2 * linux/mm/memory_hotplug.c
7 #include <linux/stddef.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/export.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 #include <linux/stop_machine.h>
34 #include <asm/tlbflush.h>
39 * online_page_callback contains pointer to current page onlining function.
40 * Initially it is generic_online_page(). If it is required it could be
41 * changed by calling set_online_page_callback() for callback registration
42 * and restore_online_page_callback() for generic callback restore.
45 static void generic_online_page(struct page
*page
);
47 static online_page_callback_t online_page_callback
= generic_online_page
;
49 DEFINE_MUTEX(mem_hotplug_mutex
);
51 void lock_memory_hotplug(void)
53 mutex_lock(&mem_hotplug_mutex
);
55 /* for exclusive hibernation if CONFIG_HIBERNATION=y */
59 void unlock_memory_hotplug(void)
61 unlock_system_sleep();
62 mutex_unlock(&mem_hotplug_mutex
);
66 /* add this memory to iomem resource */
67 static struct resource
*register_memory_resource(u64 start
, u64 size
)
70 res
= kzalloc(sizeof(struct resource
), GFP_KERNEL
);
73 res
->name
= "System RAM";
75 res
->end
= start
+ size
- 1;
76 res
->flags
= IORESOURCE_MEM
| IORESOURCE_BUSY
;
77 if (request_resource(&iomem_resource
, res
) < 0) {
78 printk("System RAM resource %pR cannot be added\n", res
);
85 static void release_memory_resource(struct resource
*res
)
89 release_resource(res
);
94 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
95 void get_page_bootmem(unsigned long info
, struct page
*page
,
98 page
->lru
.next
= (struct list_head
*) type
;
100 set_page_private(page
, info
);
101 atomic_inc(&page
->_count
);
104 /* reference to __meminit __free_pages_bootmem is valid
105 * so use __ref to tell modpost not to generate a warning */
106 void __ref
put_page_bootmem(struct page
*page
)
109 static DEFINE_MUTEX(ppb_lock
);
111 type
= (unsigned long) page
->lru
.next
;
112 BUG_ON(type
< MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE
||
113 type
> MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE
);
115 if (atomic_dec_return(&page
->_count
) == 1) {
116 ClearPagePrivate(page
);
117 set_page_private(page
, 0);
118 INIT_LIST_HEAD(&page
->lru
);
121 * Please refer to comment for __free_pages_bootmem()
122 * for why we serialize here.
124 mutex_lock(&ppb_lock
);
125 __free_pages_bootmem(page
, 0);
126 mutex_unlock(&ppb_lock
);
132 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
133 #ifndef CONFIG_SPARSEMEM_VMEMMAP
134 static void register_page_bootmem_info_section(unsigned long start_pfn
)
136 unsigned long *usemap
, mapsize
, section_nr
, i
;
137 struct mem_section
*ms
;
138 struct page
*page
, *memmap
;
140 section_nr
= pfn_to_section_nr(start_pfn
);
141 ms
= __nr_to_section(section_nr
);
143 /* Get section's memmap address */
144 memmap
= sparse_decode_mem_map(ms
->section_mem_map
, section_nr
);
147 * Get page for the memmap's phys address
148 * XXX: need more consideration for sparse_vmemmap...
150 page
= virt_to_page(memmap
);
151 mapsize
= sizeof(struct page
) * PAGES_PER_SECTION
;
152 mapsize
= PAGE_ALIGN(mapsize
) >> PAGE_SHIFT
;
154 /* remember memmap's page */
155 for (i
= 0; i
< mapsize
; i
++, page
++)
156 get_page_bootmem(section_nr
, page
, SECTION_INFO
);
158 usemap
= __nr_to_section(section_nr
)->pageblock_flags
;
159 page
= virt_to_page(usemap
);
161 mapsize
= PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT
;
163 for (i
= 0; i
< mapsize
; i
++, page
++)
164 get_page_bootmem(section_nr
, page
, MIX_SECTION_INFO
);
167 #else /* CONFIG_SPARSEMEM_VMEMMAP */
168 static void register_page_bootmem_info_section(unsigned long start_pfn
)
170 unsigned long *usemap
, mapsize
, section_nr
, i
;
171 struct mem_section
*ms
;
172 struct page
*page
, *memmap
;
174 if (!pfn_valid(start_pfn
))
177 section_nr
= pfn_to_section_nr(start_pfn
);
178 ms
= __nr_to_section(section_nr
);
180 memmap
= sparse_decode_mem_map(ms
->section_mem_map
, section_nr
);
182 register_page_bootmem_memmap(section_nr
, memmap
, PAGES_PER_SECTION
);
184 usemap
= __nr_to_section(section_nr
)->pageblock_flags
;
185 page
= virt_to_page(usemap
);
187 mapsize
= PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT
;
189 for (i
= 0; i
< mapsize
; i
++, page
++)
190 get_page_bootmem(section_nr
, page
, MIX_SECTION_INFO
);
192 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
194 void register_page_bootmem_info_node(struct pglist_data
*pgdat
)
196 unsigned long i
, pfn
, end_pfn
, nr_pages
;
197 int node
= pgdat
->node_id
;
201 nr_pages
= PAGE_ALIGN(sizeof(struct pglist_data
)) >> PAGE_SHIFT
;
202 page
= virt_to_page(pgdat
);
204 for (i
= 0; i
< nr_pages
; i
++, page
++)
205 get_page_bootmem(node
, page
, NODE_INFO
);
207 zone
= &pgdat
->node_zones
[0];
208 for (; zone
< pgdat
->node_zones
+ MAX_NR_ZONES
- 1; zone
++) {
209 if (zone
->wait_table
) {
210 nr_pages
= zone
->wait_table_hash_nr_entries
211 * sizeof(wait_queue_head_t
);
212 nr_pages
= PAGE_ALIGN(nr_pages
) >> PAGE_SHIFT
;
213 page
= virt_to_page(zone
->wait_table
);
215 for (i
= 0; i
< nr_pages
; i
++, page
++)
216 get_page_bootmem(node
, page
, NODE_INFO
);
220 pfn
= pgdat
->node_start_pfn
;
221 end_pfn
= pfn
+ pgdat
->node_spanned_pages
;
223 /* register_section info */
224 for (; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
226 * Some platforms can assign the same pfn to multiple nodes - on
227 * node0 as well as nodeN. To avoid registering a pfn against
228 * multiple nodes we check that this pfn does not already
229 * reside in some other node.
231 if (pfn_valid(pfn
) && (pfn_to_nid(pfn
) == node
))
232 register_page_bootmem_info_section(pfn
);
235 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
237 static void grow_zone_span(struct zone
*zone
, unsigned long start_pfn
,
238 unsigned long end_pfn
)
240 unsigned long old_zone_end_pfn
;
242 zone_span_writelock(zone
);
244 old_zone_end_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
245 if (!zone
->spanned_pages
|| start_pfn
< zone
->zone_start_pfn
)
246 zone
->zone_start_pfn
= start_pfn
;
248 zone
->spanned_pages
= max(old_zone_end_pfn
, end_pfn
) -
249 zone
->zone_start_pfn
;
251 zone_span_writeunlock(zone
);
254 static void resize_zone(struct zone
*zone
, unsigned long start_pfn
,
255 unsigned long end_pfn
)
257 zone_span_writelock(zone
);
259 if (end_pfn
- start_pfn
) {
260 zone
->zone_start_pfn
= start_pfn
;
261 zone
->spanned_pages
= end_pfn
- start_pfn
;
264 * make it consist as free_area_init_core(),
265 * if spanned_pages = 0, then keep start_pfn = 0
267 zone
->zone_start_pfn
= 0;
268 zone
->spanned_pages
= 0;
271 zone_span_writeunlock(zone
);
274 static void fix_zone_id(struct zone
*zone
, unsigned long start_pfn
,
275 unsigned long end_pfn
)
277 enum zone_type zid
= zone_idx(zone
);
278 int nid
= zone
->zone_pgdat
->node_id
;
281 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
282 set_page_links(pfn_to_page(pfn
), zid
, nid
, pfn
);
285 static int __meminit
move_pfn_range_left(struct zone
*z1
, struct zone
*z2
,
286 unsigned long start_pfn
, unsigned long end_pfn
)
290 unsigned long z1_start_pfn
;
292 if (!z1
->wait_table
) {
293 ret
= init_currently_empty_zone(z1
, start_pfn
,
294 end_pfn
- start_pfn
, MEMMAP_HOTPLUG
);
299 pgdat_resize_lock(z1
->zone_pgdat
, &flags
);
301 /* can't move pfns which are higher than @z2 */
302 if (end_pfn
> z2
->zone_start_pfn
+ z2
->spanned_pages
)
304 /* the move out part mast at the left most of @z2 */
305 if (start_pfn
> z2
->zone_start_pfn
)
307 /* must included/overlap */
308 if (end_pfn
<= z2
->zone_start_pfn
)
311 /* use start_pfn for z1's start_pfn if z1 is empty */
312 if (z1
->spanned_pages
)
313 z1_start_pfn
= z1
->zone_start_pfn
;
315 z1_start_pfn
= start_pfn
;
317 resize_zone(z1
, z1_start_pfn
, end_pfn
);
318 resize_zone(z2
, end_pfn
, z2
->zone_start_pfn
+ z2
->spanned_pages
);
320 pgdat_resize_unlock(z1
->zone_pgdat
, &flags
);
322 fix_zone_id(z1
, start_pfn
, end_pfn
);
326 pgdat_resize_unlock(z1
->zone_pgdat
, &flags
);
330 static int __meminit
move_pfn_range_right(struct zone
*z1
, struct zone
*z2
,
331 unsigned long start_pfn
, unsigned long end_pfn
)
335 unsigned long z2_end_pfn
;
337 if (!z2
->wait_table
) {
338 ret
= init_currently_empty_zone(z2
, start_pfn
,
339 end_pfn
- start_pfn
, MEMMAP_HOTPLUG
);
344 pgdat_resize_lock(z1
->zone_pgdat
, &flags
);
346 /* can't move pfns which are lower than @z1 */
347 if (z1
->zone_start_pfn
> start_pfn
)
349 /* the move out part mast at the right most of @z1 */
350 if (z1
->zone_start_pfn
+ z1
->spanned_pages
> end_pfn
)
352 /* must included/overlap */
353 if (start_pfn
>= z1
->zone_start_pfn
+ z1
->spanned_pages
)
356 /* use end_pfn for z2's end_pfn if z2 is empty */
357 if (z2
->spanned_pages
)
358 z2_end_pfn
= z2
->zone_start_pfn
+ z2
->spanned_pages
;
360 z2_end_pfn
= end_pfn
;
362 resize_zone(z1
, z1
->zone_start_pfn
, start_pfn
);
363 resize_zone(z2
, start_pfn
, z2_end_pfn
);
365 pgdat_resize_unlock(z1
->zone_pgdat
, &flags
);
367 fix_zone_id(z2
, start_pfn
, end_pfn
);
371 pgdat_resize_unlock(z1
->zone_pgdat
, &flags
);
375 static void grow_pgdat_span(struct pglist_data
*pgdat
, unsigned long start_pfn
,
376 unsigned long end_pfn
)
378 unsigned long old_pgdat_end_pfn
=
379 pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
381 if (!pgdat
->node_spanned_pages
|| start_pfn
< pgdat
->node_start_pfn
)
382 pgdat
->node_start_pfn
= start_pfn
;
384 pgdat
->node_spanned_pages
= max(old_pgdat_end_pfn
, end_pfn
) -
385 pgdat
->node_start_pfn
;
388 static int __meminit
__add_zone(struct zone
*zone
, unsigned long phys_start_pfn
)
390 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
391 int nr_pages
= PAGES_PER_SECTION
;
392 int nid
= pgdat
->node_id
;
396 zone_type
= zone
- pgdat
->node_zones
;
397 if (!zone
->wait_table
) {
400 ret
= init_currently_empty_zone(zone
, phys_start_pfn
,
401 nr_pages
, MEMMAP_HOTPLUG
);
405 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
406 grow_zone_span(zone
, phys_start_pfn
, phys_start_pfn
+ nr_pages
);
407 grow_pgdat_span(zone
->zone_pgdat
, phys_start_pfn
,
408 phys_start_pfn
+ nr_pages
);
409 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
410 memmap_init_zone(nr_pages
, nid
, zone_type
,
411 phys_start_pfn
, MEMMAP_HOTPLUG
);
415 static int __meminit
__add_section(int nid
, struct zone
*zone
,
416 unsigned long phys_start_pfn
)
418 int nr_pages
= PAGES_PER_SECTION
;
421 if (pfn_valid(phys_start_pfn
))
424 ret
= sparse_add_one_section(zone
, phys_start_pfn
, nr_pages
);
429 ret
= __add_zone(zone
, phys_start_pfn
);
434 return register_new_memory(nid
, __pfn_to_section(phys_start_pfn
));
437 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
438 static int find_smallest_section_pfn(int nid
, struct zone
*zone
,
439 unsigned long start_pfn
,
440 unsigned long end_pfn
)
442 struct mem_section
*ms
;
444 for (; start_pfn
< end_pfn
; start_pfn
+= PAGES_PER_SECTION
) {
445 ms
= __pfn_to_section(start_pfn
);
447 if (unlikely(!valid_section(ms
)))
450 if (unlikely(pfn_to_nid(start_pfn
) != nid
))
453 if (zone
&& zone
!= page_zone(pfn_to_page(start_pfn
)))
462 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
463 static int find_biggest_section_pfn(int nid
, struct zone
*zone
,
464 unsigned long start_pfn
,
465 unsigned long end_pfn
)
467 struct mem_section
*ms
;
470 /* pfn is the end pfn of a memory section. */
472 for (; pfn
>= start_pfn
; pfn
-= PAGES_PER_SECTION
) {
473 ms
= __pfn_to_section(pfn
);
475 if (unlikely(!valid_section(ms
)))
478 if (unlikely(pfn_to_nid(pfn
) != nid
))
481 if (zone
&& zone
!= page_zone(pfn_to_page(pfn
)))
490 static void shrink_zone_span(struct zone
*zone
, unsigned long start_pfn
,
491 unsigned long end_pfn
)
493 unsigned long zone_start_pfn
= zone
->zone_start_pfn
;
494 unsigned long zone_end_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
496 struct mem_section
*ms
;
497 int nid
= zone_to_nid(zone
);
499 zone_span_writelock(zone
);
500 if (zone_start_pfn
== start_pfn
) {
502 * If the section is smallest section in the zone, it need
503 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
504 * In this case, we find second smallest valid mem_section
505 * for shrinking zone.
507 pfn
= find_smallest_section_pfn(nid
, zone
, end_pfn
,
510 zone
->zone_start_pfn
= pfn
;
511 zone
->spanned_pages
= zone_end_pfn
- pfn
;
513 } else if (zone_end_pfn
== end_pfn
) {
515 * If the section is biggest section in the zone, it need
516 * shrink zone->spanned_pages.
517 * In this case, we find second biggest valid mem_section for
520 pfn
= find_biggest_section_pfn(nid
, zone
, zone_start_pfn
,
523 zone
->spanned_pages
= pfn
- zone_start_pfn
+ 1;
527 * The section is not biggest or smallest mem_section in the zone, it
528 * only creates a hole in the zone. So in this case, we need not
529 * change the zone. But perhaps, the zone has only hole data. Thus
530 * it check the zone has only hole or not.
532 pfn
= zone_start_pfn
;
533 for (; pfn
< zone_end_pfn
; pfn
+= PAGES_PER_SECTION
) {
534 ms
= __pfn_to_section(pfn
);
536 if (unlikely(!valid_section(ms
)))
539 if (page_zone(pfn_to_page(pfn
)) != zone
)
542 /* If the section is current section, it continues the loop */
543 if (start_pfn
== pfn
)
546 /* If we find valid section, we have nothing to do */
547 zone_span_writeunlock(zone
);
551 /* The zone has no valid section */
552 zone
->zone_start_pfn
= 0;
553 zone
->spanned_pages
= 0;
554 zone_span_writeunlock(zone
);
557 static void shrink_pgdat_span(struct pglist_data
*pgdat
,
558 unsigned long start_pfn
, unsigned long end_pfn
)
560 unsigned long pgdat_start_pfn
= pgdat
->node_start_pfn
;
561 unsigned long pgdat_end_pfn
=
562 pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
564 struct mem_section
*ms
;
565 int nid
= pgdat
->node_id
;
567 if (pgdat_start_pfn
== start_pfn
) {
569 * If the section is smallest section in the pgdat, it need
570 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
571 * In this case, we find second smallest valid mem_section
572 * for shrinking zone.
574 pfn
= find_smallest_section_pfn(nid
, NULL
, end_pfn
,
577 pgdat
->node_start_pfn
= pfn
;
578 pgdat
->node_spanned_pages
= pgdat_end_pfn
- pfn
;
580 } else if (pgdat_end_pfn
== end_pfn
) {
582 * If the section is biggest section in the pgdat, it need
583 * shrink pgdat->node_spanned_pages.
584 * In this case, we find second biggest valid mem_section for
587 pfn
= find_biggest_section_pfn(nid
, NULL
, pgdat_start_pfn
,
590 pgdat
->node_spanned_pages
= pfn
- pgdat_start_pfn
+ 1;
594 * If the section is not biggest or smallest mem_section in the pgdat,
595 * it only creates a hole in the pgdat. So in this case, we need not
597 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
598 * has only hole or not.
600 pfn
= pgdat_start_pfn
;
601 for (; pfn
< pgdat_end_pfn
; pfn
+= PAGES_PER_SECTION
) {
602 ms
= __pfn_to_section(pfn
);
604 if (unlikely(!valid_section(ms
)))
607 if (pfn_to_nid(pfn
) != nid
)
610 /* If the section is current section, it continues the loop */
611 if (start_pfn
== pfn
)
614 /* If we find valid section, we have nothing to do */
618 /* The pgdat has no valid section */
619 pgdat
->node_start_pfn
= 0;
620 pgdat
->node_spanned_pages
= 0;
623 static void __remove_zone(struct zone
*zone
, unsigned long start_pfn
)
625 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
626 int nr_pages
= PAGES_PER_SECTION
;
630 zone_type
= zone
- pgdat
->node_zones
;
632 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
633 shrink_zone_span(zone
, start_pfn
, start_pfn
+ nr_pages
);
634 shrink_pgdat_span(pgdat
, start_pfn
, start_pfn
+ nr_pages
);
635 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
638 static int __remove_section(struct zone
*zone
, struct mem_section
*ms
)
640 unsigned long start_pfn
;
644 if (!valid_section(ms
))
647 ret
= unregister_memory_section(ms
);
651 scn_nr
= __section_nr(ms
);
652 start_pfn
= section_nr_to_pfn(scn_nr
);
653 __remove_zone(zone
, start_pfn
);
655 sparse_remove_one_section(zone
, ms
);
660 * Reasonably generic function for adding memory. It is
661 * expected that archs that support memory hotplug will
662 * call this function after deciding the zone to which to
665 int __ref
__add_pages(int nid
, struct zone
*zone
, unsigned long phys_start_pfn
,
666 unsigned long nr_pages
)
670 int start_sec
, end_sec
;
671 /* during initialize mem_map, align hot-added range to section */
672 start_sec
= pfn_to_section_nr(phys_start_pfn
);
673 end_sec
= pfn_to_section_nr(phys_start_pfn
+ nr_pages
- 1);
675 for (i
= start_sec
; i
<= end_sec
; i
++) {
676 err
= __add_section(nid
, zone
, i
<< PFN_SECTION_SHIFT
);
679 * EEXIST is finally dealt with by ioresource collision
680 * check. see add_memory() => register_memory_resource()
681 * Warning will be printed if there is collision.
683 if (err
&& (err
!= -EEXIST
))
690 EXPORT_SYMBOL_GPL(__add_pages
);
693 * __remove_pages() - remove sections of pages from a zone
694 * @zone: zone from which pages need to be removed
695 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
696 * @nr_pages: number of pages to remove (must be multiple of section size)
698 * Generic helper function to remove section mappings and sysfs entries
699 * for the section of the memory we are removing. Caller needs to make
700 * sure that pages are marked reserved and zones are adjust properly by
701 * calling offline_pages().
703 int __remove_pages(struct zone
*zone
, unsigned long phys_start_pfn
,
704 unsigned long nr_pages
)
706 unsigned long i
, ret
= 0;
707 int sections_to_remove
;
710 * We can only remove entire sections
712 BUG_ON(phys_start_pfn
& ~PAGE_SECTION_MASK
);
713 BUG_ON(nr_pages
% PAGES_PER_SECTION
);
715 release_mem_region(phys_start_pfn
<< PAGE_SHIFT
, nr_pages
* PAGE_SIZE
);
717 sections_to_remove
= nr_pages
/ PAGES_PER_SECTION
;
718 for (i
= 0; i
< sections_to_remove
; i
++) {
719 unsigned long pfn
= phys_start_pfn
+ i
*PAGES_PER_SECTION
;
720 ret
= __remove_section(zone
, __pfn_to_section(pfn
));
726 EXPORT_SYMBOL_GPL(__remove_pages
);
728 int set_online_page_callback(online_page_callback_t callback
)
732 lock_memory_hotplug();
734 if (online_page_callback
== generic_online_page
) {
735 online_page_callback
= callback
;
739 unlock_memory_hotplug();
743 EXPORT_SYMBOL_GPL(set_online_page_callback
);
745 int restore_online_page_callback(online_page_callback_t callback
)
749 lock_memory_hotplug();
751 if (online_page_callback
== callback
) {
752 online_page_callback
= generic_online_page
;
756 unlock_memory_hotplug();
760 EXPORT_SYMBOL_GPL(restore_online_page_callback
);
762 void __online_page_set_limits(struct page
*page
)
764 unsigned long pfn
= page_to_pfn(page
);
766 if (pfn
>= num_physpages
)
767 num_physpages
= pfn
+ 1;
769 EXPORT_SYMBOL_GPL(__online_page_set_limits
);
771 void __online_page_increment_counters(struct page
*page
)
775 #ifdef CONFIG_HIGHMEM
776 if (PageHighMem(page
))
780 EXPORT_SYMBOL_GPL(__online_page_increment_counters
);
782 void __online_page_free(struct page
*page
)
784 ClearPageReserved(page
);
785 init_page_count(page
);
788 EXPORT_SYMBOL_GPL(__online_page_free
);
790 static void generic_online_page(struct page
*page
)
792 __online_page_set_limits(page
);
793 __online_page_increment_counters(page
);
794 __online_page_free(page
);
797 static int online_pages_range(unsigned long start_pfn
, unsigned long nr_pages
,
801 unsigned long onlined_pages
= *(unsigned long *)arg
;
803 if (PageReserved(pfn_to_page(start_pfn
)))
804 for (i
= 0; i
< nr_pages
; i
++) {
805 page
= pfn_to_page(start_pfn
+ i
);
806 (*online_page_callback
)(page
);
809 *(unsigned long *)arg
= onlined_pages
;
813 #ifdef CONFIG_MOVABLE_NODE
815 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
818 static bool can_online_high_movable(struct zone
*zone
)
822 #else /* CONFIG_MOVABLE_NODE */
823 /* ensure every online node has NORMAL memory */
824 static bool can_online_high_movable(struct zone
*zone
)
826 return node_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
828 #endif /* CONFIG_MOVABLE_NODE */
830 /* check which state of node_states will be changed when online memory */
831 static void node_states_check_changes_online(unsigned long nr_pages
,
832 struct zone
*zone
, struct memory_notify
*arg
)
834 int nid
= zone_to_nid(zone
);
835 enum zone_type zone_last
= ZONE_NORMAL
;
838 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
839 * contains nodes which have zones of 0...ZONE_NORMAL,
840 * set zone_last to ZONE_NORMAL.
842 * If we don't have HIGHMEM nor movable node,
843 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
844 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
846 if (N_MEMORY
== N_NORMAL_MEMORY
)
847 zone_last
= ZONE_MOVABLE
;
850 * if the memory to be online is in a zone of 0...zone_last, and
851 * the zones of 0...zone_last don't have memory before online, we will
852 * need to set the node to node_states[N_NORMAL_MEMORY] after
853 * the memory is online.
855 if (zone_idx(zone
) <= zone_last
&& !node_state(nid
, N_NORMAL_MEMORY
))
856 arg
->status_change_nid_normal
= nid
;
858 arg
->status_change_nid_normal
= -1;
860 #ifdef CONFIG_HIGHMEM
862 * If we have movable node, node_states[N_HIGH_MEMORY]
863 * contains nodes which have zones of 0...ZONE_HIGHMEM,
864 * set zone_last to ZONE_HIGHMEM.
866 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
867 * contains nodes which have zones of 0...ZONE_MOVABLE,
868 * set zone_last to ZONE_MOVABLE.
870 zone_last
= ZONE_HIGHMEM
;
871 if (N_MEMORY
== N_HIGH_MEMORY
)
872 zone_last
= ZONE_MOVABLE
;
874 if (zone_idx(zone
) <= zone_last
&& !node_state(nid
, N_HIGH_MEMORY
))
875 arg
->status_change_nid_high
= nid
;
877 arg
->status_change_nid_high
= -1;
879 arg
->status_change_nid_high
= arg
->status_change_nid_normal
;
883 * if the node don't have memory befor online, we will need to
884 * set the node to node_states[N_MEMORY] after the memory
887 if (!node_state(nid
, N_MEMORY
))
888 arg
->status_change_nid
= nid
;
890 arg
->status_change_nid
= -1;
893 static void node_states_set_node(int node
, struct memory_notify
*arg
)
895 if (arg
->status_change_nid_normal
>= 0)
896 node_set_state(node
, N_NORMAL_MEMORY
);
898 if (arg
->status_change_nid_high
>= 0)
899 node_set_state(node
, N_HIGH_MEMORY
);
901 node_set_state(node
, N_MEMORY
);
905 int __ref
online_pages(unsigned long pfn
, unsigned long nr_pages
, int online_type
)
907 unsigned long onlined_pages
= 0;
909 int need_zonelists_rebuild
= 0;
912 struct memory_notify arg
;
914 lock_memory_hotplug();
916 * This doesn't need a lock to do pfn_to_page().
917 * The section can't be removed here because of the
918 * memory_block->state_mutex.
920 zone
= page_zone(pfn_to_page(pfn
));
922 if ((zone_idx(zone
) > ZONE_NORMAL
|| online_type
== ONLINE_MOVABLE
) &&
923 !can_online_high_movable(zone
)) {
924 unlock_memory_hotplug();
928 if (online_type
== ONLINE_KERNEL
&& zone_idx(zone
) == ZONE_MOVABLE
) {
929 if (move_pfn_range_left(zone
- 1, zone
, pfn
, pfn
+ nr_pages
)) {
930 unlock_memory_hotplug();
934 if (online_type
== ONLINE_MOVABLE
&& zone_idx(zone
) == ZONE_MOVABLE
- 1) {
935 if (move_pfn_range_right(zone
, zone
+ 1, pfn
, pfn
+ nr_pages
)) {
936 unlock_memory_hotplug();
941 /* Previous code may changed the zone of the pfn range */
942 zone
= page_zone(pfn_to_page(pfn
));
945 arg
.nr_pages
= nr_pages
;
946 node_states_check_changes_online(nr_pages
, zone
, &arg
);
948 nid
= page_to_nid(pfn_to_page(pfn
));
950 ret
= memory_notify(MEM_GOING_ONLINE
, &arg
);
951 ret
= notifier_to_errno(ret
);
953 memory_notify(MEM_CANCEL_ONLINE
, &arg
);
954 unlock_memory_hotplug();
958 * If this zone is not populated, then it is not in zonelist.
959 * This means the page allocator ignores this zone.
960 * So, zonelist must be updated after online.
962 mutex_lock(&zonelists_mutex
);
963 if (!populated_zone(zone
)) {
964 need_zonelists_rebuild
= 1;
965 build_all_zonelists(NULL
, zone
);
968 ret
= walk_system_ram_range(pfn
, nr_pages
, &onlined_pages
,
971 if (need_zonelists_rebuild
)
972 zone_pcp_reset(zone
);
973 mutex_unlock(&zonelists_mutex
);
974 printk(KERN_DEBUG
"online_pages [mem %#010llx-%#010llx] failed\n",
975 (unsigned long long) pfn
<< PAGE_SHIFT
,
976 (((unsigned long long) pfn
+ nr_pages
)
978 memory_notify(MEM_CANCEL_ONLINE
, &arg
);
979 unlock_memory_hotplug();
983 zone
->managed_pages
+= onlined_pages
;
984 zone
->present_pages
+= onlined_pages
;
985 zone
->zone_pgdat
->node_present_pages
+= onlined_pages
;
987 node_states_set_node(zone_to_nid(zone
), &arg
);
988 if (need_zonelists_rebuild
)
989 build_all_zonelists(NULL
, NULL
);
991 zone_pcp_update(zone
);
994 mutex_unlock(&zonelists_mutex
);
996 init_per_zone_wmark_min();
999 kswapd_run(zone_to_nid(zone
));
1001 vm_total_pages
= nr_free_pagecache_pages();
1003 writeback_set_ratelimit();
1006 memory_notify(MEM_ONLINE
, &arg
);
1007 unlock_memory_hotplug();
1011 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1013 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1014 static pg_data_t __ref
*hotadd_new_pgdat(int nid
, u64 start
)
1016 struct pglist_data
*pgdat
;
1017 unsigned long zones_size
[MAX_NR_ZONES
] = {0};
1018 unsigned long zholes_size
[MAX_NR_ZONES
] = {0};
1019 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
1021 pgdat
= NODE_DATA(nid
);
1023 pgdat
= arch_alloc_nodedata(nid
);
1027 arch_refresh_nodedata(nid
, pgdat
);
1030 /* we can use NODE_DATA(nid) from here */
1032 /* init node's zones as empty zones, we don't have any present pages.*/
1033 free_area_init_node(nid
, zones_size
, start_pfn
, zholes_size
);
1036 * The node we allocated has no zone fallback lists. For avoiding
1037 * to access not-initialized zonelist, build here.
1039 mutex_lock(&zonelists_mutex
);
1040 build_all_zonelists(pgdat
, NULL
);
1041 mutex_unlock(&zonelists_mutex
);
1046 static void rollback_node_hotadd(int nid
, pg_data_t
*pgdat
)
1048 arch_refresh_nodedata(nid
, NULL
);
1049 arch_free_nodedata(pgdat
);
1055 * called by cpu_up() to online a node without onlined memory.
1057 int mem_online_node(int nid
)
1062 lock_memory_hotplug();
1063 pgdat
= hotadd_new_pgdat(nid
, 0);
1068 node_set_online(nid
);
1069 ret
= register_one_node(nid
);
1073 unlock_memory_hotplug();
1077 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1078 int __ref
add_memory(int nid
, u64 start
, u64 size
)
1080 pg_data_t
*pgdat
= NULL
;
1083 struct resource
*res
;
1086 lock_memory_hotplug();
1088 res
= register_memory_resource(start
, size
);
1093 { /* Stupid hack to suppress address-never-null warning */
1094 void *p
= NODE_DATA(nid
);
1097 new_node
= !node_online(nid
);
1099 pgdat
= hotadd_new_pgdat(nid
, start
);
1105 /* call arch's memory hotadd */
1106 ret
= arch_add_memory(nid
, start
, size
);
1111 /* we online node here. we can't roll back from here. */
1112 node_set_online(nid
);
1115 ret
= register_one_node(nid
);
1117 * If sysfs file of new node can't create, cpu on the node
1118 * can't be hot-added. There is no rollback way now.
1119 * So, check by BUG_ON() to catch it reluctantly..
1124 /* create new memmap entry */
1125 firmware_map_add_hotplug(start
, start
+ size
, "System RAM");
1130 /* rollback pgdat allocation and others */
1132 rollback_node_hotadd(nid
, pgdat
);
1133 release_memory_resource(res
);
1136 unlock_memory_hotplug();
1139 EXPORT_SYMBOL_GPL(add_memory
);
1141 #ifdef CONFIG_MEMORY_HOTREMOVE
1143 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1144 * set and the size of the free page is given by page_order(). Using this,
1145 * the function determines if the pageblock contains only free pages.
1146 * Due to buddy contraints, a free page at least the size of a pageblock will
1147 * be located at the start of the pageblock
1149 static inline int pageblock_free(struct page
*page
)
1151 return PageBuddy(page
) && page_order(page
) >= pageblock_order
;
1154 /* Return the start of the next active pageblock after a given page */
1155 static struct page
*next_active_pageblock(struct page
*page
)
1157 /* Ensure the starting page is pageblock-aligned */
1158 BUG_ON(page_to_pfn(page
) & (pageblock_nr_pages
- 1));
1160 /* If the entire pageblock is free, move to the end of free page */
1161 if (pageblock_free(page
)) {
1163 /* be careful. we don't have locks, page_order can be changed.*/
1164 order
= page_order(page
);
1165 if ((order
< MAX_ORDER
) && (order
>= pageblock_order
))
1166 return page
+ (1 << order
);
1169 return page
+ pageblock_nr_pages
;
1172 /* Checks if this range of memory is likely to be hot-removable. */
1173 int is_mem_section_removable(unsigned long start_pfn
, unsigned long nr_pages
)
1175 struct page
*page
= pfn_to_page(start_pfn
);
1176 struct page
*end_page
= page
+ nr_pages
;
1178 /* Check the starting page of each pageblock within the range */
1179 for (; page
< end_page
; page
= next_active_pageblock(page
)) {
1180 if (!is_pageblock_removable_nolock(page
))
1185 /* All pageblocks in the memory block are likely to be hot-removable */
1190 * Confirm all pages in a range [start, end) is belongs to the same zone.
1192 static int test_pages_in_a_zone(unsigned long start_pfn
, unsigned long end_pfn
)
1195 struct zone
*zone
= NULL
;
1198 for (pfn
= start_pfn
;
1200 pfn
+= MAX_ORDER_NR_PAGES
) {
1202 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1203 while ((i
< MAX_ORDER_NR_PAGES
) && !pfn_valid_within(pfn
+ i
))
1205 if (i
== MAX_ORDER_NR_PAGES
)
1207 page
= pfn_to_page(pfn
+ i
);
1208 if (zone
&& page_zone(page
) != zone
)
1210 zone
= page_zone(page
);
1216 * Scanning pfn is much easier than scanning lru list.
1217 * Scan pfn from start to end and Find LRU page.
1219 static unsigned long scan_lru_pages(unsigned long start
, unsigned long end
)
1223 for (pfn
= start
; pfn
< end
; pfn
++) {
1224 if (pfn_valid(pfn
)) {
1225 page
= pfn_to_page(pfn
);
1233 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1235 do_migrate_range(unsigned long start_pfn
, unsigned long end_pfn
)
1239 int move_pages
= NR_OFFLINE_AT_ONCE_PAGES
;
1240 int not_managed
= 0;
1244 for (pfn
= start_pfn
; pfn
< end_pfn
&& move_pages
> 0; pfn
++) {
1245 if (!pfn_valid(pfn
))
1247 page
= pfn_to_page(pfn
);
1248 if (!get_page_unless_zero(page
))
1251 * We can skip free pages. And we can only deal with pages on
1254 ret
= isolate_lru_page(page
);
1255 if (!ret
) { /* Success */
1257 list_add_tail(&page
->lru
, &source
);
1259 inc_zone_page_state(page
, NR_ISOLATED_ANON
+
1260 page_is_file_cache(page
));
1263 #ifdef CONFIG_DEBUG_VM
1264 printk(KERN_ALERT
"removing pfn %lx from LRU failed\n",
1269 /* Because we don't have big zone->lock. we should
1270 check this again here. */
1271 if (page_count(page
)) {
1278 if (!list_empty(&source
)) {
1280 putback_lru_pages(&source
);
1285 * alloc_migrate_target should be improooooved!!
1286 * migrate_pages returns # of failed pages.
1288 ret
= migrate_pages(&source
, alloc_migrate_target
, 0,
1292 putback_lru_pages(&source
);
1299 * remove from free_area[] and mark all as Reserved.
1302 offline_isolated_pages_cb(unsigned long start
, unsigned long nr_pages
,
1305 __offline_isolated_pages(start
, start
+ nr_pages
);
1310 offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
1312 walk_system_ram_range(start_pfn
, end_pfn
- start_pfn
, NULL
,
1313 offline_isolated_pages_cb
);
1317 * Check all pages in range, recoreded as memory resource, are isolated.
1320 check_pages_isolated_cb(unsigned long start_pfn
, unsigned long nr_pages
,
1324 long offlined
= *(long *)data
;
1325 ret
= test_pages_isolated(start_pfn
, start_pfn
+ nr_pages
, true);
1326 offlined
= nr_pages
;
1328 *(long *)data
+= offlined
;
1333 check_pages_isolated(unsigned long start_pfn
, unsigned long end_pfn
)
1338 ret
= walk_system_ram_range(start_pfn
, end_pfn
- start_pfn
, &offlined
,
1339 check_pages_isolated_cb
);
1341 offlined
= (long)ret
;
1345 #ifdef CONFIG_MOVABLE_NODE
1347 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1350 static bool can_offline_normal(struct zone
*zone
, unsigned long nr_pages
)
1354 #else /* CONFIG_MOVABLE_NODE */
1355 /* ensure the node has NORMAL memory if it is still online */
1356 static bool can_offline_normal(struct zone
*zone
, unsigned long nr_pages
)
1358 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1359 unsigned long present_pages
= 0;
1362 for (zt
= 0; zt
<= ZONE_NORMAL
; zt
++)
1363 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1365 if (present_pages
> nr_pages
)
1369 for (; zt
<= ZONE_MOVABLE
; zt
++)
1370 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1373 * we can't offline the last normal memory until all
1374 * higher memory is offlined.
1376 return present_pages
== 0;
1378 #endif /* CONFIG_MOVABLE_NODE */
1380 /* check which state of node_states will be changed when offline memory */
1381 static void node_states_check_changes_offline(unsigned long nr_pages
,
1382 struct zone
*zone
, struct memory_notify
*arg
)
1384 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1385 unsigned long present_pages
= 0;
1386 enum zone_type zt
, zone_last
= ZONE_NORMAL
;
1389 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1390 * contains nodes which have zones of 0...ZONE_NORMAL,
1391 * set zone_last to ZONE_NORMAL.
1393 * If we don't have HIGHMEM nor movable node,
1394 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1395 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1397 if (N_MEMORY
== N_NORMAL_MEMORY
)
1398 zone_last
= ZONE_MOVABLE
;
1401 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1402 * If the memory to be offline is in a zone of 0...zone_last,
1403 * and it is the last present memory, 0...zone_last will
1404 * become empty after offline , thus we can determind we will
1405 * need to clear the node from node_states[N_NORMAL_MEMORY].
1407 for (zt
= 0; zt
<= zone_last
; zt
++)
1408 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1409 if (zone_idx(zone
) <= zone_last
&& nr_pages
>= present_pages
)
1410 arg
->status_change_nid_normal
= zone_to_nid(zone
);
1412 arg
->status_change_nid_normal
= -1;
1414 #ifdef CONFIG_HIGHMEM
1416 * If we have movable node, node_states[N_HIGH_MEMORY]
1417 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1418 * set zone_last to ZONE_HIGHMEM.
1420 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1421 * contains nodes which have zones of 0...ZONE_MOVABLE,
1422 * set zone_last to ZONE_MOVABLE.
1424 zone_last
= ZONE_HIGHMEM
;
1425 if (N_MEMORY
== N_HIGH_MEMORY
)
1426 zone_last
= ZONE_MOVABLE
;
1428 for (; zt
<= zone_last
; zt
++)
1429 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1430 if (zone_idx(zone
) <= zone_last
&& nr_pages
>= present_pages
)
1431 arg
->status_change_nid_high
= zone_to_nid(zone
);
1433 arg
->status_change_nid_high
= -1;
1435 arg
->status_change_nid_high
= arg
->status_change_nid_normal
;
1439 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1441 zone_last
= ZONE_MOVABLE
;
1444 * check whether node_states[N_HIGH_MEMORY] will be changed
1445 * If we try to offline the last present @nr_pages from the node,
1446 * we can determind we will need to clear the node from
1447 * node_states[N_HIGH_MEMORY].
1449 for (; zt
<= zone_last
; zt
++)
1450 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1451 if (nr_pages
>= present_pages
)
1452 arg
->status_change_nid
= zone_to_nid(zone
);
1454 arg
->status_change_nid
= -1;
1457 static void node_states_clear_node(int node
, struct memory_notify
*arg
)
1459 if (arg
->status_change_nid_normal
>= 0)
1460 node_clear_state(node
, N_NORMAL_MEMORY
);
1462 if ((N_MEMORY
!= N_NORMAL_MEMORY
) &&
1463 (arg
->status_change_nid_high
>= 0))
1464 node_clear_state(node
, N_HIGH_MEMORY
);
1466 if ((N_MEMORY
!= N_HIGH_MEMORY
) &&
1467 (arg
->status_change_nid
>= 0))
1468 node_clear_state(node
, N_MEMORY
);
1471 static int __ref
__offline_pages(unsigned long start_pfn
,
1472 unsigned long end_pfn
, unsigned long timeout
)
1474 unsigned long pfn
, nr_pages
, expire
;
1475 long offlined_pages
;
1476 int ret
, drain
, retry_max
, node
;
1478 struct memory_notify arg
;
1480 BUG_ON(start_pfn
>= end_pfn
);
1481 /* at least, alignment against pageblock is necessary */
1482 if (!IS_ALIGNED(start_pfn
, pageblock_nr_pages
))
1484 if (!IS_ALIGNED(end_pfn
, pageblock_nr_pages
))
1486 /* This makes hotplug much easier...and readable.
1487 we assume this for now. .*/
1488 if (!test_pages_in_a_zone(start_pfn
, end_pfn
))
1491 lock_memory_hotplug();
1493 zone
= page_zone(pfn_to_page(start_pfn
));
1494 node
= zone_to_nid(zone
);
1495 nr_pages
= end_pfn
- start_pfn
;
1498 if (zone_idx(zone
) <= ZONE_NORMAL
&& !can_offline_normal(zone
, nr_pages
))
1501 /* set above range as isolated */
1502 ret
= start_isolate_page_range(start_pfn
, end_pfn
,
1503 MIGRATE_MOVABLE
, true);
1507 arg
.start_pfn
= start_pfn
;
1508 arg
.nr_pages
= nr_pages
;
1509 node_states_check_changes_offline(nr_pages
, zone
, &arg
);
1511 ret
= memory_notify(MEM_GOING_OFFLINE
, &arg
);
1512 ret
= notifier_to_errno(ret
);
1514 goto failed_removal
;
1517 expire
= jiffies
+ timeout
;
1521 /* start memory hot removal */
1523 if (time_after(jiffies
, expire
))
1524 goto failed_removal
;
1526 if (signal_pending(current
))
1527 goto failed_removal
;
1530 lru_add_drain_all();
1535 pfn
= scan_lru_pages(start_pfn
, end_pfn
);
1536 if (pfn
) { /* We have page on LRU */
1537 ret
= do_migrate_range(pfn
, end_pfn
);
1543 if (--retry_max
== 0)
1544 goto failed_removal
;
1550 /* drain all zone's lru pagevec, this is asynchronous... */
1551 lru_add_drain_all();
1553 /* drain pcp pages, this is synchronous. */
1556 offlined_pages
= check_pages_isolated(start_pfn
, end_pfn
);
1557 if (offlined_pages
< 0) {
1559 goto failed_removal
;
1561 printk(KERN_INFO
"Offlined Pages %ld\n", offlined_pages
);
1562 /* Ok, all of our target is isolated.
1563 We cannot do rollback at this point. */
1564 offline_isolated_pages(start_pfn
, end_pfn
);
1565 /* reset pagetype flags and makes migrate type to be MOVABLE */
1566 undo_isolate_page_range(start_pfn
, end_pfn
, MIGRATE_MOVABLE
);
1567 /* removal success */
1568 zone
->managed_pages
-= offlined_pages
;
1569 zone
->present_pages
-= offlined_pages
;
1570 zone
->zone_pgdat
->node_present_pages
-= offlined_pages
;
1571 totalram_pages
-= offlined_pages
;
1573 init_per_zone_wmark_min();
1575 if (!populated_zone(zone
)) {
1576 zone_pcp_reset(zone
);
1577 mutex_lock(&zonelists_mutex
);
1578 build_all_zonelists(NULL
, NULL
);
1579 mutex_unlock(&zonelists_mutex
);
1581 zone_pcp_update(zone
);
1583 node_states_clear_node(node
, &arg
);
1584 if (arg
.status_change_nid
>= 0)
1587 vm_total_pages
= nr_free_pagecache_pages();
1588 writeback_set_ratelimit();
1590 memory_notify(MEM_OFFLINE
, &arg
);
1591 unlock_memory_hotplug();
1595 printk(KERN_INFO
"memory offlining [mem %#010llx-%#010llx] failed\n",
1596 (unsigned long long) start_pfn
<< PAGE_SHIFT
,
1597 ((unsigned long long) end_pfn
<< PAGE_SHIFT
) - 1);
1598 memory_notify(MEM_CANCEL_OFFLINE
, &arg
);
1599 /* pushback to free area */
1600 undo_isolate_page_range(start_pfn
, end_pfn
, MIGRATE_MOVABLE
);
1603 unlock_memory_hotplug();
1607 int offline_pages(unsigned long start_pfn
, unsigned long nr_pages
)
1609 return __offline_pages(start_pfn
, start_pfn
+ nr_pages
, 120 * HZ
);
1613 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1614 * @start_pfn: start pfn of the memory range
1615 * @end_pfn: end pft of the memory range
1616 * @arg: argument passed to func
1617 * @func: callback for each memory section walked
1619 * This function walks through all present mem sections in range
1620 * [start_pfn, end_pfn) and call func on each mem section.
1622 * Returns the return value of func.
1624 static int walk_memory_range(unsigned long start_pfn
, unsigned long end_pfn
,
1625 void *arg
, int (*func
)(struct memory_block
*, void *))
1627 struct memory_block
*mem
= NULL
;
1628 struct mem_section
*section
;
1629 unsigned long pfn
, section_nr
;
1632 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
1633 section_nr
= pfn_to_section_nr(pfn
);
1634 if (!present_section_nr(section_nr
))
1637 section
= __nr_to_section(section_nr
);
1638 /* same memblock? */
1640 if ((section_nr
>= mem
->start_section_nr
) &&
1641 (section_nr
<= mem
->end_section_nr
))
1644 mem
= find_memory_block_hinted(section
, mem
);
1648 ret
= func(mem
, arg
);
1650 kobject_put(&mem
->dev
.kobj
);
1656 kobject_put(&mem
->dev
.kobj
);
1662 * offline_memory_block_cb - callback function for offlining memory block
1663 * @mem: the memory block to be offlined
1664 * @arg: buffer to hold error msg
1666 * Always return 0, and put the error msg in arg if any.
1668 static int offline_memory_block_cb(struct memory_block
*mem
, void *arg
)
1671 int error
= offline_memory_block(mem
);
1673 if (error
!= 0 && *ret
== 0)
1679 static int is_memblock_offlined_cb(struct memory_block
*mem
, void *arg
)
1681 int ret
= !is_memblock_offlined(mem
);
1684 pr_warn("removing memory fails, because memory "
1685 "[%#010llx-%#010llx] is onlined\n",
1686 PFN_PHYS(section_nr_to_pfn(mem
->start_section_nr
)),
1687 PFN_PHYS(section_nr_to_pfn(mem
->end_section_nr
+ 1))-1);
1692 static int check_cpu_on_node(void *data
)
1694 struct pglist_data
*pgdat
= data
;
1697 for_each_present_cpu(cpu
) {
1698 if (cpu_to_node(cpu
) == pgdat
->node_id
)
1700 * the cpu on this node isn't removed, and we can't
1701 * offline this node.
1709 static void unmap_cpu_on_node(void *data
)
1711 #ifdef CONFIG_ACPI_NUMA
1712 struct pglist_data
*pgdat
= data
;
1715 for_each_possible_cpu(cpu
)
1716 if (cpu_to_node(cpu
) == pgdat
->node_id
)
1717 numa_clear_node(cpu
);
1721 static int check_and_unmap_cpu_on_node(void *data
)
1723 int ret
= check_cpu_on_node(data
);
1729 * the node will be offlined when we come here, so we can clear
1730 * the cpu_to_node() now.
1733 unmap_cpu_on_node(data
);
1737 /* offline the node if all memory sections of this node are removed */
1738 void try_offline_node(int nid
)
1740 pg_data_t
*pgdat
= NODE_DATA(nid
);
1741 unsigned long start_pfn
= pgdat
->node_start_pfn
;
1742 unsigned long end_pfn
= start_pfn
+ pgdat
->node_spanned_pages
;
1744 struct page
*pgdat_page
= virt_to_page(pgdat
);
1747 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
1748 unsigned long section_nr
= pfn_to_section_nr(pfn
);
1750 if (!present_section_nr(section_nr
))
1753 if (pfn_to_nid(pfn
) != nid
)
1757 * some memory sections of this node are not removed, and we
1758 * can't offline node now.
1763 if (stop_machine(check_and_unmap_cpu_on_node
, pgdat
, NULL
))
1767 * all memory/cpu of this node are removed, we can offline this
1770 node_set_offline(nid
);
1771 unregister_one_node(nid
);
1773 if (!PageSlab(pgdat_page
) && !PageCompound(pgdat_page
))
1774 /* node data is allocated from boot memory */
1777 /* free waittable in each zone */
1778 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
1779 struct zone
*zone
= pgdat
->node_zones
+ i
;
1781 if (zone
->wait_table
)
1782 vfree(zone
->wait_table
);
1786 * Since there is no way to guarentee the address of pgdat/zone is not
1787 * on stack of any kernel threads or used by other kernel objects
1788 * without reference counting or other symchronizing method, do not
1789 * reset node_data and free pgdat here. Just reset it to 0 and reuse
1790 * the memory when the node is online again.
1792 memset(pgdat
, 0, sizeof(*pgdat
));
1794 EXPORT_SYMBOL(try_offline_node
);
1796 int __ref
remove_memory(int nid
, u64 start
, u64 size
)
1798 unsigned long start_pfn
, end_pfn
;
1802 start_pfn
= PFN_DOWN(start
);
1803 end_pfn
= start_pfn
+ PFN_DOWN(size
);
1806 * When CONFIG_MEMCG is on, one memory block may be used by other
1807 * blocks to store page cgroup when onlining pages. But we don't know
1808 * in what order pages are onlined. So we iterate twice to offline
1810 * 1st iterate: offline every non primary memory block.
1811 * 2nd iterate: offline primary (i.e. first added) memory block.
1814 walk_memory_range(start_pfn
, end_pfn
, &ret
,
1815 offline_memory_block_cb
);
1825 lock_memory_hotplug();
1828 * we have offlined all memory blocks like this:
1829 * 1. lock memory hotplug
1830 * 2. offline a memory block
1831 * 3. unlock memory hotplug
1833 * repeat step1-3 to offline the memory block. All memory blocks
1834 * must be offlined before removing memory. But we don't hold the
1835 * lock in the whole operation. So we should check whether all
1836 * memory blocks are offlined.
1839 ret
= walk_memory_range(start_pfn
, end_pfn
, NULL
,
1840 is_memblock_offlined_cb
);
1842 unlock_memory_hotplug();
1846 /* remove memmap entry */
1847 firmware_map_remove(start
, start
+ size
, "System RAM");
1849 arch_remove_memory(start
, size
);
1851 try_offline_node(nid
);
1853 unlock_memory_hotplug();
1858 int offline_pages(unsigned long start_pfn
, unsigned long nr_pages
)
1862 int remove_memory(int nid
, u64 start
, u64 size
)
1866 #endif /* CONFIG_MEMORY_HOTREMOVE */
1867 EXPORT_SYMBOL_GPL(remove_memory
);