projects / linux-2.6-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Merge branch 'master' of /pub/scm/linux/kernel/git/torvalds/linux-2.6
[linux-2.6-block.git] / mm / memory_hotplug.c
... / ...
CommitLineData
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/bootmem.h>
13#include <linux/compiler.h>
14#include <linux/module.h>
15#include <linux/pagevec.h>
16#include <linux/writeback.h>
17#include <linux/slab.h>
18#include <linux/sysctl.h>
19#include <linux/cpu.h>
20#include <linux/memory.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
33#include <asm/tlbflush.h>
34
35#include "internal.h"
36
37/* add this memory to iomem resource */
38static struct resource *register_memory_resource(u64 start, u64 size)
39{
40 struct resource *res;
41 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
42 BUG_ON(!res);
43
44 res->name = "System RAM";
45 res->start = start;
46 res->end = start + size - 1;
47 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
48 if (request_resource(&iomem_resource, res) < 0) {
49 printk("System RAM resource %llx - %llx cannot be added\n",
50 (unsigned long long)res->start, (unsigned long long)res->end);
51 kfree(res);
52 res = NULL;
53 }
54 return res;
55}
56
57static void release_memory_resource(struct resource *res)
58{
59 if (!res)
60 return;
61 release_resource(res);
62 kfree(res);
63 return;
64}
65
66#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
67#ifndef CONFIG_SPARSEMEM_VMEMMAP
68static void get_page_bootmem(unsigned long info, struct page *page, int type)
69{
70 atomic_set(&page->_mapcount, type);
71 SetPagePrivate(page);
72 set_page_private(page, info);
73 atomic_inc(&page->_count);
74}
75
76/* reference to __meminit __free_pages_bootmem is valid
77 * so use __ref to tell modpost not to generate a warning */
78void __ref put_page_bootmem(struct page *page)
79{
80 int type;
81
82 type = atomic_read(&page->_mapcount);
83 BUG_ON(type >= -1);
84
85 if (atomic_dec_return(&page->_count) == 1) {
86 ClearPagePrivate(page);
87 set_page_private(page, 0);
88 reset_page_mapcount(page);
89 __free_pages_bootmem(page, 0);
90 }
91
92}
93
94static void register_page_bootmem_info_section(unsigned long start_pfn)
95{
96 unsigned long *usemap, mapsize, section_nr, i;
97 struct mem_section *ms;
98 struct page *page, *memmap;
99
100 if (!pfn_valid(start_pfn))
101 return;
102
103 section_nr = pfn_to_section_nr(start_pfn);
104 ms = __nr_to_section(section_nr);
105
106 /* Get section's memmap address */
107 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
108
109 /*
110 * Get page for the memmap's phys address
111 * XXX: need more consideration for sparse_vmemmap...
112 */
113 page = virt_to_page(memmap);
114 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
115 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
116
117 /* remember memmap's page */
118 for (i = 0; i < mapsize; i++, page++)
119 get_page_bootmem(section_nr, page, SECTION_INFO);
120
121 usemap = __nr_to_section(section_nr)->pageblock_flags;
122 page = virt_to_page(usemap);
123
124 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
125
126 for (i = 0; i < mapsize; i++, page++)
127 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
128
129}
130
131void register_page_bootmem_info_node(struct pglist_data *pgdat)
132{
133 unsigned long i, pfn, end_pfn, nr_pages;
134 int node = pgdat->node_id;
135 struct page *page;
136 struct zone *zone;
137
138 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
139 page = virt_to_page(pgdat);
140
141 for (i = 0; i < nr_pages; i++, page++)
142 get_page_bootmem(node, page, NODE_INFO);
143
144 zone = &pgdat->node_zones[0];
145 for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
146 if (zone->wait_table) {
147 nr_pages = zone->wait_table_hash_nr_entries
148 * sizeof(wait_queue_head_t);
149 nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
150 page = virt_to_page(zone->wait_table);
151
152 for (i = 0; i < nr_pages; i++, page++)
153 get_page_bootmem(node, page, NODE_INFO);
154 }
155 }
156
157 pfn = pgdat->node_start_pfn;
158 end_pfn = pfn + pgdat->node_spanned_pages;
159
160 /* register_section info */
161 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION)
162 register_page_bootmem_info_section(pfn);
163
164}
165#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
166
167static void grow_zone_span(struct zone *zone, unsigned long start_pfn,
168 unsigned long end_pfn)
169{
170 unsigned long old_zone_end_pfn;
171
172 zone_span_writelock(zone);
173
174 old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
175 if (start_pfn < zone->zone_start_pfn)
176 zone->zone_start_pfn = start_pfn;
177
178 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
179 zone->zone_start_pfn;
180
181 zone_span_writeunlock(zone);
182}
183
184static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
185 unsigned long end_pfn)
186{
187 unsigned long old_pgdat_end_pfn =
188 pgdat->node_start_pfn + pgdat->node_spanned_pages;
189
190 if (start_pfn < pgdat->node_start_pfn)
191 pgdat->node_start_pfn = start_pfn;
192
193 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
194 pgdat->node_start_pfn;
195}
196
197static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
198{
199 struct pglist_data *pgdat = zone->zone_pgdat;
200 int nr_pages = PAGES_PER_SECTION;
201 int nid = pgdat->node_id;
202 int zone_type;
203 unsigned long flags;
204
205 zone_type = zone - pgdat->node_zones;
206 if (!zone->wait_table) {
207 int ret;
208
209 ret = init_currently_empty_zone(zone, phys_start_pfn,
210 nr_pages, MEMMAP_HOTPLUG);
211 if (ret)
212 return ret;
213 }
214 pgdat_resize_lock(zone->zone_pgdat, &flags);
215 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
216 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
217 phys_start_pfn + nr_pages);
218 pgdat_resize_unlock(zone->zone_pgdat, &flags);
219 memmap_init_zone(nr_pages, nid, zone_type,
220 phys_start_pfn, MEMMAP_HOTPLUG);
221 return 0;
222}
223
224static int __meminit __add_section(int nid, struct zone *zone,
225 unsigned long phys_start_pfn)
226{
227 int nr_pages = PAGES_PER_SECTION;
228 int ret;
229
230 if (pfn_valid(phys_start_pfn))
231 return -EEXIST;
232
233 ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
234
235 if (ret < 0)
236 return ret;
237
238 ret = __add_zone(zone, phys_start_pfn);
239
240 if (ret < 0)
241 return ret;
242
243 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
244}
245
246#ifdef CONFIG_SPARSEMEM_VMEMMAP
247static int __remove_section(struct zone *zone, struct mem_section *ms)
248{
249 /*
250 * XXX: Freeing memmap with vmemmap is not implement yet.
251 * This should be removed later.
252 */
253 return -EBUSY;
254}
255#else
256static int __remove_section(struct zone *zone, struct mem_section *ms)
257{
258 unsigned long flags;
259 struct pglist_data *pgdat = zone->zone_pgdat;
260 int ret = -EINVAL;
261
262 if (!valid_section(ms))
263 return ret;
264
265 ret = unregister_memory_section(ms);
266 if (ret)
267 return ret;
268
269 pgdat_resize_lock(pgdat, &flags);
270 sparse_remove_one_section(zone, ms);
271 pgdat_resize_unlock(pgdat, &flags);
272 return 0;
273}
274#endif
275
276/*
277 * Reasonably generic function for adding memory. It is
278 * expected that archs that support memory hotplug will
279 * call this function after deciding the zone to which to
280 * add the new pages.
281 */
282int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
283 unsigned long nr_pages)
284{
285 unsigned long i;
286 int err = 0;
287 int start_sec, end_sec;
288 /* during initialize mem_map, align hot-added range to section */
289 start_sec = pfn_to_section_nr(phys_start_pfn);
290 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
291
292 for (i = start_sec; i <= end_sec; i++) {
293 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
294
295 /*
296 * EEXIST is finally dealt with by ioresource collision
297 * check. see add_memory() => register_memory_resource()
298 * Warning will be printed if there is collision.
299 */
300 if (err && (err != -EEXIST))
301 break;
302 err = 0;
303 }
304
305 return err;
306}
307EXPORT_SYMBOL_GPL(__add_pages);
308
309/**
310 * __remove_pages() - remove sections of pages from a zone
311 * @zone: zone from which pages need to be removed
312 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
313 * @nr_pages: number of pages to remove (must be multiple of section size)
314 *
315 * Generic helper function to remove section mappings and sysfs entries
316 * for the section of the memory we are removing. Caller needs to make
317 * sure that pages are marked reserved and zones are adjust properly by
318 * calling offline_pages().
319 */
320int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
321 unsigned long nr_pages)
322{
323 unsigned long i, ret = 0;
324 int sections_to_remove;
325
326 /*
327 * We can only remove entire sections
328 */
329 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
330 BUG_ON(nr_pages % PAGES_PER_SECTION);
331
332 sections_to_remove = nr_pages / PAGES_PER_SECTION;
333 for (i = 0; i < sections_to_remove; i++) {
334 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
335 release_mem_region(pfn << PAGE_SHIFT,
336 PAGES_PER_SECTION << PAGE_SHIFT);
337 ret = __remove_section(zone, __pfn_to_section(pfn));
338 if (ret)
339 break;
340 }
341 return ret;
342}
343EXPORT_SYMBOL_GPL(__remove_pages);
344
345void online_page(struct page *page)
346{
347 unsigned long pfn = page_to_pfn(page);
348
349 totalram_pages++;
350 if (pfn >= num_physpages)
351 num_physpages = pfn + 1;
352
353#ifdef CONFIG_HIGHMEM
354 if (PageHighMem(page))
355 totalhigh_pages++;
356#endif
357
358#ifdef CONFIG_FLATMEM
359 max_mapnr = max(page_to_pfn(page), max_mapnr);
360#endif
361
362 ClearPageReserved(page);
363 init_page_count(page);
364 __free_page(page);
365}
366
367static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
368 void *arg)
369{
370 unsigned long i;
371 unsigned long onlined_pages = *(unsigned long *)arg;
372 struct page *page;
373 if (PageReserved(pfn_to_page(start_pfn)))
374 for (i = 0; i < nr_pages; i++) {
375 page = pfn_to_page(start_pfn + i);
376 online_page(page);
377 onlined_pages++;
378 }
379 *(unsigned long *)arg = onlined_pages;
380 return 0;
381}
382
383
384int online_pages(unsigned long pfn, unsigned long nr_pages)
385{
386 unsigned long onlined_pages = 0;
387 struct zone *zone;
388 int need_zonelists_rebuild = 0;
389 int nid;
390 int ret;
391 struct memory_notify arg;
392
393 arg.start_pfn = pfn;
394 arg.nr_pages = nr_pages;
395 arg.status_change_nid = -1;
396
397 nid = page_to_nid(pfn_to_page(pfn));
398 if (node_present_pages(nid) == 0)
399 arg.status_change_nid = nid;
400
401 ret = memory_notify(MEM_GOING_ONLINE, &arg);
402 ret = notifier_to_errno(ret);
403 if (ret) {
404 memory_notify(MEM_CANCEL_ONLINE, &arg);
405 return ret;
406 }
407 /*
408 * This doesn't need a lock to do pfn_to_page().
409 * The section can't be removed here because of the
410 * memory_block->state_mutex.
411 */
412 zone = page_zone(pfn_to_page(pfn));
413 /*
414 * If this zone is not populated, then it is not in zonelist.
415 * This means the page allocator ignores this zone.
416 * So, zonelist must be updated after online.
417 */
418 if (!populated_zone(zone))
419 need_zonelists_rebuild = 1;
420
421 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
422 online_pages_range);
423 if (ret) {
424 printk(KERN_DEBUG "online_pages %lx at %lx failed\n",
425 nr_pages, pfn);
426 memory_notify(MEM_CANCEL_ONLINE, &arg);
427 return ret;
428 }
429
430 zone->present_pages += onlined_pages;
431 zone->zone_pgdat->node_present_pages += onlined_pages;
432
433 zone_pcp_update(zone);
434 setup_per_zone_wmarks();
435 calculate_zone_inactive_ratio(zone);
436 if (onlined_pages) {
437 kswapd_run(zone_to_nid(zone));
438 node_set_state(zone_to_nid(zone), N_HIGH_MEMORY);
439 }
440
441 if (need_zonelists_rebuild)
442 build_all_zonelists();
443 else
444 vm_total_pages = nr_free_pagecache_pages();
445
446 writeback_set_ratelimit();
447
448 if (onlined_pages)
449 memory_notify(MEM_ONLINE, &arg);
450
451 return 0;
452}
453#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
454
455/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
456static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
457{
458 struct pglist_data *pgdat;
459 unsigned long zones_size[MAX_NR_ZONES] = {0};
460 unsigned long zholes_size[MAX_NR_ZONES] = {0};
461 unsigned long start_pfn = start >> PAGE_SHIFT;
462
463 pgdat = arch_alloc_nodedata(nid);
464 if (!pgdat)
465 return NULL;
466
467 arch_refresh_nodedata(nid, pgdat);
468
469 /* we can use NODE_DATA(nid) from here */
470
471 /* init node's zones as empty zones, we don't have any present pages.*/
472 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
473
474 return pgdat;
475}
476
477static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
478{
479 arch_refresh_nodedata(nid, NULL);
480 arch_free_nodedata(pgdat);
481 return;
482}
483
484
485/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
486int __ref add_memory(int nid, u64 start, u64 size)
487{
488 pg_data_t *pgdat = NULL;
489 int new_pgdat = 0;
490 struct resource *res;
491 int ret;
492
493 lock_system_sleep();
494
495 res = register_memory_resource(start, size);
496 ret = -EEXIST;
497 if (!res)
498 goto out;
499
500 if (!node_online(nid)) {
501 pgdat = hotadd_new_pgdat(nid, start);
502 ret = -ENOMEM;
503 if (!pgdat)
504 goto out;
505 new_pgdat = 1;
506 }
507
508 /* call arch's memory hotadd */
509 ret = arch_add_memory(nid, start, size);
510
511 if (ret < 0)
512 goto error;
513
514 /* we online node here. we can't roll back from here. */
515 node_set_online(nid);
516
517 if (new_pgdat) {
518 ret = register_one_node(nid);
519 /*
520 * If sysfs file of new node can't create, cpu on the node
521 * can't be hot-added. There is no rollback way now.
522 * So, check by BUG_ON() to catch it reluctantly..
523 */
524 BUG_ON(ret);
525 }
526
527 /* create new memmap entry */
528 firmware_map_add_hotplug(start, start + size, "System RAM");
529
530 goto out;
531
532error:
533 /* rollback pgdat allocation and others */
534 if (new_pgdat)
535 rollback_node_hotadd(nid, pgdat);
536 if (res)
537 release_memory_resource(res);
538
539out:
540 unlock_system_sleep();
541 return ret;
542}
543EXPORT_SYMBOL_GPL(add_memory);
544
545#ifdef CONFIG_MEMORY_HOTREMOVE
546/*
547 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
548 * set and the size of the free page is given by page_order(). Using this,
549 * the function determines if the pageblock contains only free pages.
550 * Due to buddy contraints, a free page at least the size of a pageblock will
551 * be located at the start of the pageblock
552 */
553static inline int pageblock_free(struct page *page)
554{
555 return PageBuddy(page) && page_order(page) >= pageblock_order;
556}
557
558/* Return the start of the next active pageblock after a given page */
559static struct page *next_active_pageblock(struct page *page)
560{
561 int pageblocks_stride;
562
563 /* Ensure the starting page is pageblock-aligned */
564 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
565
566 /* Move forward by at least 1 * pageblock_nr_pages */
567 pageblocks_stride = 1;
568
569 /* If the entire pageblock is free, move to the end of free page */
570 if (pageblock_free(page))
571 pageblocks_stride += page_order(page) - pageblock_order;
572
573 return page + (pageblocks_stride * pageblock_nr_pages);
574}
575
576/* Checks if this range of memory is likely to be hot-removable. */
577int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
578{
579 int type;
580 struct page *page = pfn_to_page(start_pfn);
581 struct page *end_page = page + nr_pages;
582
583 /* Check the starting page of each pageblock within the range */
584 for (; page < end_page; page = next_active_pageblock(page)) {
585 type = get_pageblock_migratetype(page);
586
587 /*
588 * A pageblock containing MOVABLE or free pages is considered
589 * removable
590 */
591 if (type != MIGRATE_MOVABLE && !pageblock_free(page))
592 return 0;
593
594 /*
595 * A pageblock starting with a PageReserved page is not
596 * considered removable.
597 */
598 if (PageReserved(page))
599 return 0;
600 }
601
602 /* All pageblocks in the memory block are likely to be hot-removable */
603 return 1;
604}
605
606/*
607 * Confirm all pages in a range [start, end) is belongs to the same zone.
608 */
609static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
610{
611 unsigned long pfn;
612 struct zone *zone = NULL;
613 struct page *page;
614 int i;
615 for (pfn = start_pfn;
616 pfn < end_pfn;
617 pfn += MAX_ORDER_NR_PAGES) {
618 i = 0;
619 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
620 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
621 i++;
622 if (i == MAX_ORDER_NR_PAGES)
623 continue;
624 page = pfn_to_page(pfn + i);
625 if (zone && page_zone(page) != zone)
626 return 0;
627 zone = page_zone(page);
628 }
629 return 1;
630}
631
632/*
633 * Scanning pfn is much easier than scanning lru list.
634 * Scan pfn from start to end and Find LRU page.
635 */
636int scan_lru_pages(unsigned long start, unsigned long end)
637{
638 unsigned long pfn;
639 struct page *page;
640 for (pfn = start; pfn < end; pfn++) {
641 if (pfn_valid(pfn)) {
642 page = pfn_to_page(pfn);
643 if (PageLRU(page))
644 return pfn;
645 }
646 }
647 return 0;
648}
649
650static struct page *
651hotremove_migrate_alloc(struct page *page, unsigned long private, int **x)
652{
653 /* This should be improooooved!! */
654 return alloc_page(GFP_HIGHUSER_MOVABLE);
655}
656
657#define NR_OFFLINE_AT_ONCE_PAGES (256)
658static int
659do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
660{
661 unsigned long pfn;
662 struct page *page;
663 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
664 int not_managed = 0;
665 int ret = 0;
666 LIST_HEAD(source);
667
668 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
669 if (!pfn_valid(pfn))
670 continue;
671 page = pfn_to_page(pfn);
672 if (!page_count(page))
673 continue;
674 /*
675 * We can skip free pages. And we can only deal with pages on
676 * LRU.
677 */
678 ret = isolate_lru_page(page);
679 if (!ret) { /* Success */
680 list_add_tail(&page->lru, &source);
681 move_pages--;
682 inc_zone_page_state(page, NR_ISOLATED_ANON +
683 page_is_file_cache(page));
684
685 } else {
686 /* Becasue we don't have big zone->lock. we should
687 check this again here. */
688 if (page_count(page))
689 not_managed++;
690#ifdef CONFIG_DEBUG_VM
691 printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
692 pfn);
693 dump_page(page);
694#endif
695 }
696 }
697 ret = -EBUSY;
698 if (not_managed) {
699 if (!list_empty(&source))
700 putback_lru_pages(&source);
701 goto out;
702 }
703 ret = 0;
704 if (list_empty(&source))
705 goto out;
706 /* this function returns # of failed pages */
707 ret = migrate_pages(&source, hotremove_migrate_alloc, 0, 1);
708
709out:
710 return ret;
711}
712
713/*
714 * remove from free_area[] and mark all as Reserved.
715 */
716static int
717offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
718 void *data)
719{
720 __offline_isolated_pages(start, start + nr_pages);
721 return 0;
722}
723
724static void
725offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
726{
727 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
728 offline_isolated_pages_cb);
729}
730
731/*
732 * Check all pages in range, recoreded as memory resource, are isolated.
733 */
734static int
735check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
736 void *data)
737{
738 int ret;
739 long offlined = *(long *)data;
740 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages);
741 offlined = nr_pages;
742 if (!ret)
743 *(long *)data += offlined;
744 return ret;
745}
746
747static long
748check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
749{
750 long offlined = 0;
751 int ret;
752
753 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
754 check_pages_isolated_cb);
755 if (ret < 0)
756 offlined = (long)ret;
757 return offlined;
758}
759
760static int offline_pages(unsigned long start_pfn,
761 unsigned long end_pfn, unsigned long timeout)
762{
763 unsigned long pfn, nr_pages, expire;
764 long offlined_pages;
765 int ret, drain, retry_max, node;
766 struct zone *zone;
767 struct memory_notify arg;
768
769 BUG_ON(start_pfn >= end_pfn);
770 /* at least, alignment against pageblock is necessary */
771 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
772 return -EINVAL;
773 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
774 return -EINVAL;
775 /* This makes hotplug much easier...and readable.
776 we assume this for now. .*/
777 if (!test_pages_in_a_zone(start_pfn, end_pfn))
778 return -EINVAL;
779
780 lock_system_sleep();
781
782 zone = page_zone(pfn_to_page(start_pfn));
783 node = zone_to_nid(zone);
784 nr_pages = end_pfn - start_pfn;
785
786 /* set above range as isolated */
787 ret = start_isolate_page_range(start_pfn, end_pfn);
788 if (ret)
789 goto out;
790
791 arg.start_pfn = start_pfn;
792 arg.nr_pages = nr_pages;
793 arg.status_change_nid = -1;
794 if (nr_pages >= node_present_pages(node))
795 arg.status_change_nid = node;
796
797 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
798 ret = notifier_to_errno(ret);
799 if (ret)
800 goto failed_removal;
801
802 pfn = start_pfn;
803 expire = jiffies + timeout;
804 drain = 0;
805 retry_max = 5;
806repeat:
807 /* start memory hot removal */
808 ret = -EAGAIN;
809 if (time_after(jiffies, expire))
810 goto failed_removal;
811 ret = -EINTR;
812 if (signal_pending(current))
813 goto failed_removal;
814 ret = 0;
815 if (drain) {
816 lru_add_drain_all();
817 flush_scheduled_work();
818 cond_resched();
819 drain_all_pages();
820 }
821
822 pfn = scan_lru_pages(start_pfn, end_pfn);
823 if (pfn) { /* We have page on LRU */
824 ret = do_migrate_range(pfn, end_pfn);
825 if (!ret) {
826 drain = 1;
827 goto repeat;
828 } else {
829 if (ret < 0)
830 if (--retry_max == 0)
831 goto failed_removal;
832 yield();
833 drain = 1;
834 goto repeat;
835 }
836 }
837 /* drain all zone's lru pagevec, this is asyncronous... */
838 lru_add_drain_all();
839 flush_scheduled_work();
840 yield();
841 /* drain pcp pages , this is synchrouns. */
842 drain_all_pages();
843 /* check again */
844 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
845 if (offlined_pages < 0) {
846 ret = -EBUSY;
847 goto failed_removal;
848 }
849 printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
850 /* Ok, all of our target is islaoted.
851 We cannot do rollback at this point. */
852 offline_isolated_pages(start_pfn, end_pfn);
853 /* reset pagetype flags and makes migrate type to be MOVABLE */
854 undo_isolate_page_range(start_pfn, end_pfn);
855 /* removal success */
856 zone->present_pages -= offlined_pages;
857 zone->zone_pgdat->node_present_pages -= offlined_pages;
858 totalram_pages -= offlined_pages;
859
860 setup_per_zone_wmarks();
861 calculate_zone_inactive_ratio(zone);
862 if (!node_present_pages(node)) {
863 node_clear_state(node, N_HIGH_MEMORY);
864 kswapd_stop(node);
865 }
866
867 vm_total_pages = nr_free_pagecache_pages();
868 writeback_set_ratelimit();
869
870 memory_notify(MEM_OFFLINE, &arg);
871 unlock_system_sleep();
872 return 0;
873
874failed_removal:
875 printk(KERN_INFO "memory offlining %lx to %lx failed\n",
876 start_pfn, end_pfn);
877 memory_notify(MEM_CANCEL_OFFLINE, &arg);
878 /* pushback to free area */
879 undo_isolate_page_range(start_pfn, end_pfn);
880
881out:
882 unlock_system_sleep();
883 return ret;
884}
885
886int remove_memory(u64 start, u64 size)
887{
888 unsigned long start_pfn, end_pfn;
889
890 start_pfn = PFN_DOWN(start);
891 end_pfn = start_pfn + PFN_DOWN(size);
892 return offline_pages(start_pfn, end_pfn, 120 * HZ);
893}
894#else
895int remove_memory(u64 start, u64 size)
896{
897 return -EINVAL;
898}
899#endif /* CONFIG_MEMORY_HOTREMOVE */
900EXPORT_SYMBOL_GPL(remove_memory);
Linux 6.x block layer and io_uring tree(s)