blob: b4c8d7b9ab820f61cf5d5c6080119cb90d2c9d94
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 | #include <linux/rmap.h> |
38 | |
39 | #include <asm/tlbflush.h> |
40 | |
41 | #include "internal.h" |
42 | |
43 | /* |
44 | * online_page_callback contains pointer to current page onlining function. |
45 | * Initially it is generic_online_page(). If it is required it could be |
46 | * changed by calling set_online_page_callback() for callback registration |
47 | * and restore_online_page_callback() for generic callback restore. |
48 | */ |
49 | |
50 | static void generic_online_page(struct page *page); |
51 | |
52 | static online_page_callback_t online_page_callback = generic_online_page; |
53 | static DEFINE_MUTEX(online_page_callback_lock); |
54 | |
55 | /* The same as the cpu_hotplug lock, but for memory hotplug. */ |
56 | static struct { |
57 | struct task_struct *active_writer; |
58 | struct mutex lock; /* Synchronizes accesses to refcount, */ |
59 | /* |
60 | * Also blocks the new readers during |
61 | * an ongoing mem hotplug operation. |
62 | */ |
63 | int refcount; |
64 | |
65 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
66 | struct lockdep_map dep_map; |
67 | #endif |
68 | } mem_hotplug = { |
69 | .active_writer = NULL, |
70 | .lock = __MUTEX_INITIALIZER(mem_hotplug.lock), |
71 | .refcount = 0, |
72 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
73 | .dep_map = {.name = "mem_hotplug.lock" }, |
74 | #endif |
75 | }; |
76 | |
77 | /* Lockdep annotations for get/put_online_mems() and mem_hotplug_begin/end() */ |
78 | #define memhp_lock_acquire_read() lock_map_acquire_read(&mem_hotplug.dep_map) |
79 | #define memhp_lock_acquire() lock_map_acquire(&mem_hotplug.dep_map) |
80 | #define memhp_lock_release() lock_map_release(&mem_hotplug.dep_map) |
81 | |
82 | #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE |
83 | bool memhp_auto_online; |
84 | #else |
85 | bool memhp_auto_online = true; |
86 | #endif |
87 | EXPORT_SYMBOL_GPL(memhp_auto_online); |
88 | |
89 | static int __init setup_memhp_default_state(char *str) |
90 | { |
91 | if (!strcmp(str, "online")) |
92 | memhp_auto_online = true; |
93 | else if (!strcmp(str, "offline")) |
94 | memhp_auto_online = false; |
95 | |
96 | return 1; |
97 | } |
98 | __setup("memhp_default_state=", setup_memhp_default_state); |
99 | |
100 | void get_online_mems(void) |
101 | { |
102 | might_sleep(); |
103 | if (mem_hotplug.active_writer == current) |
104 | return; |
105 | memhp_lock_acquire_read(); |
106 | mutex_lock(&mem_hotplug.lock); |
107 | mem_hotplug.refcount++; |
108 | mutex_unlock(&mem_hotplug.lock); |
109 | |
110 | } |
111 | |
112 | void put_online_mems(void) |
113 | { |
114 | if (mem_hotplug.active_writer == current) |
115 | return; |
116 | mutex_lock(&mem_hotplug.lock); |
117 | |
118 | if (WARN_ON(!mem_hotplug.refcount)) |
119 | mem_hotplug.refcount++; /* try to fix things up */ |
120 | |
121 | if (!--mem_hotplug.refcount && unlikely(mem_hotplug.active_writer)) |
122 | wake_up_process(mem_hotplug.active_writer); |
123 | mutex_unlock(&mem_hotplug.lock); |
124 | memhp_lock_release(); |
125 | |
126 | } |
127 | |
128 | void mem_hotplug_begin(void) |
129 | { |
130 | mem_hotplug.active_writer = current; |
131 | |
132 | memhp_lock_acquire(); |
133 | for (;;) { |
134 | mutex_lock(&mem_hotplug.lock); |
135 | if (likely(!mem_hotplug.refcount)) |
136 | break; |
137 | __set_current_state(TASK_UNINTERRUPTIBLE); |
138 | mutex_unlock(&mem_hotplug.lock); |
139 | schedule(); |
140 | } |
141 | } |
142 | |
143 | void mem_hotplug_done(void) |
144 | { |
145 | mem_hotplug.active_writer = NULL; |
146 | mutex_unlock(&mem_hotplug.lock); |
147 | memhp_lock_release(); |
148 | } |
149 | |
150 | /* add this memory to iomem resource */ |
151 | static struct resource *register_memory_resource(u64 start, u64 size) |
152 | { |
153 | struct resource *res; |
154 | res = kzalloc(sizeof(struct resource), GFP_KERNEL); |
155 | if (!res) |
156 | return ERR_PTR(-ENOMEM); |
157 | |
158 | res->name = "System RAM"; |
159 | res->start = start; |
160 | res->end = start + size - 1; |
161 | res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY; |
162 | if (request_resource(&iomem_resource, res) < 0) { |
163 | pr_debug("System RAM resource %pR cannot be added\n", res); |
164 | kfree(res); |
165 | return ERR_PTR(-EEXIST); |
166 | } |
167 | return res; |
168 | } |
169 | |
170 | static void release_memory_resource(struct resource *res) |
171 | { |
172 | if (!res) |
173 | return; |
174 | release_resource(res); |
175 | kfree(res); |
176 | return; |
177 | } |
178 | |
179 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
180 | void get_page_bootmem(unsigned long info, struct page *page, |
181 | unsigned long type) |
182 | { |
183 | page->freelist = (void *)type; |
184 | SetPagePrivate(page); |
185 | set_page_private(page, info); |
186 | page_ref_inc(page); |
187 | } |
188 | |
189 | void put_page_bootmem(struct page *page) |
190 | { |
191 | unsigned long type; |
192 | |
193 | type = (unsigned long) page->freelist; |
194 | BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || |
195 | type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); |
196 | |
197 | if (page_ref_dec_return(page) == 1) { |
198 | page->freelist = NULL; |
199 | ClearPagePrivate(page); |
200 | set_page_private(page, 0); |
201 | INIT_LIST_HEAD(&page->lru); |
202 | free_reserved_page(page); |
203 | } |
204 | } |
205 | |
206 | #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE |
207 | #ifndef CONFIG_SPARSEMEM_VMEMMAP |
208 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
209 | { |
210 | unsigned long *usemap, mapsize, section_nr, i; |
211 | struct mem_section *ms; |
212 | struct page *page, *memmap; |
213 | |
214 | section_nr = pfn_to_section_nr(start_pfn); |
215 | ms = __nr_to_section(section_nr); |
216 | |
217 | /* Get section's memmap address */ |
218 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); |
219 | |
220 | /* |
221 | * Get page for the memmap's phys address |
222 | * XXX: need more consideration for sparse_vmemmap... |
223 | */ |
224 | page = virt_to_page(memmap); |
225 | mapsize = sizeof(struct page) * PAGES_PER_SECTION; |
226 | mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; |
227 | |
228 | /* remember memmap's page */ |
229 | for (i = 0; i < mapsize; i++, page++) |
230 | get_page_bootmem(section_nr, page, SECTION_INFO); |
231 | |
232 | usemap = __nr_to_section(section_nr)->pageblock_flags; |
233 | page = virt_to_page(usemap); |
234 | |
235 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; |
236 | |
237 | for (i = 0; i < mapsize; i++, page++) |
238 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
239 | |
240 | } |
241 | #else /* CONFIG_SPARSEMEM_VMEMMAP */ |
242 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
243 | { |
244 | unsigned long *usemap, mapsize, section_nr, i; |
245 | struct mem_section *ms; |
246 | struct page *page, *memmap; |
247 | |
248 | if (!pfn_valid(start_pfn)) |
249 | return; |
250 | |
251 | section_nr = pfn_to_section_nr(start_pfn); |
252 | ms = __nr_to_section(section_nr); |
253 | |
254 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); |
255 | |
256 | register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION); |
257 | |
258 | usemap = __nr_to_section(section_nr)->pageblock_flags; |
259 | page = virt_to_page(usemap); |
260 | |
261 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; |
262 | |
263 | for (i = 0; i < mapsize; i++, page++) |
264 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
265 | } |
266 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ |
267 | |
268 | void __init register_page_bootmem_info_node(struct pglist_data *pgdat) |
269 | { |
270 | unsigned long i, pfn, end_pfn, nr_pages; |
271 | int node = pgdat->node_id; |
272 | struct page *page; |
273 | |
274 | nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; |
275 | page = virt_to_page(pgdat); |
276 | |
277 | for (i = 0; i < nr_pages; i++, page++) |
278 | get_page_bootmem(node, page, NODE_INFO); |
279 | |
280 | pfn = pgdat->node_start_pfn; |
281 | end_pfn = pgdat_end_pfn(pgdat); |
282 | |
283 | /* register section info */ |
284 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
285 | /* |
286 | * Some platforms can assign the same pfn to multiple nodes - on |
287 | * node0 as well as nodeN. To avoid registering a pfn against |
288 | * multiple nodes we check that this pfn does not already |
289 | * reside in some other nodes. |
290 | */ |
291 | if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node)) |
292 | register_page_bootmem_info_section(pfn); |
293 | } |
294 | } |
295 | #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */ |
296 | |
297 | static void __meminit grow_zone_span(struct zone *zone, unsigned long start_pfn, |
298 | unsigned long end_pfn) |
299 | { |
300 | unsigned long old_zone_end_pfn; |
301 | |
302 | zone_span_writelock(zone); |
303 | |
304 | old_zone_end_pfn = zone_end_pfn(zone); |
305 | if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn) |
306 | zone->zone_start_pfn = start_pfn; |
307 | |
308 | zone->spanned_pages = max(old_zone_end_pfn, end_pfn) - |
309 | zone->zone_start_pfn; |
310 | |
311 | zone_span_writeunlock(zone); |
312 | } |
313 | |
314 | static void resize_zone(struct zone *zone, unsigned long start_pfn, |
315 | unsigned long end_pfn) |
316 | { |
317 | zone_span_writelock(zone); |
318 | |
319 | if (end_pfn - start_pfn) { |
320 | zone->zone_start_pfn = start_pfn; |
321 | zone->spanned_pages = end_pfn - start_pfn; |
322 | } else { |
323 | /* |
324 | * make it consist as free_area_init_core(), |
325 | * if spanned_pages = 0, then keep start_pfn = 0 |
326 | */ |
327 | zone->zone_start_pfn = 0; |
328 | zone->spanned_pages = 0; |
329 | } |
330 | |
331 | zone_span_writeunlock(zone); |
332 | } |
333 | |
334 | static void fix_zone_id(struct zone *zone, unsigned long start_pfn, |
335 | unsigned long end_pfn) |
336 | { |
337 | enum zone_type zid = zone_idx(zone); |
338 | int nid = zone->zone_pgdat->node_id; |
339 | unsigned long pfn; |
340 | |
341 | for (pfn = start_pfn; pfn < end_pfn; pfn++) |
342 | set_page_links(pfn_to_page(pfn), zid, nid, pfn); |
343 | } |
344 | |
345 | /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or |
346 | * alloc_bootmem_node_nopanic()/memblock_virt_alloc_node_nopanic() */ |
347 | static int __ref ensure_zone_is_initialized(struct zone *zone, |
348 | unsigned long start_pfn, unsigned long num_pages) |
349 | { |
350 | if (!zone_is_initialized(zone)) |
351 | return init_currently_empty_zone(zone, start_pfn, num_pages); |
352 | |
353 | return 0; |
354 | } |
355 | |
356 | static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2, |
357 | unsigned long start_pfn, unsigned long end_pfn) |
358 | { |
359 | int ret; |
360 | unsigned long flags; |
361 | unsigned long z1_start_pfn; |
362 | |
363 | ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn); |
364 | if (ret) |
365 | return ret; |
366 | |
367 | pgdat_resize_lock(z1->zone_pgdat, &flags); |
368 | |
369 | /* can't move pfns which are higher than @z2 */ |
370 | if (end_pfn > zone_end_pfn(z2)) |
371 | goto out_fail; |
372 | /* the move out part must be at the left most of @z2 */ |
373 | if (start_pfn > z2->zone_start_pfn) |
374 | goto out_fail; |
375 | /* must included/overlap */ |
376 | if (end_pfn <= z2->zone_start_pfn) |
377 | goto out_fail; |
378 | |
379 | /* use start_pfn for z1's start_pfn if z1 is empty */ |
380 | if (!zone_is_empty(z1)) |
381 | z1_start_pfn = z1->zone_start_pfn; |
382 | else |
383 | z1_start_pfn = start_pfn; |
384 | |
385 | resize_zone(z1, z1_start_pfn, end_pfn); |
386 | resize_zone(z2, end_pfn, zone_end_pfn(z2)); |
387 | |
388 | pgdat_resize_unlock(z1->zone_pgdat, &flags); |
389 | |
390 | fix_zone_id(z1, start_pfn, end_pfn); |
391 | |
392 | return 0; |
393 | out_fail: |
394 | pgdat_resize_unlock(z1->zone_pgdat, &flags); |
395 | return -1; |
396 | } |
397 | |
398 | static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2, |
399 | unsigned long start_pfn, unsigned long end_pfn) |
400 | { |
401 | int ret; |
402 | unsigned long flags; |
403 | unsigned long z2_end_pfn; |
404 | |
405 | ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn); |
406 | if (ret) |
407 | return ret; |
408 | |
409 | pgdat_resize_lock(z1->zone_pgdat, &flags); |
410 | |
411 | /* can't move pfns which are lower than @z1 */ |
412 | if (z1->zone_start_pfn > start_pfn) |
413 | goto out_fail; |
414 | /* the move out part mast at the right most of @z1 */ |
415 | if (zone_end_pfn(z1) > end_pfn) |
416 | goto out_fail; |
417 | /* must included/overlap */ |
418 | if (start_pfn >= zone_end_pfn(z1)) |
419 | goto out_fail; |
420 | |
421 | /* use end_pfn for z2's end_pfn if z2 is empty */ |
422 | if (!zone_is_empty(z2)) |
423 | z2_end_pfn = zone_end_pfn(z2); |
424 | else |
425 | z2_end_pfn = end_pfn; |
426 | |
427 | resize_zone(z1, z1->zone_start_pfn, start_pfn); |
428 | resize_zone(z2, start_pfn, z2_end_pfn); |
429 | |
430 | pgdat_resize_unlock(z1->zone_pgdat, &flags); |
431 | |
432 | fix_zone_id(z2, start_pfn, end_pfn); |
433 | |
434 | return 0; |
435 | out_fail: |
436 | pgdat_resize_unlock(z1->zone_pgdat, &flags); |
437 | return -1; |
438 | } |
439 | |
440 | static struct zone * __meminit move_pfn_range(int zone_shift, |
441 | unsigned long start_pfn, unsigned long end_pfn) |
442 | { |
443 | struct zone *zone = page_zone(pfn_to_page(start_pfn)); |
444 | int ret = 0; |
445 | |
446 | if (zone_shift < 0) |
447 | ret = move_pfn_range_left(zone + zone_shift, zone, |
448 | start_pfn, end_pfn); |
449 | else if (zone_shift) |
450 | ret = move_pfn_range_right(zone, zone + zone_shift, |
451 | start_pfn, end_pfn); |
452 | |
453 | if (ret) |
454 | return NULL; |
455 | |
456 | return zone + zone_shift; |
457 | } |
458 | |
459 | static void __meminit grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn, |
460 | unsigned long end_pfn) |
461 | { |
462 | unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat); |
463 | |
464 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) |
465 | pgdat->node_start_pfn = start_pfn; |
466 | |
467 | pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) - |
468 | pgdat->node_start_pfn; |
469 | } |
470 | |
471 | static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn) |
472 | { |
473 | struct pglist_data *pgdat = zone->zone_pgdat; |
474 | int nr_pages = PAGES_PER_SECTION; |
475 | int nid = pgdat->node_id; |
476 | int zone_type; |
477 | unsigned long flags, pfn; |
478 | int ret; |
479 | |
480 | zone_type = zone - pgdat->node_zones; |
481 | ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages); |
482 | if (ret) |
483 | return ret; |
484 | |
485 | pgdat_resize_lock(zone->zone_pgdat, &flags); |
486 | grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages); |
487 | grow_pgdat_span(zone->zone_pgdat, phys_start_pfn, |
488 | phys_start_pfn + nr_pages); |
489 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
490 | memmap_init_zone(nr_pages, nid, zone_type, |
491 | phys_start_pfn, MEMMAP_HOTPLUG); |
492 | |
493 | /* online_page_range is called later and expects pages reserved */ |
494 | for (pfn = phys_start_pfn; pfn < phys_start_pfn + nr_pages; pfn++) { |
495 | if (!pfn_valid(pfn)) |
496 | continue; |
497 | |
498 | SetPageReserved(pfn_to_page(pfn)); |
499 | } |
500 | return 0; |
501 | } |
502 | |
503 | static int __meminit __add_section(int nid, struct zone *zone, |
504 | unsigned long phys_start_pfn) |
505 | { |
506 | int ret; |
507 | |
508 | if (pfn_valid(phys_start_pfn)) |
509 | return -EEXIST; |
510 | |
511 | ret = sparse_add_one_section(zone, phys_start_pfn); |
512 | |
513 | if (ret < 0) |
514 | return ret; |
515 | |
516 | ret = __add_zone(zone, phys_start_pfn); |
517 | |
518 | if (ret < 0) |
519 | return ret; |
520 | |
521 | return register_new_memory(nid, __pfn_to_section(phys_start_pfn)); |
522 | } |
523 | |
524 | /* |
525 | * Reasonably generic function for adding memory. It is |
526 | * expected that archs that support memory hotplug will |
527 | * call this function after deciding the zone to which to |
528 | * add the new pages. |
529 | */ |
530 | int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn, |
531 | unsigned long nr_pages) |
532 | { |
533 | unsigned long i; |
534 | int err = 0; |
535 | int start_sec, end_sec; |
536 | struct vmem_altmap *altmap; |
537 | |
538 | clear_zone_contiguous(zone); |
539 | |
540 | /* during initialize mem_map, align hot-added range to section */ |
541 | start_sec = pfn_to_section_nr(phys_start_pfn); |
542 | end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1); |
543 | |
544 | altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn)); |
545 | if (altmap) { |
546 | /* |
547 | * Validate altmap is within bounds of the total request |
548 | */ |
549 | if (altmap->base_pfn != phys_start_pfn |
550 | || vmem_altmap_offset(altmap) > nr_pages) { |
551 | pr_warn_once("memory add fail, invalid altmap\n"); |
552 | err = -EINVAL; |
553 | goto out; |
554 | } |
555 | altmap->alloc = 0; |
556 | } |
557 | |
558 | for (i = start_sec; i <= end_sec; i++) { |
559 | err = __add_section(nid, zone, section_nr_to_pfn(i)); |
560 | |
561 | /* |
562 | * EEXIST is finally dealt with by ioresource collision |
563 | * check. see add_memory() => register_memory_resource() |
564 | * Warning will be printed if there is collision. |
565 | */ |
566 | if (err && (err != -EEXIST)) |
567 | break; |
568 | err = 0; |
569 | } |
570 | vmemmap_populate_print_last(); |
571 | out: |
572 | set_zone_contiguous(zone); |
573 | return err; |
574 | } |
575 | EXPORT_SYMBOL_GPL(__add_pages); |
576 | |
577 | #ifdef CONFIG_MEMORY_HOTREMOVE |
578 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ |
579 | static int find_smallest_section_pfn(int nid, struct zone *zone, |
580 | unsigned long start_pfn, |
581 | unsigned long end_pfn) |
582 | { |
583 | struct mem_section *ms; |
584 | |
585 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) { |
586 | ms = __pfn_to_section(start_pfn); |
587 | |
588 | if (unlikely(!valid_section(ms))) |
589 | continue; |
590 | |
591 | if (unlikely(pfn_to_nid(start_pfn) != nid)) |
592 | continue; |
593 | |
594 | if (zone && zone != page_zone(pfn_to_page(start_pfn))) |
595 | continue; |
596 | |
597 | return start_pfn; |
598 | } |
599 | |
600 | return 0; |
601 | } |
602 | |
603 | /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ |
604 | static int find_biggest_section_pfn(int nid, struct zone *zone, |
605 | unsigned long start_pfn, |
606 | unsigned long end_pfn) |
607 | { |
608 | struct mem_section *ms; |
609 | unsigned long pfn; |
610 | |
611 | /* pfn is the end pfn of a memory section. */ |
612 | pfn = end_pfn - 1; |
613 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) { |
614 | ms = __pfn_to_section(pfn); |
615 | |
616 | if (unlikely(!valid_section(ms))) |
617 | continue; |
618 | |
619 | if (unlikely(pfn_to_nid(pfn) != nid)) |
620 | continue; |
621 | |
622 | if (zone && zone != page_zone(pfn_to_page(pfn))) |
623 | continue; |
624 | |
625 | return pfn; |
626 | } |
627 | |
628 | return 0; |
629 | } |
630 | |
631 | static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, |
632 | unsigned long end_pfn) |
633 | { |
634 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
635 | unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */ |
636 | unsigned long zone_end_pfn = z; |
637 | unsigned long pfn; |
638 | struct mem_section *ms; |
639 | int nid = zone_to_nid(zone); |
640 | |
641 | zone_span_writelock(zone); |
642 | if (zone_start_pfn == start_pfn) { |
643 | /* |
644 | * If the section is smallest section in the zone, it need |
645 | * shrink zone->zone_start_pfn and zone->zone_spanned_pages. |
646 | * In this case, we find second smallest valid mem_section |
647 | * for shrinking zone. |
648 | */ |
649 | pfn = find_smallest_section_pfn(nid, zone, end_pfn, |
650 | zone_end_pfn); |
651 | if (pfn) { |
652 | zone->zone_start_pfn = pfn; |
653 | zone->spanned_pages = zone_end_pfn - pfn; |
654 | } |
655 | } else if (zone_end_pfn == end_pfn) { |
656 | /* |
657 | * If the section is biggest section in the zone, it need |
658 | * shrink zone->spanned_pages. |
659 | * In this case, we find second biggest valid mem_section for |
660 | * shrinking zone. |
661 | */ |
662 | pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn, |
663 | start_pfn); |
664 | if (pfn) |
665 | zone->spanned_pages = pfn - zone_start_pfn + 1; |
666 | } |
667 | |
668 | /* |
669 | * The section is not biggest or smallest mem_section in the zone, it |
670 | * only creates a hole in the zone. So in this case, we need not |
671 | * change the zone. But perhaps, the zone has only hole data. Thus |
672 | * it check the zone has only hole or not. |
673 | */ |
674 | pfn = zone_start_pfn; |
675 | for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) { |
676 | ms = __pfn_to_section(pfn); |
677 | |
678 | if (unlikely(!valid_section(ms))) |
679 | continue; |
680 | |
681 | if (page_zone(pfn_to_page(pfn)) != zone) |
682 | continue; |
683 | |
684 | /* If the section is current section, it continues the loop */ |
685 | if (start_pfn == pfn) |
686 | continue; |
687 | |
688 | /* If we find valid section, we have nothing to do */ |
689 | zone_span_writeunlock(zone); |
690 | return; |
691 | } |
692 | |
693 | /* The zone has no valid section */ |
694 | zone->zone_start_pfn = 0; |
695 | zone->spanned_pages = 0; |
696 | zone_span_writeunlock(zone); |
697 | } |
698 | |
699 | static void shrink_pgdat_span(struct pglist_data *pgdat, |
700 | unsigned long start_pfn, unsigned long end_pfn) |
701 | { |
702 | unsigned long pgdat_start_pfn = pgdat->node_start_pfn; |
703 | unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */ |
704 | unsigned long pgdat_end_pfn = p; |
705 | unsigned long pfn; |
706 | struct mem_section *ms; |
707 | int nid = pgdat->node_id; |
708 | |
709 | if (pgdat_start_pfn == start_pfn) { |
710 | /* |
711 | * If the section is smallest section in the pgdat, it need |
712 | * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages. |
713 | * In this case, we find second smallest valid mem_section |
714 | * for shrinking zone. |
715 | */ |
716 | pfn = find_smallest_section_pfn(nid, NULL, end_pfn, |
717 | pgdat_end_pfn); |
718 | if (pfn) { |
719 | pgdat->node_start_pfn = pfn; |
720 | pgdat->node_spanned_pages = pgdat_end_pfn - pfn; |
721 | } |
722 | } else if (pgdat_end_pfn == end_pfn) { |
723 | /* |
724 | * If the section is biggest section in the pgdat, it need |
725 | * shrink pgdat->node_spanned_pages. |
726 | * In this case, we find second biggest valid mem_section for |
727 | * shrinking zone. |
728 | */ |
729 | pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn, |
730 | start_pfn); |
731 | if (pfn) |
732 | pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1; |
733 | } |
734 | |
735 | /* |
736 | * If the section is not biggest or smallest mem_section in the pgdat, |
737 | * it only creates a hole in the pgdat. So in this case, we need not |
738 | * change the pgdat. |
739 | * But perhaps, the pgdat has only hole data. Thus it check the pgdat |
740 | * has only hole or not. |
741 | */ |
742 | pfn = pgdat_start_pfn; |
743 | for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) { |
744 | ms = __pfn_to_section(pfn); |
745 | |
746 | if (unlikely(!valid_section(ms))) |
747 | continue; |
748 | |
749 | if (pfn_to_nid(pfn) != nid) |
750 | continue; |
751 | |
752 | /* If the section is current section, it continues the loop */ |
753 | if (start_pfn == pfn) |
754 | continue; |
755 | |
756 | /* If we find valid section, we have nothing to do */ |
757 | return; |
758 | } |
759 | |
760 | /* The pgdat has no valid section */ |
761 | pgdat->node_start_pfn = 0; |
762 | pgdat->node_spanned_pages = 0; |
763 | } |
764 | |
765 | static void __remove_zone(struct zone *zone, unsigned long start_pfn) |
766 | { |
767 | struct pglist_data *pgdat = zone->zone_pgdat; |
768 | int nr_pages = PAGES_PER_SECTION; |
769 | int zone_type; |
770 | unsigned long flags; |
771 | |
772 | zone_type = zone - pgdat->node_zones; |
773 | |
774 | pgdat_resize_lock(zone->zone_pgdat, &flags); |
775 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); |
776 | shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages); |
777 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
778 | } |
779 | |
780 | static int __remove_section(struct zone *zone, struct mem_section *ms, |
781 | unsigned long map_offset) |
782 | { |
783 | unsigned long start_pfn; |
784 | int scn_nr; |
785 | int ret = -EINVAL; |
786 | |
787 | if (!valid_section(ms)) |
788 | return ret; |
789 | |
790 | ret = unregister_memory_section(ms); |
791 | if (ret) |
792 | return ret; |
793 | |
794 | scn_nr = __section_nr(ms); |
795 | start_pfn = section_nr_to_pfn(scn_nr); |
796 | __remove_zone(zone, start_pfn); |
797 | |
798 | sparse_remove_one_section(zone, ms, map_offset); |
799 | return 0; |
800 | } |
801 | |
802 | /** |
803 | * __remove_pages() - remove sections of pages from a zone |
804 | * @zone: zone from which pages need to be removed |
805 | * @phys_start_pfn: starting pageframe (must be aligned to start of a section) |
806 | * @nr_pages: number of pages to remove (must be multiple of section size) |
807 | * |
808 | * Generic helper function to remove section mappings and sysfs entries |
809 | * for the section of the memory we are removing. Caller needs to make |
810 | * sure that pages are marked reserved and zones are adjust properly by |
811 | * calling offline_pages(). |
812 | */ |
813 | int __remove_pages(struct zone *zone, unsigned long phys_start_pfn, |
814 | unsigned long nr_pages) |
815 | { |
816 | unsigned long i; |
817 | unsigned long map_offset = 0; |
818 | int sections_to_remove, ret = 0; |
819 | |
820 | /* In the ZONE_DEVICE case device driver owns the memory region */ |
821 | if (is_dev_zone(zone)) { |
822 | struct page *page = pfn_to_page(phys_start_pfn); |
823 | struct vmem_altmap *altmap; |
824 | |
825 | altmap = to_vmem_altmap((unsigned long) page); |
826 | if (altmap) |
827 | map_offset = vmem_altmap_offset(altmap); |
828 | } else { |
829 | resource_size_t start, size; |
830 | |
831 | start = phys_start_pfn << PAGE_SHIFT; |
832 | size = nr_pages * PAGE_SIZE; |
833 | |
834 | ret = release_mem_region_adjustable(&iomem_resource, start, |
835 | size); |
836 | if (ret) { |
837 | resource_size_t endres = start + size - 1; |
838 | |
839 | pr_warn("Unable to release resource <%pa-%pa> (%d)\n", |
840 | &start, &endres, ret); |
841 | } |
842 | } |
843 | |
844 | clear_zone_contiguous(zone); |
845 | |
846 | /* |
847 | * We can only remove entire sections |
848 | */ |
849 | BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK); |
850 | BUG_ON(nr_pages % PAGES_PER_SECTION); |
851 | |
852 | sections_to_remove = nr_pages / PAGES_PER_SECTION; |
853 | for (i = 0; i < sections_to_remove; i++) { |
854 | unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; |
855 | |
856 | ret = __remove_section(zone, __pfn_to_section(pfn), map_offset); |
857 | map_offset = 0; |
858 | if (ret) |
859 | break; |
860 | } |
861 | |
862 | set_zone_contiguous(zone); |
863 | |
864 | return ret; |
865 | } |
866 | EXPORT_SYMBOL_GPL(__remove_pages); |
867 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
868 | |
869 | int set_online_page_callback(online_page_callback_t callback) |
870 | { |
871 | int rc = -EINVAL; |
872 | |
873 | get_online_mems(); |
874 | mutex_lock(&online_page_callback_lock); |
875 | |
876 | if (online_page_callback == generic_online_page) { |
877 | online_page_callback = callback; |
878 | rc = 0; |
879 | } |
880 | |
881 | mutex_unlock(&online_page_callback_lock); |
882 | put_online_mems(); |
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 | |
892 | get_online_mems(); |
893 | mutex_lock(&online_page_callback_lock); |
894 | |
895 | if (online_page_callback == callback) { |
896 | online_page_callback = generic_online_page; |
897 | rc = 0; |
898 | } |
899 | |
900 | mutex_unlock(&online_page_callback_lock); |
901 | put_online_mems(); |
902 | |
903 | return rc; |
904 | } |
905 | EXPORT_SYMBOL_GPL(restore_online_page_callback); |
906 | |
907 | void __online_page_set_limits(struct page *page) |
908 | { |
909 | } |
910 | EXPORT_SYMBOL_GPL(__online_page_set_limits); |
911 | |
912 | void __online_page_increment_counters(struct page *page) |
913 | { |
914 | adjust_managed_page_count(page, 1); |
915 | } |
916 | EXPORT_SYMBOL_GPL(__online_page_increment_counters); |
917 | |
918 | void __online_page_free(struct page *page) |
919 | { |
920 | __free_reserved_page(page); |
921 | } |
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 | } |
930 | |
931 | static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, |
932 | void *arg) |
933 | { |
934 | unsigned long i; |
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); |
940 | (*online_page_callback)(page); |
941 | onlined_pages++; |
942 | } |
943 | *(unsigned long *)arg = onlined_pages; |
944 | return 0; |
945 | } |
946 | |
947 | #ifdef CONFIG_MOVABLE_NODE |
948 | /* |
949 | * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have |
950 | * normal memory. |
951 | */ |
952 | static bool can_online_high_movable(struct zone *zone) |
953 | { |
954 | return true; |
955 | } |
956 | #else /* CONFIG_MOVABLE_NODE */ |
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 | } |
962 | #endif /* CONFIG_MOVABLE_NODE */ |
963 | |
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 | /* |
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. |
975 | * |
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. |
979 | */ |
980 | if (N_MEMORY == N_NORMAL_MEMORY) |
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 | |
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 | |
1016 | /* |
1017 | * if the node don't have memory befor online, we will need to |
1018 | * set the node to node_states[N_MEMORY] after the memory |
1019 | * is online. |
1020 | */ |
1021 | if (!node_state(nid, N_MEMORY)) |
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 | |
1032 | if (arg->status_change_nid_high >= 0) |
1033 | node_set_state(node, N_HIGH_MEMORY); |
1034 | |
1035 | node_set_state(node, N_MEMORY); |
1036 | } |
1037 | |
1038 | bool zone_can_shift(unsigned long pfn, unsigned long nr_pages, |
1039 | enum zone_type target, int *zone_shift) |
1040 | { |
1041 | struct zone *zone = page_zone(pfn_to_page(pfn)); |
1042 | enum zone_type idx = zone_idx(zone); |
1043 | int i; |
1044 | |
1045 | *zone_shift = 0; |
1046 | |
1047 | if (idx < target) { |
1048 | /* pages must be at end of current zone */ |
1049 | if (pfn + nr_pages != zone_end_pfn(zone)) |
1050 | return false; |
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)) |
1055 | return false; |
1056 | } |
1057 | |
1058 | if (target < idx) { |
1059 | /* pages must be at beginning of current zone */ |
1060 | if (pfn != zone->zone_start_pfn) |
1061 | return false; |
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)) |
1066 | return false; |
1067 | } |
1068 | |
1069 | *zone_shift = target - idx; |
1070 | return true; |
1071 | } |
1072 | |
1073 | /* Must be protected by mem_hotplug_begin() */ |
1074 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type) |
1075 | { |
1076 | unsigned long flags; |
1077 | unsigned long onlined_pages = 0; |
1078 | struct zone *zone; |
1079 | int need_zonelists_rebuild = 0; |
1080 | int nid; |
1081 | int ret; |
1082 | struct memory_notify arg; |
1083 | int zone_shift = 0; |
1084 | |
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 | |
1092 | if ((zone_idx(zone) > ZONE_NORMAL || |
1093 | online_type == MMOP_ONLINE_MOVABLE) && |
1094 | !can_online_high_movable(zone)) |
1095 | return -EINVAL; |
1096 | |
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 | } |
1104 | |
1105 | zone = move_pfn_range(zone_shift, pfn, pfn + nr_pages); |
1106 | if (!zone) |
1107 | return -EINVAL; |
1108 | |
1109 | arg.start_pfn = pfn; |
1110 | arg.nr_pages = nr_pages; |
1111 | node_states_check_changes_online(nr_pages, zone, &arg); |
1112 | |
1113 | nid = zone_to_nid(zone); |
1114 | |
1115 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
1116 | ret = notifier_to_errno(ret); |
1117 | if (ret) |
1118 | goto failed_addition; |
1119 | |
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 | */ |
1125 | mutex_lock(&zonelists_mutex); |
1126 | if (!populated_zone(zone)) { |
1127 | need_zonelists_rebuild = 1; |
1128 | build_all_zonelists(NULL, zone); |
1129 | } |
1130 | |
1131 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, |
1132 | online_pages_range); |
1133 | if (ret) { |
1134 | if (need_zonelists_rebuild) |
1135 | zone_pcp_reset(zone); |
1136 | mutex_unlock(&zonelists_mutex); |
1137 | goto failed_addition; |
1138 | } |
1139 | |
1140 | zone->present_pages += onlined_pages; |
1141 | |
1142 | pgdat_resize_lock(zone->zone_pgdat, &flags); |
1143 | zone->zone_pgdat->node_present_pages += onlined_pages; |
1144 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
1145 | |
1146 | if (onlined_pages) { |
1147 | node_states_set_node(nid, &arg); |
1148 | if (need_zonelists_rebuild) |
1149 | build_all_zonelists(NULL, NULL); |
1150 | else |
1151 | zone_pcp_update(zone); |
1152 | } |
1153 | |
1154 | mutex_unlock(&zonelists_mutex); |
1155 | |
1156 | init_per_zone_wmark_min(); |
1157 | |
1158 | if (onlined_pages) { |
1159 | kswapd_run(nid); |
1160 | kcompactd_run(nid); |
1161 | } |
1162 | |
1163 | vm_total_pages = nr_free_pagecache_pages(); |
1164 | |
1165 | writeback_set_ratelimit(); |
1166 | |
1167 | if (onlined_pages) |
1168 | memory_notify(MEM_ONLINE, &arg); |
1169 | return 0; |
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; |
1177 | } |
1178 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
1179 | |
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 | |
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) |
1192 | { |
1193 | struct pglist_data *pgdat; |
1194 | unsigned long zones_size[MAX_NR_ZONES] = {0}; |
1195 | unsigned long zholes_size[MAX_NR_ZONES] = {0}; |
1196 | unsigned long start_pfn = PFN_DOWN(start); |
1197 | |
1198 | pgdat = NODE_DATA(nid); |
1199 | if (!pgdat) { |
1200 | pgdat = arch_alloc_nodedata(nid); |
1201 | if (!pgdat) |
1202 | return NULL; |
1203 | |
1204 | arch_refresh_nodedata(nid, pgdat); |
1205 | } else { |
1206 | /* Reset the nr_zones, order and classzone_idx before reuse */ |
1207 | pgdat->nr_zones = 0; |
1208 | pgdat->kswapd_order = 0; |
1209 | pgdat->kswapd_classzone_idx = 0; |
1210 | } |
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.*/ |
1215 | free_area_init_node(nid, zones_size, start_pfn, zholes_size); |
1216 | pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat); |
1217 | |
1218 | /* |
1219 | * The node we allocated has no zone fallback lists. For avoiding |
1220 | * to access not-initialized zonelist, build here. |
1221 | */ |
1222 | mutex_lock(&zonelists_mutex); |
1223 | build_all_zonelists(pgdat, NULL); |
1224 | mutex_unlock(&zonelists_mutex); |
1225 | |
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 | |
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 | |
1241 | return pgdat; |
1242 | } |
1243 | |
1244 | static void rollback_node_hotadd(int nid, pg_data_t *pgdat) |
1245 | { |
1246 | arch_refresh_nodedata(nid, NULL); |
1247 | free_percpu(pgdat->per_cpu_nodestats); |
1248 | arch_free_nodedata(pgdat); |
1249 | return; |
1250 | } |
1251 | |
1252 | |
1253 | /** |
1254 | * try_online_node - online a node if offlined |
1255 | * |
1256 | * called by cpu_up() to online a node without onlined memory. |
1257 | */ |
1258 | int try_online_node(int nid) |
1259 | { |
1260 | pg_data_t *pgdat; |
1261 | int ret; |
1262 | |
1263 | if (node_online(nid)) |
1264 | return 0; |
1265 | |
1266 | mem_hotplug_begin(); |
1267 | pgdat = hotadd_new_pgdat(nid, 0); |
1268 | if (!pgdat) { |
1269 | pr_err("Cannot online node %d due to NULL pgdat\n", nid); |
1270 | ret = -ENOMEM; |
1271 | goto out; |
1272 | } |
1273 | node_set_online(nid); |
1274 | ret = register_one_node(nid); |
1275 | BUG_ON(ret); |
1276 | |
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 | |
1283 | out: |
1284 | mem_hotplug_done(); |
1285 | return ret; |
1286 | } |
1287 | |
1288 | static int check_hotplug_memory_range(u64 start, u64 size) |
1289 | { |
1290 | u64 start_pfn = PFN_DOWN(start); |
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 | |
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 | |
1325 | int zone_for_memory(int nid, u64 start, u64 size, int zone_default, |
1326 | bool for_device) |
1327 | { |
1328 | #ifdef CONFIG_ZONE_DEVICE |
1329 | if (for_device) |
1330 | return ZONE_DEVICE; |
1331 | #endif |
1332 | if (should_add_memory_movable(nid, start, size)) |
1333 | return ZONE_MOVABLE; |
1334 | |
1335 | return zone_default; |
1336 | } |
1337 | |
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 | |
1343 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
1344 | int __ref add_memory_resource(int nid, struct resource *res, bool online) |
1345 | { |
1346 | u64 start, size; |
1347 | pg_data_t *pgdat = NULL; |
1348 | bool new_pgdat; |
1349 | bool new_node; |
1350 | int ret; |
1351 | |
1352 | start = res->start; |
1353 | size = resource_size(res); |
1354 | |
1355 | ret = check_hotplug_memory_range(start, size); |
1356 | if (ret) |
1357 | return ret; |
1358 | |
1359 | { /* Stupid hack to suppress address-never-null warning */ |
1360 | void *p = NODE_DATA(nid); |
1361 | new_pgdat = !p; |
1362 | } |
1363 | |
1364 | mem_hotplug_begin(); |
1365 | |
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 | |
1374 | new_node = !node_online(nid); |
1375 | if (new_node) { |
1376 | pgdat = hotadd_new_pgdat(nid, start); |
1377 | ret = -ENOMEM; |
1378 | if (!pgdat) |
1379 | goto error; |
1380 | } |
1381 | |
1382 | /* call arch's memory hotadd */ |
1383 | ret = arch_add_memory(nid, start, size, false); |
1384 | |
1385 | if (ret < 0) |
1386 | goto error; |
1387 | |
1388 | /* we online node here. we can't roll back from here. */ |
1389 | node_set_online(nid); |
1390 | |
1391 | if (new_node) { |
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 | |
1401 | /* create new memmap entry */ |
1402 | firmware_map_add_hotplug(start, start + size, "System RAM"); |
1403 | |
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 | |
1409 | goto out; |
1410 | |
1411 | error: |
1412 | /* rollback pgdat allocation and others */ |
1413 | if (new_pgdat) |
1414 | rollback_node_hotadd(nid, pgdat); |
1415 | memblock_remove(start, size); |
1416 | |
1417 | out: |
1418 | mem_hotplug_done(); |
1419 | return ret; |
1420 | } |
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); |
1429 | if (IS_ERR(res)) |
1430 | return PTR_ERR(res); |
1431 | |
1432 | ret = add_memory_resource(nid, res, memhp_auto_online); |
1433 | if (ret < 0) |
1434 | release_memory_resource(res); |
1435 | return ret; |
1436 | } |
1437 | EXPORT_SYMBOL_GPL(add_memory); |
1438 | |
1439 | #ifdef CONFIG_MEMORY_HOTREMOVE |
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 | { |
1455 | /* Ensure the starting page is pageblock-aligned */ |
1456 | BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1)); |
1457 | |
1458 | /* If the entire pageblock is free, move to the end of free page */ |
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 | } |
1466 | |
1467 | return page + pageblock_nr_pages; |
1468 | } |
1469 | |
1470 | /* Checks if this range of memory is likely to be hot-removable. */ |
1471 | bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) |
1472 | { |
1473 | struct page *page = pfn_to_page(start_pfn); |
1474 | unsigned long end_pfn = min(start_pfn + nr_pages, zone_end_pfn(page_zone(page))); |
1475 | struct page *end_page = pfn_to_page(end_pfn); |
1476 | |
1477 | /* Check the starting page of each pageblock within the range */ |
1478 | for (; page < end_page; page = next_active_pageblock(page)) { |
1479 | if (!is_pageblock_removable_nolock(page)) |
1480 | return false; |
1481 | cond_resched(); |
1482 | } |
1483 | |
1484 | /* All pageblocks in the memory block are likely to be hot-removable */ |
1485 | return true; |
1486 | } |
1487 | |
1488 | /* |
1489 | * Confirm all pages in a range [start, end) belong to the same zone. |
1490 | * When true, return its valid [start, end). |
1491 | */ |
1492 | int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn, |
1493 | unsigned long *valid_start, unsigned long *valid_end) |
1494 | { |
1495 | unsigned long pfn, sec_end_pfn; |
1496 | unsigned long start, end; |
1497 | struct zone *zone = NULL; |
1498 | struct page *page; |
1499 | int i; |
1500 | for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1); |
1501 | pfn < end_pfn; |
1502 | pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) { |
1503 | /* Make sure the memory section is present first */ |
1504 | if (!present_section_nr(pfn_to_section_nr(pfn))) |
1505 | continue; |
1506 | for (; pfn < sec_end_pfn && pfn < end_pfn; |
1507 | pfn += MAX_ORDER_NR_PAGES) { |
1508 | i = 0; |
1509 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ |
1510 | while ((i < MAX_ORDER_NR_PAGES) && |
1511 | !pfn_valid_within(pfn + i)) |
1512 | i++; |
1513 | if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn) |
1514 | continue; |
1515 | /* Check if we got outside of the zone */ |
1516 | if (zone && !zone_spans_pfn(zone, pfn + i)) |
1517 | return 0; |
1518 | page = pfn_to_page(pfn + i); |
1519 | if (zone && page_zone(page) != zone) |
1520 | return 0; |
1521 | if (!zone) |
1522 | start = pfn + i; |
1523 | zone = page_zone(page); |
1524 | end = pfn + MAX_ORDER_NR_PAGES; |
1525 | } |
1526 | } |
1527 | |
1528 | if (zone) { |
1529 | *valid_start = start; |
1530 | *valid_end = min(end, end_pfn); |
1531 | return 1; |
1532 | } else { |
1533 | return 0; |
1534 | } |
1535 | } |
1536 | |
1537 | /* |
1538 | * Scan pfn range [start,end) to find movable/migratable pages (LRU pages |
1539 | * and hugepages). We scan pfn because it's much easier than scanning over |
1540 | * linked list. This function returns the pfn of the first found movable |
1541 | * page if it's found, otherwise 0. |
1542 | */ |
1543 | static unsigned long scan_movable_pages(unsigned long start, unsigned long end) |
1544 | { |
1545 | unsigned long pfn; |
1546 | struct page *page; |
1547 | for (pfn = start; pfn < end; pfn++) { |
1548 | if (pfn_valid(pfn)) { |
1549 | page = pfn_to_page(pfn); |
1550 | if (PageLRU(page)) |
1551 | return pfn; |
1552 | if (PageHuge(page)) { |
1553 | if (page_huge_active(page)) |
1554 | return pfn; |
1555 | else |
1556 | pfn = round_up(pfn + 1, |
1557 | 1 << compound_order(page)) - 1; |
1558 | } |
1559 | } |
1560 | } |
1561 | return 0; |
1562 | } |
1563 | |
1564 | static struct page *new_node_page(struct page *page, unsigned long private, |
1565 | int **result) |
1566 | { |
1567 | gfp_t gfp_mask = GFP_USER | __GFP_MOVABLE; |
1568 | int nid = page_to_nid(page); |
1569 | nodemask_t nmask = node_states[N_MEMORY]; |
1570 | struct page *new_page = NULL; |
1571 | |
1572 | /* |
1573 | * TODO: allocate a destination hugepage from a nearest neighbor node, |
1574 | * accordance with memory policy of the user process if possible. For |
1575 | * now as a simple work-around, we use the next node for destination. |
1576 | */ |
1577 | if (PageHuge(page)) |
1578 | return alloc_huge_page_node(page_hstate(compound_head(page)), |
1579 | next_node_in(nid, nmask)); |
1580 | |
1581 | node_clear(nid, nmask); |
1582 | |
1583 | if (PageHighMem(page) |
1584 | || (zone_idx(page_zone(page)) == ZONE_MOVABLE)) |
1585 | gfp_mask |= __GFP_HIGHMEM; |
1586 | |
1587 | if (!nodes_empty(nmask)) |
1588 | new_page = __alloc_pages_nodemask(gfp_mask, 0, |
1589 | node_zonelist(nid, gfp_mask), &nmask); |
1590 | if (!new_page) |
1591 | new_page = __alloc_pages(gfp_mask, 0, |
1592 | node_zonelist(nid, gfp_mask)); |
1593 | |
1594 | return new_page; |
1595 | } |
1596 | |
1597 | #define NR_OFFLINE_AT_ONCE_PAGES (256) |
1598 | static int |
1599 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) |
1600 | { |
1601 | unsigned long pfn; |
1602 | struct page *page; |
1603 | int move_pages = NR_OFFLINE_AT_ONCE_PAGES; |
1604 | int not_managed = 0; |
1605 | int ret = 0; |
1606 | LIST_HEAD(source); |
1607 | |
1608 | for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) { |
1609 | if (!pfn_valid(pfn)) |
1610 | continue; |
1611 | page = pfn_to_page(pfn); |
1612 | |
1613 | if (PageHuge(page)) { |
1614 | struct page *head = compound_head(page); |
1615 | pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1; |
1616 | if (compound_order(head) > PFN_SECTION_SHIFT) { |
1617 | ret = -EBUSY; |
1618 | break; |
1619 | } |
1620 | if (isolate_huge_page(page, &source)) |
1621 | move_pages -= 1 << compound_order(head); |
1622 | continue; |
1623 | } |
1624 | |
1625 | /* |
1626 | * HWPoison pages have elevated reference counts so the migration would |
1627 | * fail on them. It also doesn't make any sense to migrate them in the |
1628 | * first place. Still try to unmap such a page in case it is still mapped |
1629 | * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep |
1630 | * the unmap as the catch all safety net). |
1631 | */ |
1632 | if (PageHWPoison(page)) { |
1633 | if (WARN_ON(PageLRU(page))) |
1634 | isolate_lru_page(page); |
1635 | if (page_mapped(page)) |
1636 | try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS); |
1637 | continue; |
1638 | } |
1639 | |
1640 | if (!get_page_unless_zero(page)) |
1641 | continue; |
1642 | /* |
1643 | * We can skip free pages. And we can only deal with pages on |
1644 | * LRU. |
1645 | */ |
1646 | ret = isolate_lru_page(page); |
1647 | if (!ret) { /* Success */ |
1648 | put_page(page); |
1649 | list_add_tail(&page->lru, &source); |
1650 | move_pages--; |
1651 | inc_node_page_state(page, NR_ISOLATED_ANON + |
1652 | page_is_file_cache(page)); |
1653 | |
1654 | } else { |
1655 | #ifdef CONFIG_DEBUG_VM |
1656 | pr_alert("removing pfn %lx from LRU failed\n", pfn); |
1657 | dump_page(page, "failed to remove from LRU"); |
1658 | #endif |
1659 | put_page(page); |
1660 | /* Because we don't have big zone->lock. we should |
1661 | check this again here. */ |
1662 | if (page_count(page)) { |
1663 | not_managed++; |
1664 | ret = -EBUSY; |
1665 | break; |
1666 | } |
1667 | } |
1668 | } |
1669 | if (!list_empty(&source)) { |
1670 | if (not_managed) { |
1671 | putback_movable_pages(&source); |
1672 | goto out; |
1673 | } |
1674 | |
1675 | /* Allocate a new page from the nearest neighbor node */ |
1676 | ret = migrate_pages(&source, new_node_page, NULL, 0, |
1677 | MIGRATE_SYNC, MR_MEMORY_HOTPLUG); |
1678 | if (ret) |
1679 | putback_movable_pages(&source); |
1680 | } |
1681 | out: |
1682 | return ret; |
1683 | } |
1684 | |
1685 | /* |
1686 | * remove from free_area[] and mark all as Reserved. |
1687 | */ |
1688 | static int |
1689 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, |
1690 | void *data) |
1691 | { |
1692 | __offline_isolated_pages(start, start + nr_pages); |
1693 | return 0; |
1694 | } |
1695 | |
1696 | static void |
1697 | offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) |
1698 | { |
1699 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL, |
1700 | offline_isolated_pages_cb); |
1701 | } |
1702 | |
1703 | /* |
1704 | * Check all pages in range, recoreded as memory resource, are isolated. |
1705 | */ |
1706 | static int |
1707 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, |
1708 | void *data) |
1709 | { |
1710 | int ret; |
1711 | long offlined = *(long *)data; |
1712 | ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true); |
1713 | offlined = nr_pages; |
1714 | if (!ret) |
1715 | *(long *)data += offlined; |
1716 | return ret; |
1717 | } |
1718 | |
1719 | static long |
1720 | check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn) |
1721 | { |
1722 | long offlined = 0; |
1723 | int ret; |
1724 | |
1725 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined, |
1726 | check_pages_isolated_cb); |
1727 | if (ret < 0) |
1728 | offlined = (long)ret; |
1729 | return offlined; |
1730 | } |
1731 | |
1732 | #ifdef CONFIG_MOVABLE_NODE |
1733 | /* |
1734 | * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have |
1735 | * normal memory. |
1736 | */ |
1737 | static bool can_offline_normal(struct zone *zone, unsigned long nr_pages) |
1738 | { |
1739 | return true; |
1740 | } |
1741 | #else /* CONFIG_MOVABLE_NODE */ |
1742 | /* ensure the node has NORMAL memory if it is still online */ |
1743 | static bool can_offline_normal(struct zone *zone, unsigned long nr_pages) |
1744 | { |
1745 | struct pglist_data *pgdat = zone->zone_pgdat; |
1746 | unsigned long present_pages = 0; |
1747 | enum zone_type zt; |
1748 | |
1749 | for (zt = 0; zt <= ZONE_NORMAL; zt++) |
1750 | present_pages += pgdat->node_zones[zt].present_pages; |
1751 | |
1752 | if (present_pages > nr_pages) |
1753 | return true; |
1754 | |
1755 | present_pages = 0; |
1756 | for (; zt <= ZONE_MOVABLE; zt++) |
1757 | present_pages += pgdat->node_zones[zt].present_pages; |
1758 | |
1759 | /* |
1760 | * we can't offline the last normal memory until all |
1761 | * higher memory is offlined. |
1762 | */ |
1763 | return present_pages == 0; |
1764 | } |
1765 | #endif /* CONFIG_MOVABLE_NODE */ |
1766 | |
1767 | static int __init cmdline_parse_movable_node(char *p) |
1768 | { |
1769 | #ifdef CONFIG_MOVABLE_NODE |
1770 | /* |
1771 | * Memory used by the kernel cannot be hot-removed because Linux |
1772 | * cannot migrate the kernel pages. When memory hotplug is |
1773 | * enabled, we should prevent memblock from allocating memory |
1774 | * for the kernel. |
1775 | * |
1776 | * ACPI SRAT records all hotpluggable memory ranges. But before |
1777 | * SRAT is parsed, we don't know about it. |
1778 | * |
1779 | * The kernel image is loaded into memory at very early time. We |
1780 | * cannot prevent this anyway. So on NUMA system, we set any |
1781 | * node the kernel resides in as un-hotpluggable. |
1782 | * |
1783 | * Since on modern servers, one node could have double-digit |
1784 | * gigabytes memory, we can assume the memory around the kernel |
1785 | * image is also un-hotpluggable. So before SRAT is parsed, just |
1786 | * allocate memory near the kernel image to try the best to keep |
1787 | * the kernel away from hotpluggable memory. |
1788 | */ |
1789 | memblock_set_bottom_up(true); |
1790 | movable_node_enabled = true; |
1791 | #else |
1792 | pr_warn("movable_node option not supported\n"); |
1793 | #endif |
1794 | return 0; |
1795 | } |
1796 | early_param("movable_node", cmdline_parse_movable_node); |
1797 | |
1798 | /* check which state of node_states will be changed when offline memory */ |
1799 | static void node_states_check_changes_offline(unsigned long nr_pages, |
1800 | struct zone *zone, struct memory_notify *arg) |
1801 | { |
1802 | struct pglist_data *pgdat = zone->zone_pgdat; |
1803 | unsigned long present_pages = 0; |
1804 | enum zone_type zt, zone_last = ZONE_NORMAL; |
1805 | |
1806 | /* |
1807 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
1808 | * contains nodes which have zones of 0...ZONE_NORMAL, |
1809 | * set zone_last to ZONE_NORMAL. |
1810 | * |
1811 | * If we don't have HIGHMEM nor movable node, |
1812 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of |
1813 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. |
1814 | */ |
1815 | if (N_MEMORY == N_NORMAL_MEMORY) |
1816 | zone_last = ZONE_MOVABLE; |
1817 | |
1818 | /* |
1819 | * check whether node_states[N_NORMAL_MEMORY] will be changed. |
1820 | * If the memory to be offline is in a zone of 0...zone_last, |
1821 | * and it is the last present memory, 0...zone_last will |
1822 | * become empty after offline , thus we can determind we will |
1823 | * need to clear the node from node_states[N_NORMAL_MEMORY]. |
1824 | */ |
1825 | for (zt = 0; zt <= zone_last; zt++) |
1826 | present_pages += pgdat->node_zones[zt].present_pages; |
1827 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) |
1828 | arg->status_change_nid_normal = zone_to_nid(zone); |
1829 | else |
1830 | arg->status_change_nid_normal = -1; |
1831 | |
1832 | #ifdef CONFIG_HIGHMEM |
1833 | /* |
1834 | * If we have movable node, node_states[N_HIGH_MEMORY] |
1835 | * contains nodes which have zones of 0...ZONE_HIGHMEM, |
1836 | * set zone_last to ZONE_HIGHMEM. |
1837 | * |
1838 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] |
1839 | * contains nodes which have zones of 0...ZONE_MOVABLE, |
1840 | * set zone_last to ZONE_MOVABLE. |
1841 | */ |
1842 | zone_last = ZONE_HIGHMEM; |
1843 | if (N_MEMORY == N_HIGH_MEMORY) |
1844 | zone_last = ZONE_MOVABLE; |
1845 | |
1846 | for (; zt <= zone_last; zt++) |
1847 | present_pages += pgdat->node_zones[zt].present_pages; |
1848 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) |
1849 | arg->status_change_nid_high = zone_to_nid(zone); |
1850 | else |
1851 | arg->status_change_nid_high = -1; |
1852 | #else |
1853 | arg->status_change_nid_high = arg->status_change_nid_normal; |
1854 | #endif |
1855 | |
1856 | /* |
1857 | * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE |
1858 | */ |
1859 | zone_last = ZONE_MOVABLE; |
1860 | |
1861 | /* |
1862 | * check whether node_states[N_HIGH_MEMORY] will be changed |
1863 | * If we try to offline the last present @nr_pages from the node, |
1864 | * we can determind we will need to clear the node from |
1865 | * node_states[N_HIGH_MEMORY]. |
1866 | */ |
1867 | for (; zt <= zone_last; zt++) |
1868 | present_pages += pgdat->node_zones[zt].present_pages; |
1869 | if (nr_pages >= present_pages) |
1870 | arg->status_change_nid = zone_to_nid(zone); |
1871 | else |
1872 | arg->status_change_nid = -1; |
1873 | } |
1874 | |
1875 | static void node_states_clear_node(int node, struct memory_notify *arg) |
1876 | { |
1877 | if (arg->status_change_nid_normal >= 0) |
1878 | node_clear_state(node, N_NORMAL_MEMORY); |
1879 | |
1880 | if ((N_MEMORY != N_NORMAL_MEMORY) && |
1881 | (arg->status_change_nid_high >= 0)) |
1882 | node_clear_state(node, N_HIGH_MEMORY); |
1883 | |
1884 | if ((N_MEMORY != N_HIGH_MEMORY) && |
1885 | (arg->status_change_nid >= 0)) |
1886 | node_clear_state(node, N_MEMORY); |
1887 | } |
1888 | |
1889 | static int __ref __offline_pages(unsigned long start_pfn, |
1890 | unsigned long end_pfn, unsigned long timeout) |
1891 | { |
1892 | unsigned long pfn, nr_pages, expire; |
1893 | long offlined_pages; |
1894 | int ret, drain, retry_max, node; |
1895 | unsigned long flags; |
1896 | unsigned long valid_start, valid_end; |
1897 | struct zone *zone; |
1898 | struct memory_notify arg; |
1899 | |
1900 | /* at least, alignment against pageblock is necessary */ |
1901 | if (!IS_ALIGNED(start_pfn, pageblock_nr_pages)) |
1902 | return -EINVAL; |
1903 | if (!IS_ALIGNED(end_pfn, pageblock_nr_pages)) |
1904 | return -EINVAL; |
1905 | /* This makes hotplug much easier...and readable. |
1906 | we assume this for now. .*/ |
1907 | if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, &valid_end)) |
1908 | return -EINVAL; |
1909 | |
1910 | zone = page_zone(pfn_to_page(valid_start)); |
1911 | node = zone_to_nid(zone); |
1912 | nr_pages = end_pfn - start_pfn; |
1913 | |
1914 | if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages)) |
1915 | return -EINVAL; |
1916 | |
1917 | /* set above range as isolated */ |
1918 | ret = start_isolate_page_range(start_pfn, end_pfn, |
1919 | MIGRATE_MOVABLE, true); |
1920 | if (ret) |
1921 | return ret; |
1922 | |
1923 | arg.start_pfn = start_pfn; |
1924 | arg.nr_pages = nr_pages; |
1925 | node_states_check_changes_offline(nr_pages, zone, &arg); |
1926 | |
1927 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); |
1928 | ret = notifier_to_errno(ret); |
1929 | if (ret) |
1930 | goto failed_removal; |
1931 | |
1932 | pfn = start_pfn; |
1933 | expire = jiffies + timeout; |
1934 | drain = 0; |
1935 | retry_max = 5; |
1936 | repeat: |
1937 | /* start memory hot removal */ |
1938 | ret = -EAGAIN; |
1939 | if (time_after(jiffies, expire)) |
1940 | goto failed_removal; |
1941 | ret = -EINTR; |
1942 | if (signal_pending(current)) |
1943 | goto failed_removal; |
1944 | ret = 0; |
1945 | if (drain) { |
1946 | lru_add_drain_all(); |
1947 | cond_resched(); |
1948 | drain_all_pages(zone); |
1949 | } |
1950 | |
1951 | pfn = scan_movable_pages(start_pfn, end_pfn); |
1952 | if (pfn) { /* We have movable pages */ |
1953 | ret = do_migrate_range(pfn, end_pfn); |
1954 | if (!ret) { |
1955 | drain = 1; |
1956 | goto repeat; |
1957 | } else { |
1958 | if (ret < 0) |
1959 | if (--retry_max == 0) |
1960 | goto failed_removal; |
1961 | yield(); |
1962 | drain = 1; |
1963 | goto repeat; |
1964 | } |
1965 | } |
1966 | /* drain all zone's lru pagevec, this is asynchronous... */ |
1967 | lru_add_drain_all(); |
1968 | yield(); |
1969 | /* drain pcp pages, this is synchronous. */ |
1970 | drain_all_pages(zone); |
1971 | /* |
1972 | * dissolve free hugepages in the memory block before doing offlining |
1973 | * actually in order to make hugetlbfs's object counting consistent. |
1974 | */ |
1975 | ret = dissolve_free_huge_pages(start_pfn, end_pfn); |
1976 | if (ret) |
1977 | goto failed_removal; |
1978 | /* check again */ |
1979 | offlined_pages = check_pages_isolated(start_pfn, end_pfn); |
1980 | if (offlined_pages < 0) { |
1981 | ret = -EBUSY; |
1982 | goto failed_removal; |
1983 | } |
1984 | pr_info("Offlined Pages %ld\n", offlined_pages); |
1985 | /* Ok, all of our target is isolated. |
1986 | We cannot do rollback at this point. */ |
1987 | offline_isolated_pages(start_pfn, end_pfn); |
1988 | /* reset pagetype flags and makes migrate type to be MOVABLE */ |
1989 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
1990 | /* removal success */ |
1991 | adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages); |
1992 | zone->present_pages -= offlined_pages; |
1993 | |
1994 | pgdat_resize_lock(zone->zone_pgdat, &flags); |
1995 | zone->zone_pgdat->node_present_pages -= offlined_pages; |
1996 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
1997 | |
1998 | init_per_zone_wmark_min(); |
1999 | |
2000 | if (!populated_zone(zone)) { |
2001 | zone_pcp_reset(zone); |
2002 | mutex_lock(&zonelists_mutex); |
2003 | build_all_zonelists(NULL, NULL); |
2004 | mutex_unlock(&zonelists_mutex); |
2005 | } else |
2006 | zone_pcp_update(zone); |
2007 | |
2008 | node_states_clear_node(node, &arg); |
2009 | if (arg.status_change_nid >= 0) { |
2010 | kswapd_stop(node); |
2011 | kcompactd_stop(node); |
2012 | } |
2013 | |
2014 | vm_total_pages = nr_free_pagecache_pages(); |
2015 | writeback_set_ratelimit(); |
2016 | |
2017 | memory_notify(MEM_OFFLINE, &arg); |
2018 | return 0; |
2019 | |
2020 | failed_removal: |
2021 | pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n", |
2022 | (unsigned long long) start_pfn << PAGE_SHIFT, |
2023 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1); |
2024 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
2025 | /* pushback to free area */ |
2026 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
2027 | return ret; |
2028 | } |
2029 | |
2030 | /* Must be protected by mem_hotplug_begin() */ |
2031 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
2032 | { |
2033 | return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ); |
2034 | } |
2035 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
2036 | |
2037 | /** |
2038 | * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn) |
2039 | * @start_pfn: start pfn of the memory range |
2040 | * @end_pfn: end pfn of the memory range |
2041 | * @arg: argument passed to func |
2042 | * @func: callback for each memory section walked |
2043 | * |
2044 | * This function walks through all present mem sections in range |
2045 | * [start_pfn, end_pfn) and call func on each mem section. |
2046 | * |
2047 | * Returns the return value of func. |
2048 | */ |
2049 | int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn, |
2050 | void *arg, int (*func)(struct memory_block *, void *)) |
2051 | { |
2052 | struct memory_block *mem = NULL; |
2053 | struct mem_section *section; |
2054 | unsigned long pfn, section_nr; |
2055 | int ret; |
2056 | |
2057 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
2058 | section_nr = pfn_to_section_nr(pfn); |
2059 | if (!present_section_nr(section_nr)) |
2060 | continue; |
2061 | |
2062 | section = __nr_to_section(section_nr); |
2063 | /* same memblock? */ |
2064 | if (mem) |
2065 | if ((section_nr >= mem->start_section_nr) && |
2066 | (section_nr <= mem->end_section_nr)) |
2067 | continue; |
2068 | |
2069 | mem = find_memory_block_hinted(section, mem); |
2070 | if (!mem) |
2071 | continue; |
2072 | |
2073 | ret = func(mem, arg); |
2074 | if (ret) { |
2075 | kobject_put(&mem->dev.kobj); |
2076 | return ret; |
2077 | } |
2078 | } |
2079 | |
2080 | if (mem) |
2081 | kobject_put(&mem->dev.kobj); |
2082 | |
2083 | return 0; |
2084 | } |
2085 | |
2086 | #ifdef CONFIG_MEMORY_HOTREMOVE |
2087 | static int check_memblock_offlined_cb(struct memory_block *mem, void *arg) |
2088 | { |
2089 | int ret = !is_memblock_offlined(mem); |
2090 | |
2091 | if (unlikely(ret)) { |
2092 | phys_addr_t beginpa, endpa; |
2093 | |
2094 | beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)); |
2095 | endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1; |
2096 | pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n", |
2097 | &beginpa, &endpa); |
2098 | } |
2099 | |
2100 | return ret; |
2101 | } |
2102 | |
2103 | static int check_cpu_on_node(pg_data_t *pgdat) |
2104 | { |
2105 | int cpu; |
2106 | |
2107 | for_each_present_cpu(cpu) { |
2108 | if (cpu_to_node(cpu) == pgdat->node_id) |
2109 | /* |
2110 | * the cpu on this node isn't removed, and we can't |
2111 | * offline this node. |
2112 | */ |
2113 | return -EBUSY; |
2114 | } |
2115 | |
2116 | return 0; |
2117 | } |
2118 | |
2119 | static void unmap_cpu_on_node(pg_data_t *pgdat) |
2120 | { |
2121 | #ifdef CONFIG_ACPI_NUMA |
2122 | int cpu; |
2123 | |
2124 | for_each_possible_cpu(cpu) |
2125 | if (cpu_to_node(cpu) == pgdat->node_id) |
2126 | numa_clear_node(cpu); |
2127 | #endif |
2128 | } |
2129 | |
2130 | static int check_and_unmap_cpu_on_node(pg_data_t *pgdat) |
2131 | { |
2132 | int ret; |
2133 | |
2134 | ret = check_cpu_on_node(pgdat); |
2135 | if (ret) |
2136 | return ret; |
2137 | |
2138 | /* |
2139 | * the node will be offlined when we come here, so we can clear |
2140 | * the cpu_to_node() now. |
2141 | */ |
2142 | |
2143 | unmap_cpu_on_node(pgdat); |
2144 | return 0; |
2145 | } |
2146 | |
2147 | /** |
2148 | * try_offline_node |
2149 | * |
2150 | * Offline a node if all memory sections and cpus of the node are removed. |
2151 | * |
2152 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug |
2153 | * and online/offline operations before this call. |
2154 | */ |
2155 | void try_offline_node(int nid) |
2156 | { |
2157 | pg_data_t *pgdat = NODE_DATA(nid); |
2158 | unsigned long start_pfn = pgdat->node_start_pfn; |
2159 | unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages; |
2160 | unsigned long pfn; |
2161 | |
2162 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
2163 | unsigned long section_nr = pfn_to_section_nr(pfn); |
2164 | |
2165 | if (!present_section_nr(section_nr)) |
2166 | continue; |
2167 | |
2168 | if (pfn_to_nid(pfn) != nid) |
2169 | continue; |
2170 | |
2171 | /* |
2172 | * some memory sections of this node are not removed, and we |
2173 | * can't offline node now. |
2174 | */ |
2175 | return; |
2176 | } |
2177 | |
2178 | if (check_and_unmap_cpu_on_node(pgdat)) |
2179 | return; |
2180 | |
2181 | /* |
2182 | * all memory/cpu of this node are removed, we can offline this |
2183 | * node now. |
2184 | */ |
2185 | node_set_offline(nid); |
2186 | unregister_one_node(nid); |
2187 | } |
2188 | EXPORT_SYMBOL(try_offline_node); |
2189 | |
2190 | /** |
2191 | * remove_memory |
2192 | * |
2193 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug |
2194 | * and online/offline operations before this call, as required by |
2195 | * try_offline_node(). |
2196 | */ |
2197 | void __ref remove_memory(int nid, u64 start, u64 size) |
2198 | { |
2199 | int ret; |
2200 | |
2201 | BUG_ON(check_hotplug_memory_range(start, size)); |
2202 | |
2203 | mem_hotplug_begin(); |
2204 | |
2205 | /* |
2206 | * All memory blocks must be offlined before removing memory. Check |
2207 | * whether all memory blocks in question are offline and trigger a BUG() |
2208 | * if this is not the case. |
2209 | */ |
2210 | ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL, |
2211 | check_memblock_offlined_cb); |
2212 | if (ret) |
2213 | BUG(); |
2214 | |
2215 | /* remove memmap entry */ |
2216 | firmware_map_remove(start, start + size, "System RAM"); |
2217 | memblock_free(start, size); |
2218 | memblock_remove(start, size); |
2219 | |
2220 | arch_remove_memory(start, size); |
2221 | |
2222 | try_offline_node(nid); |
2223 | |
2224 | mem_hotplug_done(); |
2225 | } |
2226 | EXPORT_SYMBOL_GPL(remove_memory); |
2227 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
2228 |