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
);
131 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
132 #ifndef CONFIG_SPARSEMEM_VMEMMAP
133 static void register_page_bootmem_info_section(unsigned long start_pfn
)
135 unsigned long *usemap
, mapsize
, section_nr
, i
;
136 struct mem_section
*ms
;
137 struct page
*page
, *memmap
;
139 section_nr
= pfn_to_section_nr(start_pfn
);
140 ms
= __nr_to_section(section_nr
);
142 /* Get section's memmap address */
143 memmap
= sparse_decode_mem_map(ms
->section_mem_map
, section_nr
);
146 * Get page for the memmap's phys address
147 * XXX: need more consideration for sparse_vmemmap...
149 page
= virt_to_page(memmap
);
150 mapsize
= sizeof(struct page
) * PAGES_PER_SECTION
;
151 mapsize
= PAGE_ALIGN(mapsize
) >> PAGE_SHIFT
;
153 /* remember memmap's page */
154 for (i
= 0; i
< mapsize
; i
++, page
++)
155 get_page_bootmem(section_nr
, page
, SECTION_INFO
);
157 usemap
= __nr_to_section(section_nr
)->pageblock_flags
;
158 page
= virt_to_page(usemap
);
160 mapsize
= PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT
;
162 for (i
= 0; i
< mapsize
; i
++, page
++)
163 get_page_bootmem(section_nr
, page
, MIX_SECTION_INFO
);
166 #else /* CONFIG_SPARSEMEM_VMEMMAP */
167 static void register_page_bootmem_info_section(unsigned long start_pfn
)
169 unsigned long *usemap
, mapsize
, section_nr
, i
;
170 struct mem_section
*ms
;
171 struct page
*page
, *memmap
;
173 if (!pfn_valid(start_pfn
))
176 section_nr
= pfn_to_section_nr(start_pfn
);
177 ms
= __nr_to_section(section_nr
);
179 memmap
= sparse_decode_mem_map(ms
->section_mem_map
, section_nr
);
181 register_page_bootmem_memmap(section_nr
, memmap
, PAGES_PER_SECTION
);
183 usemap
= __nr_to_section(section_nr
)->pageblock_flags
;
184 page
= virt_to_page(usemap
);
186 mapsize
= PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT
;
188 for (i
= 0; i
< mapsize
; i
++, page
++)
189 get_page_bootmem(section_nr
, page
, MIX_SECTION_INFO
);
191 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
193 void register_page_bootmem_info_node(struct pglist_data
*pgdat
)
195 unsigned long i
, pfn
, end_pfn
, nr_pages
;
196 int node
= pgdat
->node_id
;
200 nr_pages
= PAGE_ALIGN(sizeof(struct pglist_data
)) >> PAGE_SHIFT
;
201 page
= virt_to_page(pgdat
);
203 for (i
= 0; i
< nr_pages
; i
++, page
++)
204 get_page_bootmem(node
, page
, NODE_INFO
);
206 zone
= &pgdat
->node_zones
[0];
207 for (; zone
< pgdat
->node_zones
+ MAX_NR_ZONES
- 1; zone
++) {
208 if (zone
->wait_table
) {
209 nr_pages
= zone
->wait_table_hash_nr_entries
210 * sizeof(wait_queue_head_t
);
211 nr_pages
= PAGE_ALIGN(nr_pages
) >> PAGE_SHIFT
;
212 page
= virt_to_page(zone
->wait_table
);
214 for (i
= 0; i
< nr_pages
; i
++, page
++)
215 get_page_bootmem(node
, page
, NODE_INFO
);
219 pfn
= pgdat
->node_start_pfn
;
220 end_pfn
= pfn
+ pgdat
->node_spanned_pages
;
222 /* register_section info */
223 for (; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
225 * Some platforms can assign the same pfn to multiple nodes - on
226 * node0 as well as nodeN. To avoid registering a pfn against
227 * multiple nodes we check that this pfn does not already
228 * reside in some other node.
230 if (pfn_valid(pfn
) && (pfn_to_nid(pfn
) == node
))
231 register_page_bootmem_info_section(pfn
);
234 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
236 static void grow_zone_span(struct zone
*zone
, unsigned long start_pfn
,
237 unsigned long end_pfn
)
239 unsigned long old_zone_end_pfn
;
241 zone_span_writelock(zone
);
243 old_zone_end_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
244 if (!zone
->spanned_pages
|| start_pfn
< zone
->zone_start_pfn
)
245 zone
->zone_start_pfn
= start_pfn
;
247 zone
->spanned_pages
= max(old_zone_end_pfn
, end_pfn
) -
248 zone
->zone_start_pfn
;
250 zone_span_writeunlock(zone
);
253 static void resize_zone(struct zone
*zone
, unsigned long start_pfn
,
254 unsigned long end_pfn
)
256 zone_span_writelock(zone
);
258 if (end_pfn
- start_pfn
) {
259 zone
->zone_start_pfn
= start_pfn
;
260 zone
->spanned_pages
= end_pfn
- start_pfn
;
263 * make it consist as free_area_init_core(),
264 * if spanned_pages = 0, then keep start_pfn = 0
266 zone
->zone_start_pfn
= 0;
267 zone
->spanned_pages
= 0;
270 zone_span_writeunlock(zone
);
273 static void fix_zone_id(struct zone
*zone
, unsigned long start_pfn
,
274 unsigned long end_pfn
)
276 enum zone_type zid
= zone_idx(zone
);
277 int nid
= zone
->zone_pgdat
->node_id
;
280 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
281 set_page_links(pfn_to_page(pfn
), zid
, nid
, pfn
);
284 static int __meminit
move_pfn_range_left(struct zone
*z1
, struct zone
*z2
,
285 unsigned long start_pfn
, unsigned long end_pfn
)
289 unsigned long z1_start_pfn
;
291 if (!z1
->wait_table
) {
292 ret
= init_currently_empty_zone(z1
, start_pfn
,
293 end_pfn
- start_pfn
, MEMMAP_HOTPLUG
);
298 pgdat_resize_lock(z1
->zone_pgdat
, &flags
);
300 /* can't move pfns which are higher than @z2 */
301 if (end_pfn
> z2
->zone_start_pfn
+ z2
->spanned_pages
)
303 /* the move out part mast at the left most of @z2 */
304 if (start_pfn
> z2
->zone_start_pfn
)
306 /* must included/overlap */
307 if (end_pfn
<= z2
->zone_start_pfn
)
310 /* use start_pfn for z1's start_pfn if z1 is empty */
311 if (z1
->spanned_pages
)
312 z1_start_pfn
= z1
->zone_start_pfn
;
314 z1_start_pfn
= start_pfn
;
316 resize_zone(z1
, z1_start_pfn
, end_pfn
);
317 resize_zone(z2
, end_pfn
, z2
->zone_start_pfn
+ z2
->spanned_pages
);
319 pgdat_resize_unlock(z1
->zone_pgdat
, &flags
);
321 fix_zone_id(z1
, start_pfn
, end_pfn
);
325 pgdat_resize_unlock(z1
->zone_pgdat
, &flags
);
329 static int __meminit
move_pfn_range_right(struct zone
*z1
, struct zone
*z2
,
330 unsigned long start_pfn
, unsigned long end_pfn
)
334 unsigned long z2_end_pfn
;
336 if (!z2
->wait_table
) {
337 ret
= init_currently_empty_zone(z2
, start_pfn
,
338 end_pfn
- start_pfn
, MEMMAP_HOTPLUG
);
343 pgdat_resize_lock(z1
->zone_pgdat
, &flags
);
345 /* can't move pfns which are lower than @z1 */
346 if (z1
->zone_start_pfn
> start_pfn
)
348 /* the move out part mast at the right most of @z1 */
349 if (z1
->zone_start_pfn
+ z1
->spanned_pages
> end_pfn
)
351 /* must included/overlap */
352 if (start_pfn
>= z1
->zone_start_pfn
+ z1
->spanned_pages
)
355 /* use end_pfn for z2's end_pfn if z2 is empty */
356 if (z2
->spanned_pages
)
357 z2_end_pfn
= z2
->zone_start_pfn
+ z2
->spanned_pages
;
359 z2_end_pfn
= end_pfn
;
361 resize_zone(z1
, z1
->zone_start_pfn
, start_pfn
);
362 resize_zone(z2
, start_pfn
, z2_end_pfn
);
364 pgdat_resize_unlock(z1
->zone_pgdat
, &flags
);
366 fix_zone_id(z2
, start_pfn
, end_pfn
);
370 pgdat_resize_unlock(z1
->zone_pgdat
, &flags
);
374 static void grow_pgdat_span(struct pglist_data
*pgdat
, unsigned long start_pfn
,
375 unsigned long end_pfn
)
377 unsigned long old_pgdat_end_pfn
=
378 pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
380 if (!pgdat
->node_spanned_pages
|| start_pfn
< pgdat
->node_start_pfn
)
381 pgdat
->node_start_pfn
= start_pfn
;
383 pgdat
->node_spanned_pages
= max(old_pgdat_end_pfn
, end_pfn
) -
384 pgdat
->node_start_pfn
;
387 static int __meminit
__add_zone(struct zone
*zone
, unsigned long phys_start_pfn
)
389 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
390 int nr_pages
= PAGES_PER_SECTION
;
391 int nid
= pgdat
->node_id
;
395 zone_type
= zone
- pgdat
->node_zones
;
396 if (!zone
->wait_table
) {
399 ret
= init_currently_empty_zone(zone
, phys_start_pfn
,
400 nr_pages
, MEMMAP_HOTPLUG
);
404 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
405 grow_zone_span(zone
, phys_start_pfn
, phys_start_pfn
+ nr_pages
);
406 grow_pgdat_span(zone
->zone_pgdat
, phys_start_pfn
,
407 phys_start_pfn
+ nr_pages
);
408 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
409 memmap_init_zone(nr_pages
, nid
, zone_type
,
410 phys_start_pfn
, MEMMAP_HOTPLUG
);
414 static int __meminit
__add_section(int nid
, struct zone
*zone
,
415 unsigned long phys_start_pfn
)
417 int nr_pages
= PAGES_PER_SECTION
;
420 if (pfn_valid(phys_start_pfn
))
423 ret
= sparse_add_one_section(zone
, phys_start_pfn
, nr_pages
);
428 ret
= __add_zone(zone
, phys_start_pfn
);
433 return register_new_memory(nid
, __pfn_to_section(phys_start_pfn
));
436 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
437 static int find_smallest_section_pfn(int nid
, struct zone
*zone
,
438 unsigned long start_pfn
,
439 unsigned long end_pfn
)
441 struct mem_section
*ms
;
443 for (; start_pfn
< end_pfn
; start_pfn
+= PAGES_PER_SECTION
) {
444 ms
= __pfn_to_section(start_pfn
);
446 if (unlikely(!valid_section(ms
)))
449 if (unlikely(pfn_to_nid(start_pfn
) != nid
))
452 if (zone
&& zone
!= page_zone(pfn_to_page(start_pfn
)))
461 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
462 static int find_biggest_section_pfn(int nid
, struct zone
*zone
,
463 unsigned long start_pfn
,
464 unsigned long end_pfn
)
466 struct mem_section
*ms
;
469 /* pfn is the end pfn of a memory section. */
471 for (; pfn
>= start_pfn
; pfn
-= PAGES_PER_SECTION
) {
472 ms
= __pfn_to_section(pfn
);
474 if (unlikely(!valid_section(ms
)))
477 if (unlikely(pfn_to_nid(pfn
) != nid
))
480 if (zone
&& zone
!= page_zone(pfn_to_page(pfn
)))
489 static void shrink_zone_span(struct zone
*zone
, unsigned long start_pfn
,
490 unsigned long end_pfn
)
492 unsigned long zone_start_pfn
= zone
->zone_start_pfn
;
493 unsigned long zone_end_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
495 struct mem_section
*ms
;
496 int nid
= zone_to_nid(zone
);
498 zone_span_writelock(zone
);
499 if (zone_start_pfn
== start_pfn
) {
501 * If the section is smallest section in the zone, it need
502 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
503 * In this case, we find second smallest valid mem_section
504 * for shrinking zone.
506 pfn
= find_smallest_section_pfn(nid
, zone
, end_pfn
,
509 zone
->zone_start_pfn
= pfn
;
510 zone
->spanned_pages
= zone_end_pfn
- pfn
;
512 } else if (zone_end_pfn
== end_pfn
) {
514 * If the section is biggest section in the zone, it need
515 * shrink zone->spanned_pages.
516 * In this case, we find second biggest valid mem_section for
519 pfn
= find_biggest_section_pfn(nid
, zone
, zone_start_pfn
,
522 zone
->spanned_pages
= pfn
- zone_start_pfn
+ 1;
526 * The section is not biggest or smallest mem_section in the zone, it
527 * only creates a hole in the zone. So in this case, we need not
528 * change the zone. But perhaps, the zone has only hole data. Thus
529 * it check the zone has only hole or not.
531 pfn
= zone_start_pfn
;
532 for (; pfn
< zone_end_pfn
; pfn
+= PAGES_PER_SECTION
) {
533 ms
= __pfn_to_section(pfn
);
535 if (unlikely(!valid_section(ms
)))
538 if (page_zone(pfn_to_page(pfn
)) != zone
)
541 /* If the section is current section, it continues the loop */
542 if (start_pfn
== pfn
)
545 /* If we find valid section, we have nothing to do */
546 zone_span_writeunlock(zone
);
550 /* The zone has no valid section */
551 zone
->zone_start_pfn
= 0;
552 zone
->spanned_pages
= 0;
553 zone_span_writeunlock(zone
);
556 static void shrink_pgdat_span(struct pglist_data
*pgdat
,
557 unsigned long start_pfn
, unsigned long end_pfn
)
559 unsigned long pgdat_start_pfn
= pgdat
->node_start_pfn
;
560 unsigned long pgdat_end_pfn
=
561 pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
563 struct mem_section
*ms
;
564 int nid
= pgdat
->node_id
;
566 if (pgdat_start_pfn
== start_pfn
) {
568 * If the section is smallest section in the pgdat, it need
569 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
570 * In this case, we find second smallest valid mem_section
571 * for shrinking zone.
573 pfn
= find_smallest_section_pfn(nid
, NULL
, end_pfn
,
576 pgdat
->node_start_pfn
= pfn
;
577 pgdat
->node_spanned_pages
= pgdat_end_pfn
- pfn
;
579 } else if (pgdat_end_pfn
== end_pfn
) {
581 * If the section is biggest section in the pgdat, it need
582 * shrink pgdat->node_spanned_pages.
583 * In this case, we find second biggest valid mem_section for
586 pfn
= find_biggest_section_pfn(nid
, NULL
, pgdat_start_pfn
,
589 pgdat
->node_spanned_pages
= pfn
- pgdat_start_pfn
+ 1;
593 * If the section is not biggest or smallest mem_section in the pgdat,
594 * it only creates a hole in the pgdat. So in this case, we need not
596 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
597 * has only hole or not.
599 pfn
= pgdat_start_pfn
;
600 for (; pfn
< pgdat_end_pfn
; pfn
+= PAGES_PER_SECTION
) {
601 ms
= __pfn_to_section(pfn
);
603 if (unlikely(!valid_section(ms
)))
606 if (pfn_to_nid(pfn
) != nid
)
609 /* If the section is current section, it continues the loop */
610 if (start_pfn
== pfn
)
613 /* If we find valid section, we have nothing to do */
617 /* The pgdat has no valid section */
618 pgdat
->node_start_pfn
= 0;
619 pgdat
->node_spanned_pages
= 0;
622 static void __remove_zone(struct zone
*zone
, unsigned long start_pfn
)
624 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
625 int nr_pages
= PAGES_PER_SECTION
;
629 zone_type
= zone
- pgdat
->node_zones
;
631 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
632 shrink_zone_span(zone
, start_pfn
, start_pfn
+ nr_pages
);
633 shrink_pgdat_span(pgdat
, start_pfn
, start_pfn
+ nr_pages
);
634 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
637 static int __remove_section(struct zone
*zone
, struct mem_section
*ms
)
639 unsigned long start_pfn
;
643 if (!valid_section(ms
))
646 ret
= unregister_memory_section(ms
);
650 scn_nr
= __section_nr(ms
);
651 start_pfn
= section_nr_to_pfn(scn_nr
);
652 __remove_zone(zone
, start_pfn
);
654 sparse_remove_one_section(zone
, ms
);
659 * Reasonably generic function for adding memory. It is
660 * expected that archs that support memory hotplug will
661 * call this function after deciding the zone to which to
664 int __ref
__add_pages(int nid
, struct zone
*zone
, unsigned long phys_start_pfn
,
665 unsigned long nr_pages
)
669 int start_sec
, end_sec
;
670 /* during initialize mem_map, align hot-added range to section */
671 start_sec
= pfn_to_section_nr(phys_start_pfn
);
672 end_sec
= pfn_to_section_nr(phys_start_pfn
+ nr_pages
- 1);
674 for (i
= start_sec
; i
<= end_sec
; i
++) {
675 err
= __add_section(nid
, zone
, i
<< PFN_SECTION_SHIFT
);
678 * EEXIST is finally dealt with by ioresource collision
679 * check. see add_memory() => register_memory_resource()
680 * Warning will be printed if there is collision.
682 if (err
&& (err
!= -EEXIST
))
689 EXPORT_SYMBOL_GPL(__add_pages
);
692 * __remove_pages() - remove sections of pages from a zone
693 * @zone: zone from which pages need to be removed
694 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
695 * @nr_pages: number of pages to remove (must be multiple of section size)
697 * Generic helper function to remove section mappings and sysfs entries
698 * for the section of the memory we are removing. Caller needs to make
699 * sure that pages are marked reserved and zones are adjust properly by
700 * calling offline_pages().
702 int __remove_pages(struct zone
*zone
, unsigned long phys_start_pfn
,
703 unsigned long nr_pages
)
705 unsigned long i
, ret
= 0;
706 int sections_to_remove
;
709 * We can only remove entire sections
711 BUG_ON(phys_start_pfn
& ~PAGE_SECTION_MASK
);
712 BUG_ON(nr_pages
% PAGES_PER_SECTION
);
714 release_mem_region(phys_start_pfn
<< PAGE_SHIFT
, nr_pages
* PAGE_SIZE
);
716 sections_to_remove
= nr_pages
/ PAGES_PER_SECTION
;
717 for (i
= 0; i
< sections_to_remove
; i
++) {
718 unsigned long pfn
= phys_start_pfn
+ i
*PAGES_PER_SECTION
;
719 ret
= __remove_section(zone
, __pfn_to_section(pfn
));
725 EXPORT_SYMBOL_GPL(__remove_pages
);
727 int set_online_page_callback(online_page_callback_t callback
)
731 lock_memory_hotplug();
733 if (online_page_callback
== generic_online_page
) {
734 online_page_callback
= callback
;
738 unlock_memory_hotplug();
742 EXPORT_SYMBOL_GPL(set_online_page_callback
);
744 int restore_online_page_callback(online_page_callback_t callback
)
748 lock_memory_hotplug();
750 if (online_page_callback
== callback
) {
751 online_page_callback
= generic_online_page
;
755 unlock_memory_hotplug();
759 EXPORT_SYMBOL_GPL(restore_online_page_callback
);
761 void __online_page_set_limits(struct page
*page
)
763 unsigned long pfn
= page_to_pfn(page
);
765 if (pfn
>= num_physpages
)
766 num_physpages
= pfn
+ 1;
768 EXPORT_SYMBOL_GPL(__online_page_set_limits
);
770 void __online_page_increment_counters(struct page
*page
)
774 #ifdef CONFIG_HIGHMEM
775 if (PageHighMem(page
))
779 EXPORT_SYMBOL_GPL(__online_page_increment_counters
);
781 void __online_page_free(struct page
*page
)
783 ClearPageReserved(page
);
784 init_page_count(page
);
787 EXPORT_SYMBOL_GPL(__online_page_free
);
789 static void generic_online_page(struct page
*page
)
791 __online_page_set_limits(page
);
792 __online_page_increment_counters(page
);
793 __online_page_free(page
);
796 static int online_pages_range(unsigned long start_pfn
, unsigned long nr_pages
,
800 unsigned long onlined_pages
= *(unsigned long *)arg
;
802 if (PageReserved(pfn_to_page(start_pfn
)))
803 for (i
= 0; i
< nr_pages
; i
++) {
804 page
= pfn_to_page(start_pfn
+ i
);
805 (*online_page_callback
)(page
);
808 *(unsigned long *)arg
= onlined_pages
;
812 #ifdef CONFIG_MOVABLE_NODE
814 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
817 static bool can_online_high_movable(struct zone
*zone
)
821 #else /* CONFIG_MOVABLE_NODE */
822 /* ensure every online node has NORMAL memory */
823 static bool can_online_high_movable(struct zone
*zone
)
825 return node_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
827 #endif /* CONFIG_MOVABLE_NODE */
829 /* check which state of node_states will be changed when online memory */
830 static void node_states_check_changes_online(unsigned long nr_pages
,
831 struct zone
*zone
, struct memory_notify
*arg
)
833 int nid
= zone_to_nid(zone
);
834 enum zone_type zone_last
= ZONE_NORMAL
;
837 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
838 * contains nodes which have zones of 0...ZONE_NORMAL,
839 * set zone_last to ZONE_NORMAL.
841 * If we don't have HIGHMEM nor movable node,
842 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
843 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
845 if (N_MEMORY
== N_NORMAL_MEMORY
)
846 zone_last
= ZONE_MOVABLE
;
849 * if the memory to be online is in a zone of 0...zone_last, and
850 * the zones of 0...zone_last don't have memory before online, we will
851 * need to set the node to node_states[N_NORMAL_MEMORY] after
852 * the memory is online.
854 if (zone_idx(zone
) <= zone_last
&& !node_state(nid
, N_NORMAL_MEMORY
))
855 arg
->status_change_nid_normal
= nid
;
857 arg
->status_change_nid_normal
= -1;
859 #ifdef CONFIG_HIGHMEM
861 * If we have movable node, node_states[N_HIGH_MEMORY]
862 * contains nodes which have zones of 0...ZONE_HIGHMEM,
863 * set zone_last to ZONE_HIGHMEM.
865 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
866 * contains nodes which have zones of 0...ZONE_MOVABLE,
867 * set zone_last to ZONE_MOVABLE.
869 zone_last
= ZONE_HIGHMEM
;
870 if (N_MEMORY
== N_HIGH_MEMORY
)
871 zone_last
= ZONE_MOVABLE
;
873 if (zone_idx(zone
) <= zone_last
&& !node_state(nid
, N_HIGH_MEMORY
))
874 arg
->status_change_nid_high
= nid
;
876 arg
->status_change_nid_high
= -1;
878 arg
->status_change_nid_high
= arg
->status_change_nid_normal
;
882 * if the node don't have memory befor online, we will need to
883 * set the node to node_states[N_MEMORY] after the memory
886 if (!node_state(nid
, N_MEMORY
))
887 arg
->status_change_nid
= nid
;
889 arg
->status_change_nid
= -1;
892 static void node_states_set_node(int node
, struct memory_notify
*arg
)
894 if (arg
->status_change_nid_normal
>= 0)
895 node_set_state(node
, N_NORMAL_MEMORY
);
897 if (arg
->status_change_nid_high
>= 0)
898 node_set_state(node
, N_HIGH_MEMORY
);
900 node_set_state(node
, N_MEMORY
);
904 int __ref
online_pages(unsigned long pfn
, unsigned long nr_pages
, int online_type
)
906 unsigned long onlined_pages
= 0;
908 int need_zonelists_rebuild
= 0;
911 struct memory_notify arg
;
913 lock_memory_hotplug();
915 * This doesn't need a lock to do pfn_to_page().
916 * The section can't be removed here because of the
917 * memory_block->state_mutex.
919 zone
= page_zone(pfn_to_page(pfn
));
921 if ((zone_idx(zone
) > ZONE_NORMAL
|| online_type
== ONLINE_MOVABLE
) &&
922 !can_online_high_movable(zone
)) {
923 unlock_memory_hotplug();
927 if (online_type
== ONLINE_KERNEL
&& zone_idx(zone
) == ZONE_MOVABLE
) {
928 if (move_pfn_range_left(zone
- 1, zone
, pfn
, pfn
+ nr_pages
)) {
929 unlock_memory_hotplug();
933 if (online_type
== ONLINE_MOVABLE
&& zone_idx(zone
) == ZONE_MOVABLE
- 1) {
934 if (move_pfn_range_right(zone
, zone
+ 1, pfn
, pfn
+ nr_pages
)) {
935 unlock_memory_hotplug();
940 /* Previous code may changed the zone of the pfn range */
941 zone
= page_zone(pfn_to_page(pfn
));
944 arg
.nr_pages
= nr_pages
;
945 node_states_check_changes_online(nr_pages
, zone
, &arg
);
947 nid
= page_to_nid(pfn_to_page(pfn
));
949 ret
= memory_notify(MEM_GOING_ONLINE
, &arg
);
950 ret
= notifier_to_errno(ret
);
952 memory_notify(MEM_CANCEL_ONLINE
, &arg
);
953 unlock_memory_hotplug();
957 * If this zone is not populated, then it is not in zonelist.
958 * This means the page allocator ignores this zone.
959 * So, zonelist must be updated after online.
961 mutex_lock(&zonelists_mutex
);
962 if (!populated_zone(zone
)) {
963 need_zonelists_rebuild
= 1;
964 build_all_zonelists(NULL
, zone
);
967 ret
= walk_system_ram_range(pfn
, nr_pages
, &onlined_pages
,
970 if (need_zonelists_rebuild
)
971 zone_pcp_reset(zone
);
972 mutex_unlock(&zonelists_mutex
);
973 printk(KERN_DEBUG
"online_pages [mem %#010llx-%#010llx] failed\n",
974 (unsigned long long) pfn
<< PAGE_SHIFT
,
975 (((unsigned long long) pfn
+ nr_pages
)
977 memory_notify(MEM_CANCEL_ONLINE
, &arg
);
978 unlock_memory_hotplug();
982 zone
->managed_pages
+= onlined_pages
;
983 zone
->present_pages
+= onlined_pages
;
984 zone
->zone_pgdat
->node_present_pages
+= onlined_pages
;
986 node_states_set_node(zone_to_nid(zone
), &arg
);
987 if (need_zonelists_rebuild
)
988 build_all_zonelists(NULL
, NULL
);
990 zone_pcp_update(zone
);
993 mutex_unlock(&zonelists_mutex
);
995 init_per_zone_wmark_min();
998 kswapd_run(zone_to_nid(zone
));
1000 vm_total_pages
= nr_free_pagecache_pages();
1002 writeback_set_ratelimit();
1005 memory_notify(MEM_ONLINE
, &arg
);
1006 unlock_memory_hotplug();
1010 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1012 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1013 static pg_data_t __ref
*hotadd_new_pgdat(int nid
, u64 start
)
1015 struct pglist_data
*pgdat
;
1016 unsigned long zones_size
[MAX_NR_ZONES
] = {0};
1017 unsigned long zholes_size
[MAX_NR_ZONES
] = {0};
1018 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
1020 pgdat
= NODE_DATA(nid
);
1022 pgdat
= arch_alloc_nodedata(nid
);
1026 arch_refresh_nodedata(nid
, pgdat
);
1029 /* we can use NODE_DATA(nid) from here */
1031 /* init node's zones as empty zones, we don't have any present pages.*/
1032 free_area_init_node(nid
, zones_size
, start_pfn
, zholes_size
);
1035 * The node we allocated has no zone fallback lists. For avoiding
1036 * to access not-initialized zonelist, build here.
1038 mutex_lock(&zonelists_mutex
);
1039 build_all_zonelists(pgdat
, NULL
);
1040 mutex_unlock(&zonelists_mutex
);
1045 static void rollback_node_hotadd(int nid
, pg_data_t
*pgdat
)
1047 arch_refresh_nodedata(nid
, NULL
);
1048 arch_free_nodedata(pgdat
);
1054 * called by cpu_up() to online a node without onlined memory.
1056 int mem_online_node(int nid
)
1061 lock_memory_hotplug();
1062 pgdat
= hotadd_new_pgdat(nid
, 0);
1067 node_set_online(nid
);
1068 ret
= register_one_node(nid
);
1072 unlock_memory_hotplug();
1076 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1077 int __ref
add_memory(int nid
, u64 start
, u64 size
)
1079 pg_data_t
*pgdat
= NULL
;
1082 struct resource
*res
;
1085 lock_memory_hotplug();
1087 res
= register_memory_resource(start
, size
);
1092 { /* Stupid hack to suppress address-never-null warning */
1093 void *p
= NODE_DATA(nid
);
1096 new_node
= !node_online(nid
);
1098 pgdat
= hotadd_new_pgdat(nid
, start
);
1104 /* call arch's memory hotadd */
1105 ret
= arch_add_memory(nid
, start
, size
);
1110 /* we online node here. we can't roll back from here. */
1111 node_set_online(nid
);
1114 ret
= register_one_node(nid
);
1116 * If sysfs file of new node can't create, cpu on the node
1117 * can't be hot-added. There is no rollback way now.
1118 * So, check by BUG_ON() to catch it reluctantly..
1123 /* create new memmap entry */
1124 firmware_map_add_hotplug(start
, start
+ size
, "System RAM");
1129 /* rollback pgdat allocation and others */
1131 rollback_node_hotadd(nid
, pgdat
);
1132 release_memory_resource(res
);
1135 unlock_memory_hotplug();
1138 EXPORT_SYMBOL_GPL(add_memory
);
1140 #ifdef CONFIG_MEMORY_HOTREMOVE
1142 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1143 * set and the size of the free page is given by page_order(). Using this,
1144 * the function determines if the pageblock contains only free pages.
1145 * Due to buddy contraints, a free page at least the size of a pageblock will
1146 * be located at the start of the pageblock
1148 static inline int pageblock_free(struct page
*page
)
1150 return PageBuddy(page
) && page_order(page
) >= pageblock_order
;
1153 /* Return the start of the next active pageblock after a given page */
1154 static struct page
*next_active_pageblock(struct page
*page
)
1156 /* Ensure the starting page is pageblock-aligned */
1157 BUG_ON(page_to_pfn(page
) & (pageblock_nr_pages
- 1));
1159 /* If the entire pageblock is free, move to the end of free page */
1160 if (pageblock_free(page
)) {
1162 /* be careful. we don't have locks, page_order can be changed.*/
1163 order
= page_order(page
);
1164 if ((order
< MAX_ORDER
) && (order
>= pageblock_order
))
1165 return page
+ (1 << order
);
1168 return page
+ pageblock_nr_pages
;
1171 /* Checks if this range of memory is likely to be hot-removable. */
1172 int is_mem_section_removable(unsigned long start_pfn
, unsigned long nr_pages
)
1174 struct page
*page
= pfn_to_page(start_pfn
);
1175 struct page
*end_page
= page
+ nr_pages
;
1177 /* Check the starting page of each pageblock within the range */
1178 for (; page
< end_page
; page
= next_active_pageblock(page
)) {
1179 if (!is_pageblock_removable_nolock(page
))
1184 /* All pageblocks in the memory block are likely to be hot-removable */
1189 * Confirm all pages in a range [start, end) is belongs to the same zone.
1191 static int test_pages_in_a_zone(unsigned long start_pfn
, unsigned long end_pfn
)
1194 struct zone
*zone
= NULL
;
1197 for (pfn
= start_pfn
;
1199 pfn
+= MAX_ORDER_NR_PAGES
) {
1201 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1202 while ((i
< MAX_ORDER_NR_PAGES
) && !pfn_valid_within(pfn
+ i
))
1204 if (i
== MAX_ORDER_NR_PAGES
)
1206 page
= pfn_to_page(pfn
+ i
);
1207 if (zone
&& page_zone(page
) != zone
)
1209 zone
= page_zone(page
);
1215 * Scanning pfn is much easier than scanning lru list.
1216 * Scan pfn from start to end and Find LRU page.
1218 static unsigned long scan_lru_pages(unsigned long start
, unsigned long end
)
1222 for (pfn
= start
; pfn
< end
; pfn
++) {
1223 if (pfn_valid(pfn
)) {
1224 page
= pfn_to_page(pfn
);
1232 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1234 do_migrate_range(unsigned long start_pfn
, unsigned long end_pfn
)
1238 int move_pages
= NR_OFFLINE_AT_ONCE_PAGES
;
1239 int not_managed
= 0;
1243 for (pfn
= start_pfn
; pfn
< end_pfn
&& move_pages
> 0; pfn
++) {
1244 if (!pfn_valid(pfn
))
1246 page
= pfn_to_page(pfn
);
1247 if (!get_page_unless_zero(page
))
1250 * We can skip free pages. And we can only deal with pages on
1253 ret
= isolate_lru_page(page
);
1254 if (!ret
) { /* Success */
1256 list_add_tail(&page
->lru
, &source
);
1258 inc_zone_page_state(page
, NR_ISOLATED_ANON
+
1259 page_is_file_cache(page
));
1262 #ifdef CONFIG_DEBUG_VM
1263 printk(KERN_ALERT
"removing pfn %lx from LRU failed\n",
1268 /* Because we don't have big zone->lock. we should
1269 check this again here. */
1270 if (page_count(page
)) {
1277 if (!list_empty(&source
)) {
1279 putback_lru_pages(&source
);
1284 * alloc_migrate_target should be improooooved!!
1285 * migrate_pages returns # of failed pages.
1287 ret
= migrate_pages(&source
, alloc_migrate_target
, 0,
1291 putback_lru_pages(&source
);
1298 * remove from free_area[] and mark all as Reserved.
1301 offline_isolated_pages_cb(unsigned long start
, unsigned long nr_pages
,
1304 __offline_isolated_pages(start
, start
+ nr_pages
);
1309 offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
1311 walk_system_ram_range(start_pfn
, end_pfn
- start_pfn
, NULL
,
1312 offline_isolated_pages_cb
);
1316 * Check all pages in range, recoreded as memory resource, are isolated.
1319 check_pages_isolated_cb(unsigned long start_pfn
, unsigned long nr_pages
,
1323 long offlined
= *(long *)data
;
1324 ret
= test_pages_isolated(start_pfn
, start_pfn
+ nr_pages
, true);
1325 offlined
= nr_pages
;
1327 *(long *)data
+= offlined
;
1332 check_pages_isolated(unsigned long start_pfn
, unsigned long end_pfn
)
1337 ret
= walk_system_ram_range(start_pfn
, end_pfn
- start_pfn
, &offlined
,
1338 check_pages_isolated_cb
);
1340 offlined
= (long)ret
;
1344 #ifdef CONFIG_MOVABLE_NODE
1346 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1349 static bool can_offline_normal(struct zone
*zone
, unsigned long nr_pages
)
1353 #else /* CONFIG_MOVABLE_NODE */
1354 /* ensure the node has NORMAL memory if it is still online */
1355 static bool can_offline_normal(struct zone
*zone
, unsigned long nr_pages
)
1357 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1358 unsigned long present_pages
= 0;
1361 for (zt
= 0; zt
<= ZONE_NORMAL
; zt
++)
1362 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1364 if (present_pages
> nr_pages
)
1368 for (; zt
<= ZONE_MOVABLE
; zt
++)
1369 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1372 * we can't offline the last normal memory until all
1373 * higher memory is offlined.
1375 return present_pages
== 0;
1377 #endif /* CONFIG_MOVABLE_NODE */
1379 /* check which state of node_states will be changed when offline memory */
1380 static void node_states_check_changes_offline(unsigned long nr_pages
,
1381 struct zone
*zone
, struct memory_notify
*arg
)
1383 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1384 unsigned long present_pages
= 0;
1385 enum zone_type zt
, zone_last
= ZONE_NORMAL
;
1388 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1389 * contains nodes which have zones of 0...ZONE_NORMAL,
1390 * set zone_last to ZONE_NORMAL.
1392 * If we don't have HIGHMEM nor movable node,
1393 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1394 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1396 if (N_MEMORY
== N_NORMAL_MEMORY
)
1397 zone_last
= ZONE_MOVABLE
;
1400 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1401 * If the memory to be offline is in a zone of 0...zone_last,
1402 * and it is the last present memory, 0...zone_last will
1403 * become empty after offline , thus we can determind we will
1404 * need to clear the node from node_states[N_NORMAL_MEMORY].
1406 for (zt
= 0; zt
<= zone_last
; zt
++)
1407 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1408 if (zone_idx(zone
) <= zone_last
&& nr_pages
>= present_pages
)
1409 arg
->status_change_nid_normal
= zone_to_nid(zone
);
1411 arg
->status_change_nid_normal
= -1;
1413 #ifdef CONFIG_HIGHMEM
1415 * If we have movable node, node_states[N_HIGH_MEMORY]
1416 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1417 * set zone_last to ZONE_HIGHMEM.
1419 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1420 * contains nodes which have zones of 0...ZONE_MOVABLE,
1421 * set zone_last to ZONE_MOVABLE.
1423 zone_last
= ZONE_HIGHMEM
;
1424 if (N_MEMORY
== N_HIGH_MEMORY
)
1425 zone_last
= ZONE_MOVABLE
;
1427 for (; zt
<= zone_last
; zt
++)
1428 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1429 if (zone_idx(zone
) <= zone_last
&& nr_pages
>= present_pages
)
1430 arg
->status_change_nid_high
= zone_to_nid(zone
);
1432 arg
->status_change_nid_high
= -1;
1434 arg
->status_change_nid_high
= arg
->status_change_nid_normal
;
1438 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1440 zone_last
= ZONE_MOVABLE
;
1443 * check whether node_states[N_HIGH_MEMORY] will be changed
1444 * If we try to offline the last present @nr_pages from the node,
1445 * we can determind we will need to clear the node from
1446 * node_states[N_HIGH_MEMORY].
1448 for (; zt
<= zone_last
; zt
++)
1449 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1450 if (nr_pages
>= present_pages
)
1451 arg
->status_change_nid
= zone_to_nid(zone
);
1453 arg
->status_change_nid
= -1;
1456 static void node_states_clear_node(int node
, struct memory_notify
*arg
)
1458 if (arg
->status_change_nid_normal
>= 0)
1459 node_clear_state(node
, N_NORMAL_MEMORY
);
1461 if ((N_MEMORY
!= N_NORMAL_MEMORY
) &&
1462 (arg
->status_change_nid_high
>= 0))
1463 node_clear_state(node
, N_HIGH_MEMORY
);
1465 if ((N_MEMORY
!= N_HIGH_MEMORY
) &&
1466 (arg
->status_change_nid
>= 0))
1467 node_clear_state(node
, N_MEMORY
);
1470 static int __ref
__offline_pages(unsigned long start_pfn
,
1471 unsigned long end_pfn
, unsigned long timeout
)
1473 unsigned long pfn
, nr_pages
, expire
;
1474 long offlined_pages
;
1475 int ret
, drain
, retry_max
, node
;
1477 struct memory_notify arg
;
1479 BUG_ON(start_pfn
>= end_pfn
);
1480 /* at least, alignment against pageblock is necessary */
1481 if (!IS_ALIGNED(start_pfn
, pageblock_nr_pages
))
1483 if (!IS_ALIGNED(end_pfn
, pageblock_nr_pages
))
1485 /* This makes hotplug much easier...and readable.
1486 we assume this for now. .*/
1487 if (!test_pages_in_a_zone(start_pfn
, end_pfn
))
1490 lock_memory_hotplug();
1492 zone
= page_zone(pfn_to_page(start_pfn
));
1493 node
= zone_to_nid(zone
);
1494 nr_pages
= end_pfn
- start_pfn
;
1497 if (zone_idx(zone
) <= ZONE_NORMAL
&& !can_offline_normal(zone
, nr_pages
))
1500 /* set above range as isolated */
1501 ret
= start_isolate_page_range(start_pfn
, end_pfn
,
1502 MIGRATE_MOVABLE
, true);
1506 arg
.start_pfn
= start_pfn
;
1507 arg
.nr_pages
= nr_pages
;
1508 node_states_check_changes_offline(nr_pages
, zone
, &arg
);
1510 ret
= memory_notify(MEM_GOING_OFFLINE
, &arg
);
1511 ret
= notifier_to_errno(ret
);
1513 goto failed_removal
;
1516 expire
= jiffies
+ timeout
;
1520 /* start memory hot removal */
1522 if (time_after(jiffies
, expire
))
1523 goto failed_removal
;
1525 if (signal_pending(current
))
1526 goto failed_removal
;
1529 lru_add_drain_all();
1534 pfn
= scan_lru_pages(start_pfn
, end_pfn
);
1535 if (pfn
) { /* We have page on LRU */
1536 ret
= do_migrate_range(pfn
, end_pfn
);
1542 if (--retry_max
== 0)
1543 goto failed_removal
;
1549 /* drain all zone's lru pagevec, this is asynchronous... */
1550 lru_add_drain_all();
1552 /* drain pcp pages, this is synchronous. */
1555 offlined_pages
= check_pages_isolated(start_pfn
, end_pfn
);
1556 if (offlined_pages
< 0) {
1558 goto failed_removal
;
1560 printk(KERN_INFO
"Offlined Pages %ld\n", offlined_pages
);
1561 /* Ok, all of our target is isolated.
1562 We cannot do rollback at this point. */
1563 offline_isolated_pages(start_pfn
, end_pfn
);
1564 /* reset pagetype flags and makes migrate type to be MOVABLE */
1565 undo_isolate_page_range(start_pfn
, end_pfn
, MIGRATE_MOVABLE
);
1566 /* removal success */
1567 zone
->managed_pages
-= offlined_pages
;
1568 zone
->present_pages
-= offlined_pages
;
1569 zone
->zone_pgdat
->node_present_pages
-= offlined_pages
;
1570 totalram_pages
-= offlined_pages
;
1572 init_per_zone_wmark_min();
1574 if (!populated_zone(zone
)) {
1575 zone_pcp_reset(zone
);
1576 mutex_lock(&zonelists_mutex
);
1577 build_all_zonelists(NULL
, NULL
);
1578 mutex_unlock(&zonelists_mutex
);
1580 zone_pcp_update(zone
);
1582 node_states_clear_node(node
, &arg
);
1583 if (arg
.status_change_nid
>= 0)
1586 vm_total_pages
= nr_free_pagecache_pages();
1587 writeback_set_ratelimit();
1589 memory_notify(MEM_OFFLINE
, &arg
);
1590 unlock_memory_hotplug();
1594 printk(KERN_INFO
"memory offlining [mem %#010llx-%#010llx] failed\n",
1595 (unsigned long long) start_pfn
<< PAGE_SHIFT
,
1596 ((unsigned long long) end_pfn
<< PAGE_SHIFT
) - 1);
1597 memory_notify(MEM_CANCEL_OFFLINE
, &arg
);
1598 /* pushback to free area */
1599 undo_isolate_page_range(start_pfn
, end_pfn
, MIGRATE_MOVABLE
);
1602 unlock_memory_hotplug();
1606 int offline_pages(unsigned long start_pfn
, unsigned long nr_pages
)
1608 return __offline_pages(start_pfn
, start_pfn
+ nr_pages
, 120 * HZ
);
1612 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1613 * @start_pfn: start pfn of the memory range
1614 * @end_pfn: end pft of the memory range
1615 * @arg: argument passed to func
1616 * @func: callback for each memory section walked
1618 * This function walks through all present mem sections in range
1619 * [start_pfn, end_pfn) and call func on each mem section.
1621 * Returns the return value of func.
1623 static int walk_memory_range(unsigned long start_pfn
, unsigned long end_pfn
,
1624 void *arg
, int (*func
)(struct memory_block
*, void *))
1626 struct memory_block
*mem
= NULL
;
1627 struct mem_section
*section
;
1628 unsigned long pfn
, section_nr
;
1631 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
1632 section_nr
= pfn_to_section_nr(pfn
);
1633 if (!present_section_nr(section_nr
))
1636 section
= __nr_to_section(section_nr
);
1637 /* same memblock? */
1639 if ((section_nr
>= mem
->start_section_nr
) &&
1640 (section_nr
<= mem
->end_section_nr
))
1643 mem
= find_memory_block_hinted(section
, mem
);
1647 ret
= func(mem
, arg
);
1649 kobject_put(&mem
->dev
.kobj
);
1655 kobject_put(&mem
->dev
.kobj
);
1661 * offline_memory_block_cb - callback function for offlining memory block
1662 * @mem: the memory block to be offlined
1663 * @arg: buffer to hold error msg
1665 * Always return 0, and put the error msg in arg if any.
1667 static int offline_memory_block_cb(struct memory_block
*mem
, void *arg
)
1670 int error
= offline_memory_block(mem
);
1672 if (error
!= 0 && *ret
== 0)
1678 static int is_memblock_offlined_cb(struct memory_block
*mem
, void *arg
)
1680 int ret
= !is_memblock_offlined(mem
);
1683 pr_warn("removing memory fails, because memory "
1684 "[%#010llx-%#010llx] is onlined\n",
1685 PFN_PHYS(section_nr_to_pfn(mem
->start_section_nr
)),
1686 PFN_PHYS(section_nr_to_pfn(mem
->end_section_nr
+ 1))-1);
1691 static int check_cpu_on_node(void *data
)
1693 struct pglist_data
*pgdat
= data
;
1696 for_each_present_cpu(cpu
) {
1697 if (cpu_to_node(cpu
) == pgdat
->node_id
)
1699 * the cpu on this node isn't removed, and we can't
1700 * offline this node.
1708 static void unmap_cpu_on_node(void *data
)
1710 #ifdef CONFIG_ACPI_NUMA
1711 struct pglist_data
*pgdat
= data
;
1714 for_each_possible_cpu(cpu
)
1715 if (cpu_to_node(cpu
) == pgdat
->node_id
)
1716 numa_clear_node(cpu
);
1720 static int check_and_unmap_cpu_on_node(void *data
)
1722 int ret
= check_cpu_on_node(data
);
1728 * the node will be offlined when we come here, so we can clear
1729 * the cpu_to_node() now.
1732 unmap_cpu_on_node(data
);
1736 /* offline the node if all memory sections of this node are removed */
1737 void try_offline_node(int nid
)
1739 pg_data_t
*pgdat
= NODE_DATA(nid
);
1740 unsigned long start_pfn
= pgdat
->node_start_pfn
;
1741 unsigned long end_pfn
= start_pfn
+ pgdat
->node_spanned_pages
;
1743 struct page
*pgdat_page
= virt_to_page(pgdat
);
1746 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
1747 unsigned long section_nr
= pfn_to_section_nr(pfn
);
1749 if (!present_section_nr(section_nr
))
1752 if (pfn_to_nid(pfn
) != nid
)
1756 * some memory sections of this node are not removed, and we
1757 * can't offline node now.
1762 if (stop_machine(check_and_unmap_cpu_on_node
, pgdat
, NULL
))
1766 * all memory/cpu of this node are removed, we can offline this
1769 node_set_offline(nid
);
1770 unregister_one_node(nid
);
1772 if (!PageSlab(pgdat_page
) && !PageCompound(pgdat_page
))
1773 /* node data is allocated from boot memory */
1776 /* free waittable in each zone */
1777 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
1778 struct zone
*zone
= pgdat
->node_zones
+ i
;
1780 if (zone
->wait_table
)
1781 vfree(zone
->wait_table
);
1785 * Since there is no way to guarentee the address of pgdat/zone is not
1786 * on stack of any kernel threads or used by other kernel objects
1787 * without reference counting or other symchronizing method, do not
1788 * reset node_data and free pgdat here. Just reset it to 0 and reuse
1789 * the memory when the node is online again.
1791 memset(pgdat
, 0, sizeof(*pgdat
));
1793 EXPORT_SYMBOL(try_offline_node
);
1795 int __ref
remove_memory(int nid
, u64 start
, u64 size
)
1797 unsigned long start_pfn
, end_pfn
;
1801 start_pfn
= PFN_DOWN(start
);
1802 end_pfn
= start_pfn
+ PFN_DOWN(size
);
1805 * When CONFIG_MEMCG is on, one memory block may be used by other
1806 * blocks to store page cgroup when onlining pages. But we don't know
1807 * in what order pages are onlined. So we iterate twice to offline
1809 * 1st iterate: offline every non primary memory block.
1810 * 2nd iterate: offline primary (i.e. first added) memory block.
1813 walk_memory_range(start_pfn
, end_pfn
, &ret
,
1814 offline_memory_block_cb
);
1824 lock_memory_hotplug();
1827 * we have offlined all memory blocks like this:
1828 * 1. lock memory hotplug
1829 * 2. offline a memory block
1830 * 3. unlock memory hotplug
1832 * repeat step1-3 to offline the memory block. All memory blocks
1833 * must be offlined before removing memory. But we don't hold the
1834 * lock in the whole operation. So we should check whether all
1835 * memory blocks are offlined.
1838 ret
= walk_memory_range(start_pfn
, end_pfn
, NULL
,
1839 is_memblock_offlined_cb
);
1841 unlock_memory_hotplug();
1845 /* remove memmap entry */
1846 firmware_map_remove(start
, start
+ size
, "System RAM");
1848 arch_remove_memory(start
, size
);
1850 try_offline_node(nid
);
1852 unlock_memory_hotplug();
1857 int offline_pages(unsigned long start_pfn
, unsigned long nr_pages
)
1861 int remove_memory(int nid
, u64 start
, u64 size
)
1865 #endif /* CONFIG_MEMORY_HOTREMOVE */
1866 EXPORT_SYMBOL_GPL(remove_memory
);