345 | { | |
1346 | return memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE); | |
1347 | } | |
1348 | ||
1349 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ | |
1350 | int __ref add_memory_resource(int nid, struct resource *res, bool online) | |
1351 | { | |
1352 | u64 start, size; | |
1353 | pg_data_t *pgdat = NULL; | |
1354 | bool new_pgdat; | |
1355 | bool new_node; | |
1356 | int ret; | |
1357 | ||
1358 | start = res->start; | |
1359 | size = resource_size(res); | |
1360 | ||
1361 | ret = check_hotplug_memory_range(start, size); | |
1362 | if (ret) | |
1363 | return ret; | |
1364 | ||
1365 | { /* Stupid hack to suppress address-never-null warning */ | |
1366 | void *p = NODE_DATA(nid); | |
1367 | new_pgdat = !p; | |
1368 | } | |
1369 | ||
1370 | mem_hotplug_begin(); | |
1371 | ||
1372 | /* | |
1373 | * Add new range to memblock so that when hotadd_new_pgdat() is called | |
1374 | * to allocate new pgdat, get_pfn_range_for_nid() will be able to find | |
1375 | * this new range and calculate total pages correctly. The range will | |
1376 | * be removed at hot-remove time. | |
1377 | */ | |
1378 | memblock_add_node(start, size, nid); | |
1379 | ||
1380 | new_node = !node_online(nid); | |
1381 | if (new_node) { | |
1382 | pgdat = hotadd_new_pgdat(nid, start); | |
1383 | ret = -ENOMEM; | |
1384 | if (!pgdat) | |
1385 | goto error; | |
1386 | } | |
1387 | ||
1388 | /* call arch's memory hotadd */ | |
1389 | ret = arch_add_memory(nid, start, size, false); | |
1390 | ||
1391 | if (ret < 0) | |
1392 | goto error; | |
1393 | ||
1394 | /* we online node here. we can't roll back from here. */ | |
1395 | node_set_online(nid); | |
1396 | ||
1397 | if (new_node) { | |
1398 | ret = register_one_node(nid); | |
1399 | /* | |
1400 | * If sysfs file of new node can't create, cpu on the node | |
1401 | * can't be hot-added. There is no rollback way now. | |
1402 | * So, check by BUG_ON() to catch it reluctantly.. | |
1403 | */ | |
1404 | BUG_ON(ret); | |
1405 | } | |
1406 | ||
1407 | /* create new memmap entry */ | |
1408 | firmware_map_add_hotplug(start, start + size, "System RAM"); | |
1409 | ||
1410 | /* online pages if requested */ | |
1411 | if (online) | |
1412 | walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), | |
1413 | NULL, online_memory_block); | |
1414 | ||
1415 | goto out; | |
1416 | ||
1417 | error: | |
1418 | /* rollback pgdat allocation and others */ | |
1419 | if (new_pgdat) | |
1420 | rollback_node_hotadd(nid, pgdat); | |
1421 | memblock_remove(start, size); | |
1422 | ||
1423 | out: | |
1424 | mem_hotplug_done(); | |
1425 | return ret; | |
1426 | } | |
1427 | EXPORT_SYMBOL_GPL(add_memory_resource); | |
1428 | ||
1429 | int __ref add_memory(int nid, u64 start, u64 size) | |
1430 | { | |
1431 | struct resource *res; | |
1432 | int ret; | |
1433 | ||
1434 | res = register_memory_resource(start, size); | |
1435 | if (IS_ERR(res)) | |
1436 | return PTR_ERR(res); | |
1437 | ||
1438 | ret = add_memory_resource(nid, res, memhp_auto_online); | |
1439 | if (ret < 0) | |
1440 | release_memory_resource(res); | |
1441 | return ret; | |
1442 | } | |
1443 | EXPORT_SYMBOL_GPL(add_memory); | |
1444 | ||
1445 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
1446 | /* | |
1447 | * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy | |
1448 | * set and the size of the free page is given by page_order(). Using this, | |
1449 | * the function determines if the pageblock contains only free pages. | |
1450 | * Due to buddy contraints, a free page at least the size of a pageblock will | |
1451 | * be located at the start of the pageblock | |
1452 | */ | |
1453 | static inline int pageblock_free(struct page *page) | |
1454 | { | |
1455 | return PageBuddy(page) && page_order(page) >= pageblock_order; | |
1456 | } | |
1457 | ||
1458 | /* Return the start of the next active pageblock after a given page */ | |
1459 | static struct page *next_active_pageblock(struct page *page) | |
1460 | { | |
1461 | /* Ensure the starting page is pageblock-aligned */ | |
1462 | BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1)); | |
1463 | ||
1464 | /* If the entire pageblock is free, move to the end of free page */ | |
1465 | if (pageblock_free(page)) { | |
1466 | int order; | |
1467 | /* be careful. we don't have locks, page_order can be changed.*/ | |
1468 | order = page_order(page); | |
1469 | if ((order < MAX_ORDER) && (order >= pageblock_order)) | |
1470 | return page + (1 << order); | |
1471 | } | |
1472 | ||
1473 | return page + pageblock_nr_pages; | |
1474 | } | |
1475 | ||
1476 | /* Checks if this range of memory is likely to be hot-removable. */ | |
1477 | bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) | |
1478 | { | |
1479 | struct page *page = pfn_to_page(start_pfn); | |
1480 | struct page *end_page = page + nr_pages; | |
1481 | ||
1482 | /* Check the starting page of each pageblock within the range */ | |
1483 | for (; page < end_page; page = next_active_pageblock(page)) { | |
1484 | if (!is_pageblock_removable_nolock(page)) | |
1485 | return false; | |
1486 | cond_resched(); | |
1487 | } | |
1488 | ||
1489 | /* All pageblocks in the memory block are likely to be hot-removable */ | |
1490 | return true; | |
1491 | } | |
1492 | ||
1493 | /* | |
1494 | * Confirm all pages in a range [start, end) is belongs to the same zone. | |
1495 | */ | |
1496 | int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn) | |
1497 | { | |
1498 | unsigned long pfn, sec_end_pfn; | |
1499 | struct zone *zone = NULL; | |
1500 | struct page *page; | |
1501 | int i; | |
1502 | for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn); | |
1503 | pfn < end_pfn; | |
1504 | pfn = sec_end_pfn + 1, sec_end_pfn += PAGES_PER_SECTION) { | |
1505 | /* Make sure the memory section is present first */ | |
1506 | if (!present_section_nr(pfn_to_section_nr(pfn))) | |
1507 | continue; | |
1508 | for (; pfn < sec_end_pfn && pfn < end_pfn; | |
1509 | pfn += MAX_ORDER_NR_PAGES) { | |
1510 | i = 0; | |
1511 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ | |
1512 | while ((i < MAX_ORDER_NR_PAGES) && | |
1513 | !pfn_valid_within(pfn + i)) | |
1514 | i++; | |
1515 | if (i == MAX_ORDER_NR_PAGES) | |
1516 | continue; | |
1517 | page = pfn_to_page(pfn + i); | |
1518 | if (zone && page_zone(page) != zone) | |
1519 | return 0; | |
1520 | zone = page_zone(page); | |
1521 | } | |
1522 | } | |
1523 | return 1; | |
1524 | } | |
1525 | ||
1526 | /* | |
1527 | * Scan pfn range [start,end) to find movable/migratable pages (LRU pages | |
1528 | * and hugepages). We scan pfn because it's much easier than scanning over | |
1529 | * linked list. This function returns the pfn of the first found movable | |
1530 | * page if it's found, otherwise 0. | |
1531 | */ | |
1532 | static unsigned long scan_movable_pages(unsigned long start, unsigned long end) | |
1533 | { | |
1534 | unsigned long pfn; | |
1535 | struct page *page; | |
1536 | for (pfn = start; pfn < end; pfn++) { | |
1537 | if (pfn_valid(pfn)) { | |
1538 | page = pfn_to_page(pfn); | |
1539 | if (PageLRU(page)) | |
1540 | return pfn; | |
1541 | if (PageHuge(page)) { | |
1542 | if (page_huge_active(page)) | |
1543 | return pfn; | |
1544 | else | |
1545 | pfn = round_up(pfn + 1, | |
1546 | 1 << compound_order(page)) - 1; | |
1547 | } | |
1548 | } | |
1549 | } | |
1550 | return 0; | |
1551 | } | |
1552 | ||
1553 | static struct page *new_node_page(struct page *page, unsigned long private, | |
1554 | int **result) | |
1555 | { | |
1556 | gfp_t gfp_mask = GFP_USER | __GFP_MOVABLE; | |
1557 | int nid = page_to_nid(page); | |
1558 | nodemask_t nmask = node_online_map; | |
1559 | struct page *new_page; | |
1560 | ||
1561 | /* | |
1562 | * TODO: allocate a destination hugepage from a nearest neighbor node, | |
1563 | * accordance with memory policy of the user process if possible. For | |
1564 | * now as a simple work-around, we use the next node for destination. | |
1565 | */ | |
1566 | if (PageHuge(page)) | |
1567 | return alloc_huge_page_node(page_hstate(compound_head(page)), | |
1568 | next_node_in(nid, nmask)); | |
1569 | ||
1570 | node_clear(nid, nmask); | |
1571 | if (PageHighMem(page) | |
1572 | || (zone_idx(page_zone(page)) == ZONE_MOVABLE)) | |
1573 | gfp_mask |= __GFP_HIGHMEM; | |
1574 | ||
1575 | new_page = __alloc_pages_nodemask(gfp_mask, 0, | |
1576 | node_zonelist(nid, gfp_mask), &nmask); | |
1577 | if (!new_page) | |
1578 | new_page = __alloc_pages(gfp_mask, 0, | |
1579 | node_zonelist(nid, gfp_mask)); | |
1580 | ||
1581 | return new_page; | |
1582 | } | |
1583 | ||
1584 | #define NR_OFFLINE_AT_ONCE_PAGES (256) | |
1585 | static int | |
1586 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) | |
1587 | { | |
1588 | unsigned long pfn; | |
1589 | struct page *page; | |
1590 | int move_pages = NR_OFFLINE_AT_ONCE_PAGES; | |
1591 | int not_managed = 0; | |
1592 | int ret = 0; | |
1593 | LIST_HEAD(source); | |
1594 | ||
1595 | for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) { | |
1596 | if (!pfn_valid(pfn)) | |
1597 | continue; | |
1598 | page = pfn_to_page(pfn); | |
1599 | ||
1600 | if (PageHuge(page)) { | |
1601 | struct page *head = compound_head(page); | |
1602 | pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1; | |
1603 | if (compound_order(head) > PFN_SECTION_SHIFT) { | |
1604 | ret = -EBUSY; | |
1605 | break; | |
1606 | } | |
1607 | if (isolate_huge_page(page, &source)) | |
1608 | move_pages -= 1 << compound_order(head); | |
1609 | continue; | |
1610 | } | |
1611 | ||
1612 | if (!get_page_unless_zero(page)) | |
1613 | continue; | |
1614 | /* | |
1615 | * We can skip free pages. And we can only deal with pages on | |
1616 | * LRU. | |
1617 | */ | |
1618 | ret = isolate_lru_page(page); | |
1619 | if (!ret) { /* Success */ | |
1620 | put_page(page); | |
1621 | list_add_tail(&page->lru, &source); | |
1622 | move_pages--; | |
1623 | inc_node_page_state(page, NR_ISOLATED_ANON + | |
1624 | page_is_file_cache(page)); | |
1625 | ||
1626 | } else { | |
1627 | #ifdef CONFIG_DEBUG_VM | |
1628 | pr_alert("removing pfn %lx from LRU failed\n", pfn); | |
1629 | dump_page(page, "failed to remove from LRU"); | |
1630 | #endif | |
1631 | put_page(page); | |
1632 | /* Because we don't have big zone->lock. we should | |
1633 | check this again here. */ | |
1634 | if (page_count(page)) { | |
1635 | not_managed++; | |
1636 | ret = -EBUSY; | |
1637 | break; | |
1638 | } | |
1639 | } | |
1640 | } | |
1641 | if (!list_empty(&source)) { | |
1642 | if (not_managed) { | |
1643 | putback_movable_pages(&source); | |
1644 | goto out; | |
1645 | } | |
1646 | ||
1647 | /* Allocate a new page from the nearest neighbor node */ | |
1648 | ret = migrate_pages(&source, new_node_page, NULL, 0, | |
1649 | MIGRATE_SYNC, MR_MEMORY_HOTPLUG); | |
1650 | if (ret) | |
1651 | putback_movable_pages(&source); | |
1652 | } | |
1653 | out: | |
1654 | return ret; | |
1655 | } | |
1656 | ||
1657 | /* | |
1658 | * remove from free_area[] and mark all as Reserved. | |
1659 | */ | |
1660 | static int | |
1661 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, | |
1662 | void *data) | |
1663 | { | |
1664 | __offline_isolated_pages(start, start + nr_pages); | |
1665 | return 0; | |
1666 | } | |
1667 | ||
1668 | static void | |
1669 | offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
1670 | { | |
1671 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL, | |
1672 | offline_isolated_pages_cb); | |
1673 | } | |
1674 | ||
1675 | /* | |
1676 | * Check all pages in range, recoreded as memory resource, are isolated. | |
1677 | */ | |
1678 | static int | |
1679 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, | |
1680 | void *data) | |
1681 | { | |
1682 | int ret; | |
1683 | long offlined = *(long *)data; | |
1684 | ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true); | |
1685 | offlined = nr_pages; | |
1686 | if (!ret) | |
1687 | *(long *)data += offlined; | |
1688 | return ret; | |
1689 | } | |
1690 | ||
1691 | static long | |
1692 | check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn) | |
1693 | { | |
1694 | long offlined = 0; | |
1695 | int ret; | |
1696 | ||
1697 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined, | |
1698 | check_pages_isolated_cb); | |
1699 | if (ret < 0) | |
1700 | offlined = (long)ret; | |
1701 | return offlined; | |
1702 | } | |
1703 | ||
1704 | #ifdef CONFIG_MOVABLE_NODE | |
1705 | /* | |
1706 | * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have | |
1707 | * normal memory. | |
1708 | */ | |
1709 | static bool can_offline_normal(struct zone *zone, unsigned long nr_pages) | |
1710 | { | |
1711 | return true; | |
1712 | } | |
1713 | #else /* CONFIG_MOVABLE_NODE */ | |
1714 | /* ensure the node has NORMAL memory if it is still online */ | |
1715 | static bool can_offline_normal(struct zone *zone, unsigned long nr_pages) | |
1716 | { | |
1717 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1718 | unsigned long present_pages = 0; | |
1719 | enum zone_type zt; | |
1720 | ||
1721 | for (zt = 0; zt <= ZONE_NORMAL; zt++) | |
1722 | present_pages += pgdat->node_zones[zt].present_pages; | |
1723 | ||
1724 | if (present_pages > nr_pages) | |
1725 | return true; | |
1726 | ||
1727 | present_pages = 0; | |
1728 | for (; zt <= ZONE_MOVABLE; zt++) | |
1729 | present_pages += pgdat->node_zones[zt].present_pages; | |
1730 | ||
1731 | /* | |
1732 | * we can't offline the last normal memory until all | |
1733 | * higher memory is offlined. | |
1734 | */ | |
1735 | return present_pages == 0; | |
1736 | } | |
1737 | #endif /* CONFIG_MOVABLE_NODE */ | |
1738 | ||
1739 | static int __init cmdline_parse_movable_node(char *p) | |
1740 | { | |
1741 | #ifdef CONFIG_MOVABLE_NODE | |
1742 | /* | |
1743 | * Memory used by the kernel cannot be hot-removed because Linux | |
1744 | * cannot migrate the kernel pages. When memory hotplug is | |
1745 | * enabled, we should prevent memblock from allocating memory | |
1746 | * for the kernel. | |
1747 | * | |
1748 | * ACPI SRAT records all hotpluggable memory ranges. But before | |
1749 | * SRAT is parsed, we don't know about it. | |
1750 | * | |
1751 | * The kernel image is loaded into memory at very early time. We | |
1752 | * cannot prevent this anyway. So on NUMA system, we set any | |
1753 | * node the kernel resides in as un-hotpluggable. | |
1754 | * | |
1755 | * Since on modern servers, one node could have double-digit | |
1756 | * gigabytes memory, we can assume the memory around the kernel | |
1757 | * image is also un-hotpluggable. So before SRAT is parsed, just | |
1758 | * allocate memory near the kernel image to try the best to keep | |
1759 | * the kernel away from hotpluggable memory. | |
1760 | */ | |
1761 | memblock_set_bottom_up(true); | |
1762 | movable_node_enabled = true; | |
1763 | #else | |
1764 | pr_warn("movable_node option not supported\n"); | |
1765 | #endif | |
1766 | return 0; | |
1767 | } | |
1768 | early_param("movable_node", cmdline_parse_movable_node); | |
1769 | ||
1770 | /* check which state of node_states will be changed when offline memory */ | |
1771 | static void node_states_check_changes_offline(unsigned long nr_pages, | |
1772 | struct zone *zone, struct memory_notify *arg) | |
1773 | { | |
1774 | struct pglist_data *pgdat = zone->zone_pgdat; | |
1775 | unsigned long present_pages = 0; | |
1776 | enum zone_type zt, zone_last = ZONE_NORMAL; | |
1777 | ||
1778 | /* | |
1779 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] | |
1780 | * contains nodes which have zones of 0...ZONE_NORMAL, | |
1781 | * set zone_last to ZONE_NORMAL. | |
1782 | * | |
1783 | * If we don't have HIGHMEM nor movable node, | |
1784 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of | |
1785 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. | |
1786 | */ | |
1787 | if (N_MEMORY == N_NORMAL_MEMORY) | |
1788 | zone_last = ZONE_MOVABLE; | |
1789 | ||
1790 | /* | |
1791 | * check whether node_states[N_NORMAL_MEMORY] will be changed. | |
1792 | * If the memory to be offline is in a zone of 0...zone_last, | |
1793 | * and it is the last present memory, 0...zone_last will | |
1794 | * become empty after offline , thus we can determind we will | |
1795 | * need to clear the node from node_states[N_NORMAL_MEMORY]. | |
1796 | */ | |
1797 | for (zt = 0; zt <= zone_last; zt++) | |
1798 | present_pages += pgdat->node_zones[zt].present_pages; | |
1799 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1800 | arg->status_change_nid_normal = zone_to_nid(zone); | |
1801 | else | |
1802 | arg->status_change_nid_normal = -1; | |
1803 | ||
1804 | #ifdef CONFIG_HIGHMEM | |
1805 | /* | |
1806 | * If we have movable node, node_states[N_HIGH_MEMORY] | |
1807 | * contains nodes which have zones of 0...ZONE_HIGHMEM, | |
1808 | * set zone_last to ZONE_HIGHMEM. | |
1809 | * | |
1810 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] | |
1811 | * contains nodes which have zones of 0...ZONE_MOVABLE, | |
1812 | * set zone_last to ZONE_MOVABLE. | |
1813 | */ | |
1814 | zone_last = ZONE_HIGHMEM; | |
1815 | if (N_MEMORY == N_HIGH_MEMORY) | |
1816 | zone_last = ZONE_MOVABLE; | |
1817 | ||
1818 | for (; zt <= zone_last; zt++) | |
1819 | present_pages += pgdat->node_zones[zt].present_pages; | |
1820 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) | |
1821 | arg->status_change_nid_high = zone_to_nid(zone); | |
1822 | else | |
1823 | arg->status_change_nid_high = -1; | |
1824 | #else | |
1825 | arg->status_change_nid_high = arg->status_change_nid_normal; | |
1826 | #endif | |
1827 | ||
1828 | /* | |
1829 | * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE | |
1830 | */ | |
1831 | zone_last = ZONE_MOVABLE; | |
1832 | ||
1833 | /* | |
1834 | * check whether node_states[N_HIGH_MEMORY] will be changed | |
1835 | * If we try to offline the last present @nr_pages from the node, | |
1836 | * we can determind we will need to clear the node from | |
1837 | * node_states[N_HIGH_MEMORY]. | |
1838 | */ | |
1839 | for (; zt <= zone_last; zt++) | |
1840 | present_pages += pgdat->node_zones[zt].present_pages; | |
1841 | if (nr_pages >= present_pages) | |
1842 | arg->status_change_nid = zone_to_nid(zone); | |
1843 | else | |
1844 | arg->status_change_nid = -1; | |
1845 | } | |
1846 | ||
1847 | static void node_states_clear_node(int node, struct memory_notify *arg) | |
1848 | { | |
1849 | if (arg->status_change_nid_normal >= 0) | |
1850 | node_clear_state(node, N_NORMAL_MEMORY); | |
1851 | ||
1852 | if ((N_MEMORY != N_NORMAL_MEMORY) && | |
1853 | (arg->status_change_nid_high >= 0)) | |
1854 | node_clear_state(node, N_HIGH_MEMORY); | |
1855 | ||
1856 | if ((N_MEMORY != N_HIGH_MEMORY) && | |
1857 | (arg->status_change_nid >= 0)) | |
1858 | node_clear_state(node, N_MEMORY); | |
1859 | } | |
1860 | ||
1861 | static int __ref __offline_pages(unsigned long start_pfn, | |
1862 | unsigned long end_pfn, unsigned long timeout) | |
1863 | { | |
1864 | unsigned long pfn, nr_pages, expire; | |
1865 | long offlined_pages; | |
1866 | int ret, drain, retry_max, node; | |
1867 | unsigned long flags; | |
1868 | struct zone *zone; | |
1869 | struct memory_notify arg; | |
1870 | ||
1871 | /* at least, alignment against pageblock is necessary */ | |
1872 | if (!IS_ALIGNED(start_pfn, pageblock_nr_pages)) | |
1873 | return -EINVAL; | |
1874 | if (!IS_ALIGNED(end_pfn, pageblock_nr_pages)) | |
1875 | return -EINVAL; | |
1876 | /* This makes hotplug much easier...and readable. | |
1877 | we assume this for now. .*/ | |
1878 | if (!test_pages_in_a_zone(start_pfn, end_pfn)) | |
1879 | return -EINVAL; | |
1880 | ||
1881 | zone = page_zone(pfn_to_page(start_pfn)); | |
1882 | node = zone_to_nid(zone); | |
1883 | nr_pages = end_pfn - start_pfn; | |
1884 | ||
1885 | if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages)) | |
1886 | return -EINVAL; | |
1887 | ||
1888 | /* set above range as isolated */ | |
1889 | ret = start_isolate_page_range(start_pfn, end_pfn, | |
1890 | MIGRATE_MOVABLE, true); | |
1891 | if (ret) | |
1892 | return ret; | |
1893 | ||
1894 | arg.start_pfn = start_pfn; | |
1895 | arg.nr_pages = nr_pages; | |
1896 | node_states_check_changes_offline(nr_pages, zone, &arg); | |
1897 | ||
1898 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); | |
1899 | ret = notifier_to_errno(ret); | |
1900 | if (ret) | |
1901 | goto failed_removal; | |
1902 | ||
1903 | pfn = start_pfn; | |
1904 | expire = jiffies + timeout; | |
1905 | drain = 0; | |
1906 | retry_max = 5; | |
1907 | repeat: | |
1908 | /* start memory hot removal */ | |
1909 | ret = -EAGAIN; | |
1910 | if (time_after(jiffies, expire)) | |
1911 | goto failed_removal; | |
1912 | ret = -EINTR; | |
1913 | if (signal_pending(current)) | |
1914 | goto failed_removal; | |
1915 | ret = 0; | |
1916 | if (drain) { | |
1917 | lru_add_drain_all(); | |
1918 | cond_resched(); | |
1919 | drain_all_pages(zone); | |
1920 | } | |
1921 | ||
1922 | pfn = scan_movable_pages(start_pfn, end_pfn); | |
1923 | if (pfn) { /* We have movable pages */ | |
1924 | ret = do_migrate_range(pfn, end_pfn); | |
1925 | if (!ret) { | |
1926 | drain = 1; | |
1927 | goto repeat; | |
1928 | } else { | |
1929 | if (ret < 0) | |
1930 | if (--retry_max == 0) | |
1931 | goto failed_removal; | |
1932 | yield(); | |
1933 | drain = 1; | |
1934 | goto repeat; | |
1935 | } | |
1936 | } | |
1937 | /* drain all zone's lru pagevec, this is asynchronous... */ | |
1938 | lru_add_drain_all(); | |
1939 | yield(); | |
1940 | /* drain pcp pages, this is synchronous. */ | |
1941 | drain_all_pages(zone); | |
1942 | /* | |
1943 | * dissolve free hugepages in the memory block before doing offlining | |
1944 | * actually in order to make hugetlbfs's object counting consistent. | |
1945 | */ | |
1946 | dissolve_free_huge_pages(start_pfn, end_pfn); | |
1947 | /* check again */ | |
1948 | offlined_pages = check_pages_isolated(start_pfn, end_pfn); | |
1949 | if (offlined_pages < 0) { | |
1950 | ret = -EBUSY; | |
1951 | goto failed_removal; | |
1952 | } | |
1953 | pr_info("Offlined Pages %ld\n", offlined_pages); | |
1954 | /* Ok, all of our target is isolated. | |
1955 | We cannot do rollback at this point. */ | |
1956 | offline_isolated_pages(start_pfn, end_pfn); | |
1957 | /* reset pagetype flags and makes migrate type to be MOVABLE */ | |
1958 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); | |
1959 | /* removal success */ | |
1960 | adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages); | |
1961 | zone->present_pages -= offlined_pages; | |
1962 | ||
1963 | pgdat_resize_lock(zone->zone_pgdat, &flags); | |
1964 | zone->zone_pgdat->node_present_pages -= offlined_pages; | |
1965 | pgdat_resize_unlock(zone->zone_pgdat, &flags); | |
1966 | ||
1967 | init_per_zone_wmark_min(); | |
1968 | ||
1969 | if (!populated_zone(zone)) { | |
1970 | zone_pcp_reset(zone); | |
1971 | mutex_lock(&zonelists_mutex); | |
1972 | build_all_zonelists(NULL, NULL); | |
1973 | mutex_unlock(&zonelists_mutex); | |
1974 | } else | |
1975 | zone_pcp_update(zone); | |
1976 | ||
1977 | node_states_clear_node(node, &arg); | |
1978 | if (arg.status_change_nid >= 0) { | |
1979 | kswapd_stop(node); | |
1980 | kcompactd_stop(node); | |
1981 | } | |
1982 | ||
1983 | vm_total_pages = nr_free_pagecache_pages(); | |
1984 | writeback_set_ratelimit(); | |
1985 | ||
1986 | memory_notify(MEM_OFFLINE, &arg); | |
1987 | return 0; | |
1988 | ||
1989 | failed_removal: | |
1990 | pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n", | |
1991 | (unsigned long long) start_pfn << PAGE_SHIFT, | |
1992 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1); | |
1993 | memory_notify(MEM_CANCEL_OFFLINE, &arg); | |
1994 | /* pushback to free area */ | |
1995 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); | |
1996 | return ret; | |
1997 | } | |
1998 | ||
1999 | /* Must be protected by mem_hotplug_begin() */ | |
2000 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) | |
2001 | { | |
2002 | return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ); | |
2003 | } | |
2004 | #endif /* CONFIG_MEMORY_HOTREMOVE */ | |
2005 | ||
2006 | /** | |
2007 | * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn) | |
2008 | * @start_pfn: start pfn of the memory range | |
2009 | * @end_pfn: end pfn of the memory range | |
2010 | * @arg: argument passed to func | |
2011 | * @func: callback for each memory section walked | |
2012 | * | |
2013 | * This function walks through all present mem sections in range | |
2014 | * [start_pfn, end_pfn) and call func on each mem section. | |
2015 | * | |
2016 | * Returns the return value of func. | |
2017 | */ | |
2018 | int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn, | |
2019 | void *arg, int (*func)(struct memory_block *, void *)) | |
2020 | { | |
2021 | struct memory_block *mem = NULL; | |
2022 | struct mem_section *section; | |
2023 | unsigned long pfn, section_nr; | |
2024 | int ret; | |
2025 | ||
2026 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
2027 | section_nr = pfn_to_section_nr(pfn); | |
2028 | if (!present_section_nr(section_nr)) | |
2029 | continue; | |
2030 | ||
2031 | section = __nr_to_section(section_nr); | |
2032 | /* same memblock? */ | |
2033 | if (mem) | |
2034 | if ((section_nr >= mem->start_section_nr) && | |
2035 | (section_nr <= mem->end_section_nr)) | |
2036 | continue; | |
2037 | ||
2038 | mem = find_memory_block_hinted(section, mem); | |
2039 | if (!mem) | |
2040 | continue; | |
2041 | ||
2042 | ret = func(mem, arg); | |
2043 | if (ret) { | |
2044 | kobject_put(&mem->dev.kobj); | |
2045 | return ret; | |
2046 | } | |
2047 | } | |
2048 | ||
2049 | if (mem) | |
2050 | kobject_put(&mem->dev.kobj); | |
2051 | ||
2052 | return 0; | |
2053 | } | |
2054 | ||
2055 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
2056 | static int check_memblock_offlined_cb(struct memory_block *mem, void *arg) | |
2057 | { | |
2058 | int ret = !is_memblock_offlined(mem); | |
2059 | ||
2060 | if (unlikely(ret)) { | |
2061 | phys_addr_t beginpa, endpa; | |
2062 | ||
2063 | beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)); | |
2064 | endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1; | |
2065 | pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n", | |
2066 | &beginpa, &endpa); | |
2067 | } | |
2068 | ||
2069 | return ret; | |
2070 | } | |
2071 | ||
2072 | static int check_cpu_on_node(pg_data_t *pgdat) | |
2073 | { | |
2074 | int cpu; | |
2075 | ||
2076 | for_each_present_cpu(cpu) { | |
2077 | if (cpu_to_node(cpu) == pgdat->node_id) | |
2078 | /* | |
2079 | * the cpu on this node isn't removed, and we can't | |
2080 | * offline this node. | |
2081 | */ | |
2082 | return -EBUSY; | |
2083 | } | |
2084 | ||
2085 | return 0; | |
2086 | } | |
2087 | ||
2088 | static void unmap_cpu_on_node(pg_data_t *pgdat) | |
2089 | { | |
2090 | #ifdef CONFIG_ACPI_NUMA | |
2091 | int cpu; | |
2092 | ||
2093 | for_each_possible_cpu(cpu) | |
2094 | if (cpu_to_node(cpu) == pgdat->node_id) | |
2095 | numa_clear_node(cpu); | |
2096 | #endif | |
2097 | } | |
2098 | ||
2099 | static int check_and_unmap_cpu_on_node(pg_data_t *pgdat) | |
2100 | { | |
2101 | int ret; | |
2102 | ||
2103 | ret = check_cpu_on_node(pgdat); | |
2104 | if (ret) | |
2105 | return ret; | |
2106 | ||
2107 | /* | |
2108 | * the node will be offlined when we come here, so we can clear | |
2109 | * the cpu_to_node() now. | |
2110 | */ | |
2111 | ||
2112 | unmap_cpu_on_node(pgdat); | |
2113 | return 0; | |
2114 | } | |
2115 | ||
2116 | /** | |
2117 | * try_offline_node | |
2118 | * | |
2119 | * Offline a node if all memory sections and cpus of the node are removed. | |
2120 | * | |
2121 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
2122 | * and online/offline operations before this call. | |
2123 | */ | |
2124 | void try_offline_node(int nid) | |
2125 | { | |
2126 | pg_data_t *pgdat = NODE_DATA(nid); | |
2127 | unsigned long start_pfn = pgdat->node_start_pfn; | |
2128 | unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages; | |
2129 | unsigned long pfn; | |
2130 | int i; | |
2131 | ||
2132 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
2133 | unsigned long section_nr = pfn_to_section_nr(pfn); | |
2134 | ||
2135 | if (!present_section_nr(section_nr)) | |
2136 | continue; | |
2137 | ||
2138 | if (pfn_to_nid(pfn) != nid) | |
2139 | continue; | |
2140 | ||
2141 | /* | |
2142 | * some memory sections of this node are not removed, and we | |
2143 | * can't offline node now. | |
2144 | */ | |
2145 | return; | |
2146 | } | |
2147 | ||
2148 | if (check_and_unmap_cpu_on_node(pgdat)) | |
2149 | return; | |
2150 | ||
2151 | /* | |
2152 | * all memory/cpu of this node are removed, we can offline this | |
2153 | * node now. | |
2154 | */ | |
2155 | node_set_offline(nid); | |
2156 | unregister_one_node(nid); | |
2157 | ||
2158 | /* free waittable in each zone */ | |
2159 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2160 | struct zone *zone = pgdat->node_zones + i; | |
2161 | ||
2162 | /* | |
2163 | * wait_table may be allocated from boot memory, | |
2164 | * here only free if it's allocated by vmalloc. | |
2165 | */ | |
2166 | if (is_vmalloc_addr(zone->wait_table)) { | |
2167 | vfree(zone->wait_table); | |
2168 | zone->wait_table = NULL; | |
2169 | } | |
2170 | } | |
2171 | } | |
2172 | EXPORT_SYMBOL(try_offline_node); | |
2173 | ||
2174 | /** | |
2175 | * remove_memory | |
2176 | * | |
2177 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug | |
2178 | * and online/offline operations before this call, as required by | |
2179 | * try_offline_node(). | |
2180 | */ | |
2181 | void __ref remove_memory(int nid, u64 start, u64 size) | |
2182 | { | |
2183 | int ret; | |
2184 | ||
2185 | BUG_ON(check_hotplug_memory_range(start, size)); | |
2186 | ||
2187 | mem_hotplug_begin(); | |
2188 | ||
2189 | /* | |
2190 | * All memory blocks must be offlined before removing memory. Check | |
2191 | * whether all memory blocks in question are offline and trigger a BUG() | |
2192 | * if this is not the case. | |
2193 | */ | |
2194 | ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL, | |
2195 | check_memblock_offlined_cb); | |
2196 | if (ret) | |
2197 | BUG(); | |
2198 | ||
2199 | /* remove memmap entry */ | |
2200 | firmware_map_remove(start, start + size, "System RAM"); | |
2201 | memblock_free(start, size); | |
2202 | memblock_remove(start, size); | |
2203 | ||
2204 | arch_remove_memory(start, size); | |
2205 | ||
2206 | try_offline_node(nid); | |
2207 | ||
2208 | mem_hotplug_done(); | |
2209 | } | |
2210 | EXPORT_SYMBOL_GPL(remove_memory); | |
2211 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |