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>
33 #include <asm/tlbflush.h>
38 * online_page_callback contains pointer to current page onlining function.
39 * Initially it is generic_online_page(). If it is required it could be
40 * changed by calling set_online_page_callback() for callback registration
41 * and restore_online_page_callback() for generic callback restore.
44 static void generic_online_page(struct page
*page
);
46 static online_page_callback_t online_page_callback
= generic_online_page
;
48 DEFINE_MUTEX(mem_hotplug_mutex
);
50 void lock_memory_hotplug(void)
52 mutex_lock(&mem_hotplug_mutex
);
54 /* for exclusive hibernation if CONFIG_HIBERNATION=y */
58 void unlock_memory_hotplug(void)
60 unlock_system_sleep();
61 mutex_unlock(&mem_hotplug_mutex
);
65 /* add this memory to iomem resource */
66 static struct resource
*register_memory_resource(u64 start
, u64 size
)
69 res
= kzalloc(sizeof(struct resource
), GFP_KERNEL
);
72 res
->name
= "System RAM";
74 res
->end
= start
+ size
- 1;
75 res
->flags
= IORESOURCE_MEM
| IORESOURCE_BUSY
;
76 if (request_resource(&iomem_resource
, res
) < 0) {
77 printk("System RAM resource %pR cannot be added\n", res
);
84 static void release_memory_resource(struct resource
*res
)
88 release_resource(res
);
93 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
94 void get_page_bootmem(unsigned long info
, struct page
*page
,
97 page
->lru
.next
= (struct list_head
*) type
;
99 set_page_private(page
, info
);
100 atomic_inc(&page
->_count
);
103 /* reference to __meminit __free_pages_bootmem is valid
104 * so use __ref to tell modpost not to generate a warning */
105 void __ref
put_page_bootmem(struct page
*page
)
108 static DEFINE_MUTEX(ppb_lock
);
110 type
= (unsigned long) page
->lru
.next
;
111 BUG_ON(type
< MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE
||
112 type
> MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE
);
114 if (atomic_dec_return(&page
->_count
) == 1) {
115 ClearPagePrivate(page
);
116 set_page_private(page
, 0);
117 INIT_LIST_HEAD(&page
->lru
);
120 * Please refer to comment for __free_pages_bootmem()
121 * for why we serialize here.
123 mutex_lock(&ppb_lock
);
124 __free_pages_bootmem(page
, 0);
125 mutex_unlock(&ppb_lock
);
130 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
131 #ifndef CONFIG_SPARSEMEM_VMEMMAP
132 static void register_page_bootmem_info_section(unsigned long start_pfn
)
134 unsigned long *usemap
, mapsize
, section_nr
, i
;
135 struct mem_section
*ms
;
136 struct page
*page
, *memmap
;
138 section_nr
= pfn_to_section_nr(start_pfn
);
139 ms
= __nr_to_section(section_nr
);
141 /* Get section's memmap address */
142 memmap
= sparse_decode_mem_map(ms
->section_mem_map
, section_nr
);
145 * Get page for the memmap's phys address
146 * XXX: need more consideration for sparse_vmemmap...
148 page
= virt_to_page(memmap
);
149 mapsize
= sizeof(struct page
) * PAGES_PER_SECTION
;
150 mapsize
= PAGE_ALIGN(mapsize
) >> PAGE_SHIFT
;
152 /* remember memmap's page */
153 for (i
= 0; i
< mapsize
; i
++, page
++)
154 get_page_bootmem(section_nr
, page
, SECTION_INFO
);
156 usemap
= __nr_to_section(section_nr
)->pageblock_flags
;
157 page
= virt_to_page(usemap
);
159 mapsize
= PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT
;
161 for (i
= 0; i
< mapsize
; i
++, page
++)
162 get_page_bootmem(section_nr
, page
, MIX_SECTION_INFO
);
165 #else /* CONFIG_SPARSEMEM_VMEMMAP */
166 static void register_page_bootmem_info_section(unsigned long start_pfn
)
168 unsigned long *usemap
, mapsize
, section_nr
, i
;
169 struct mem_section
*ms
;
170 struct page
*page
, *memmap
;
172 if (!pfn_valid(start_pfn
))
175 section_nr
= pfn_to_section_nr(start_pfn
);
176 ms
= __nr_to_section(section_nr
);
178 memmap
= sparse_decode_mem_map(ms
->section_mem_map
, section_nr
);
180 register_page_bootmem_memmap(section_nr
, memmap
, PAGES_PER_SECTION
);
182 usemap
= __nr_to_section(section_nr
)->pageblock_flags
;
183 page
= virt_to_page(usemap
);
185 mapsize
= PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT
;
187 for (i
= 0; i
< mapsize
; i
++, page
++)
188 get_page_bootmem(section_nr
, page
, MIX_SECTION_INFO
);
190 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
192 void register_page_bootmem_info_node(struct pglist_data
*pgdat
)
194 unsigned long i
, pfn
, end_pfn
, nr_pages
;
195 int node
= pgdat
->node_id
;
199 nr_pages
= PAGE_ALIGN(sizeof(struct pglist_data
)) >> PAGE_SHIFT
;
200 page
= virt_to_page(pgdat
);
202 for (i
= 0; i
< nr_pages
; i
++, page
++)
203 get_page_bootmem(node
, page
, NODE_INFO
);
205 zone
= &pgdat
->node_zones
[0];
206 for (; zone
< pgdat
->node_zones
+ MAX_NR_ZONES
- 1; zone
++) {
207 if (zone
->wait_table
) {
208 nr_pages
= zone
->wait_table_hash_nr_entries
209 * sizeof(wait_queue_head_t
);
210 nr_pages
= PAGE_ALIGN(nr_pages
) >> PAGE_SHIFT
;
211 page
= virt_to_page(zone
->wait_table
);
213 for (i
= 0; i
< nr_pages
; i
++, page
++)
214 get_page_bootmem(node
, page
, NODE_INFO
);
218 pfn
= pgdat
->node_start_pfn
;
219 end_pfn
= pfn
+ pgdat
->node_spanned_pages
;
221 /* register_section info */
222 for (; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
224 * Some platforms can assign the same pfn to multiple nodes - on
225 * node0 as well as nodeN. To avoid registering a pfn against
226 * multiple nodes we check that this pfn does not already
227 * reside in some other node.
229 if (pfn_valid(pfn
) && (pfn_to_nid(pfn
) == node
))
230 register_page_bootmem_info_section(pfn
);
233 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
235 static void grow_zone_span(struct zone
*zone
, unsigned long start_pfn
,
236 unsigned long end_pfn
)
238 unsigned long old_zone_end_pfn
;
240 zone_span_writelock(zone
);
242 old_zone_end_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
243 if (!zone
->spanned_pages
|| start_pfn
< zone
->zone_start_pfn
)
244 zone
->zone_start_pfn
= start_pfn
;
246 zone
->spanned_pages
= max(old_zone_end_pfn
, end_pfn
) -
247 zone
->zone_start_pfn
;
249 zone_span_writeunlock(zone
);
252 static void resize_zone(struct zone
*zone
, unsigned long start_pfn
,
253 unsigned long end_pfn
)
255 zone_span_writelock(zone
);
257 if (end_pfn
- start_pfn
) {
258 zone
->zone_start_pfn
= start_pfn
;
259 zone
->spanned_pages
= end_pfn
- start_pfn
;
262 * make it consist as free_area_init_core(),
263 * if spanned_pages = 0, then keep start_pfn = 0
265 zone
->zone_start_pfn
= 0;
266 zone
->spanned_pages
= 0;
269 zone_span_writeunlock(zone
);
272 static void fix_zone_id(struct zone
*zone
, unsigned long start_pfn
,
273 unsigned long end_pfn
)
275 enum zone_type zid
= zone_idx(zone
);
276 int nid
= zone
->zone_pgdat
->node_id
;
279 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
280 set_page_links(pfn_to_page(pfn
), zid
, nid
, pfn
);
283 static int __meminit
move_pfn_range_left(struct zone
*z1
, struct zone
*z2
,
284 unsigned long start_pfn
, unsigned long end_pfn
)
288 unsigned long z1_start_pfn
;
290 if (!z1
->wait_table
) {
291 ret
= init_currently_empty_zone(z1
, start_pfn
,
292 end_pfn
- start_pfn
, MEMMAP_HOTPLUG
);
297 pgdat_resize_lock(z1
->zone_pgdat
, &flags
);
299 /* can't move pfns which are higher than @z2 */
300 if (end_pfn
> z2
->zone_start_pfn
+ z2
->spanned_pages
)
302 /* the move out part mast at the left most of @z2 */
303 if (start_pfn
> z2
->zone_start_pfn
)
305 /* must included/overlap */
306 if (end_pfn
<= z2
->zone_start_pfn
)
309 /* use start_pfn for z1's start_pfn if z1 is empty */
310 if (z1
->spanned_pages
)
311 z1_start_pfn
= z1
->zone_start_pfn
;
313 z1_start_pfn
= start_pfn
;
315 resize_zone(z1
, z1_start_pfn
, end_pfn
);
316 resize_zone(z2
, end_pfn
, z2
->zone_start_pfn
+ z2
->spanned_pages
);
318 pgdat_resize_unlock(z1
->zone_pgdat
, &flags
);
320 fix_zone_id(z1
, start_pfn
, end_pfn
);
324 pgdat_resize_unlock(z1
->zone_pgdat
, &flags
);
328 static int __meminit
move_pfn_range_right(struct zone
*z1
, struct zone
*z2
,
329 unsigned long start_pfn
, unsigned long end_pfn
)
333 unsigned long z2_end_pfn
;
335 if (!z2
->wait_table
) {
336 ret
= init_currently_empty_zone(z2
, start_pfn
,
337 end_pfn
- start_pfn
, MEMMAP_HOTPLUG
);
342 pgdat_resize_lock(z1
->zone_pgdat
, &flags
);
344 /* can't move pfns which are lower than @z1 */
345 if (z1
->zone_start_pfn
> start_pfn
)
347 /* the move out part mast at the right most of @z1 */
348 if (z1
->zone_start_pfn
+ z1
->spanned_pages
> end_pfn
)
350 /* must included/overlap */
351 if (start_pfn
>= z1
->zone_start_pfn
+ z1
->spanned_pages
)
354 /* use end_pfn for z2's end_pfn if z2 is empty */
355 if (z2
->spanned_pages
)
356 z2_end_pfn
= z2
->zone_start_pfn
+ z2
->spanned_pages
;
358 z2_end_pfn
= end_pfn
;
360 resize_zone(z1
, z1
->zone_start_pfn
, start_pfn
);
361 resize_zone(z2
, start_pfn
, z2_end_pfn
);
363 pgdat_resize_unlock(z1
->zone_pgdat
, &flags
);
365 fix_zone_id(z2
, start_pfn
, end_pfn
);
369 pgdat_resize_unlock(z1
->zone_pgdat
, &flags
);
373 static void grow_pgdat_span(struct pglist_data
*pgdat
, unsigned long start_pfn
,
374 unsigned long end_pfn
)
376 unsigned long old_pgdat_end_pfn
=
377 pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
379 if (!pgdat
->node_spanned_pages
|| start_pfn
< pgdat
->node_start_pfn
)
380 pgdat
->node_start_pfn
= start_pfn
;
382 pgdat
->node_spanned_pages
= max(old_pgdat_end_pfn
, end_pfn
) -
383 pgdat
->node_start_pfn
;
386 static int __meminit
__add_zone(struct zone
*zone
, unsigned long phys_start_pfn
)
388 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
389 int nr_pages
= PAGES_PER_SECTION
;
390 int nid
= pgdat
->node_id
;
394 zone_type
= zone
- pgdat
->node_zones
;
395 if (!zone
->wait_table
) {
398 ret
= init_currently_empty_zone(zone
, phys_start_pfn
,
399 nr_pages
, MEMMAP_HOTPLUG
);
403 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
404 grow_zone_span(zone
, phys_start_pfn
, phys_start_pfn
+ nr_pages
);
405 grow_pgdat_span(zone
->zone_pgdat
, phys_start_pfn
,
406 phys_start_pfn
+ nr_pages
);
407 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
408 memmap_init_zone(nr_pages
, nid
, zone_type
,
409 phys_start_pfn
, MEMMAP_HOTPLUG
);
413 static int __meminit
__add_section(int nid
, struct zone
*zone
,
414 unsigned long phys_start_pfn
)
416 int nr_pages
= PAGES_PER_SECTION
;
419 if (pfn_valid(phys_start_pfn
))
422 ret
= sparse_add_one_section(zone
, phys_start_pfn
, nr_pages
);
427 ret
= __add_zone(zone
, phys_start_pfn
);
432 return register_new_memory(nid
, __pfn_to_section(phys_start_pfn
));
435 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
436 static int find_smallest_section_pfn(int nid
, struct zone
*zone
,
437 unsigned long start_pfn
,
438 unsigned long end_pfn
)
440 struct mem_section
*ms
;
442 for (; start_pfn
< end_pfn
; start_pfn
+= PAGES_PER_SECTION
) {
443 ms
= __pfn_to_section(start_pfn
);
445 if (unlikely(!valid_section(ms
)))
448 if (unlikely(pfn_to_nid(start_pfn
) != nid
))
451 if (zone
&& zone
!= page_zone(pfn_to_page(start_pfn
)))
460 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
461 static int find_biggest_section_pfn(int nid
, struct zone
*zone
,
462 unsigned long start_pfn
,
463 unsigned long end_pfn
)
465 struct mem_section
*ms
;
468 /* pfn is the end pfn of a memory section. */
470 for (; pfn
>= start_pfn
; pfn
-= PAGES_PER_SECTION
) {
471 ms
= __pfn_to_section(pfn
);
473 if (unlikely(!valid_section(ms
)))
476 if (unlikely(pfn_to_nid(pfn
) != nid
))
479 if (zone
&& zone
!= page_zone(pfn_to_page(pfn
)))
488 static void shrink_zone_span(struct zone
*zone
, unsigned long start_pfn
,
489 unsigned long end_pfn
)
491 unsigned long zone_start_pfn
= zone
->zone_start_pfn
;
492 unsigned long zone_end_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
494 struct mem_section
*ms
;
495 int nid
= zone_to_nid(zone
);
497 zone_span_writelock(zone
);
498 if (zone_start_pfn
== start_pfn
) {
500 * If the section is smallest section in the zone, it need
501 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
502 * In this case, we find second smallest valid mem_section
503 * for shrinking zone.
505 pfn
= find_smallest_section_pfn(nid
, zone
, end_pfn
,
508 zone
->zone_start_pfn
= pfn
;
509 zone
->spanned_pages
= zone_end_pfn
- pfn
;
511 } else if (zone_end_pfn
== end_pfn
) {
513 * If the section is biggest section in the zone, it need
514 * shrink zone->spanned_pages.
515 * In this case, we find second biggest valid mem_section for
518 pfn
= find_biggest_section_pfn(nid
, zone
, zone_start_pfn
,
521 zone
->spanned_pages
= pfn
- zone_start_pfn
+ 1;
525 * The section is not biggest or smallest mem_section in the zone, it
526 * only creates a hole in the zone. So in this case, we need not
527 * change the zone. But perhaps, the zone has only hole data. Thus
528 * it check the zone has only hole or not.
530 pfn
= zone_start_pfn
;
531 for (; pfn
< zone_end_pfn
; pfn
+= PAGES_PER_SECTION
) {
532 ms
= __pfn_to_section(pfn
);
534 if (unlikely(!valid_section(ms
)))
537 if (page_zone(pfn_to_page(pfn
)) != zone
)
540 /* If the section is current section, it continues the loop */
541 if (start_pfn
== pfn
)
544 /* If we find valid section, we have nothing to do */
545 zone_span_writeunlock(zone
);
549 /* The zone has no valid section */
550 zone
->zone_start_pfn
= 0;
551 zone
->spanned_pages
= 0;
552 zone_span_writeunlock(zone
);
555 static void shrink_pgdat_span(struct pglist_data
*pgdat
,
556 unsigned long start_pfn
, unsigned long end_pfn
)
558 unsigned long pgdat_start_pfn
= pgdat
->node_start_pfn
;
559 unsigned long pgdat_end_pfn
=
560 pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
562 struct mem_section
*ms
;
563 int nid
= pgdat
->node_id
;
565 if (pgdat_start_pfn
== start_pfn
) {
567 * If the section is smallest section in the pgdat, it need
568 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
569 * In this case, we find second smallest valid mem_section
570 * for shrinking zone.
572 pfn
= find_smallest_section_pfn(nid
, NULL
, end_pfn
,
575 pgdat
->node_start_pfn
= pfn
;
576 pgdat
->node_spanned_pages
= pgdat_end_pfn
- pfn
;
578 } else if (pgdat_end_pfn
== end_pfn
) {
580 * If the section is biggest section in the pgdat, it need
581 * shrink pgdat->node_spanned_pages.
582 * In this case, we find second biggest valid mem_section for
585 pfn
= find_biggest_section_pfn(nid
, NULL
, pgdat_start_pfn
,
588 pgdat
->node_spanned_pages
= pfn
- pgdat_start_pfn
+ 1;
592 * If the section is not biggest or smallest mem_section in the pgdat,
593 * it only creates a hole in the pgdat. So in this case, we need not
595 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
596 * has only hole or not.
598 pfn
= pgdat_start_pfn
;
599 for (; pfn
< pgdat_end_pfn
; pfn
+= PAGES_PER_SECTION
) {
600 ms
= __pfn_to_section(pfn
);
602 if (unlikely(!valid_section(ms
)))
605 if (pfn_to_nid(pfn
) != nid
)
608 /* If the section is current section, it continues the loop */
609 if (start_pfn
== pfn
)
612 /* If we find valid section, we have nothing to do */
616 /* The pgdat has no valid section */
617 pgdat
->node_start_pfn
= 0;
618 pgdat
->node_spanned_pages
= 0;
621 static void __remove_zone(struct zone
*zone
, unsigned long start_pfn
)
623 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
624 int nr_pages
= PAGES_PER_SECTION
;
628 zone_type
= zone
- pgdat
->node_zones
;
630 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
631 shrink_zone_span(zone
, start_pfn
, start_pfn
+ nr_pages
);
632 shrink_pgdat_span(pgdat
, start_pfn
, start_pfn
+ nr_pages
);
633 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
636 static int __remove_section(struct zone
*zone
, struct mem_section
*ms
)
638 unsigned long start_pfn
;
642 if (!valid_section(ms
))
645 ret
= unregister_memory_section(ms
);
649 scn_nr
= __section_nr(ms
);
650 start_pfn
= section_nr_to_pfn(scn_nr
);
651 __remove_zone(zone
, start_pfn
);
653 sparse_remove_one_section(zone
, ms
);
658 * Reasonably generic function for adding memory. It is
659 * expected that archs that support memory hotplug will
660 * call this function after deciding the zone to which to
663 int __ref
__add_pages(int nid
, struct zone
*zone
, unsigned long phys_start_pfn
,
664 unsigned long nr_pages
)
668 int start_sec
, end_sec
;
669 /* during initialize mem_map, align hot-added range to section */
670 start_sec
= pfn_to_section_nr(phys_start_pfn
);
671 end_sec
= pfn_to_section_nr(phys_start_pfn
+ nr_pages
- 1);
673 for (i
= start_sec
; i
<= end_sec
; i
++) {
674 err
= __add_section(nid
, zone
, i
<< PFN_SECTION_SHIFT
);
677 * EEXIST is finally dealt with by ioresource collision
678 * check. see add_memory() => register_memory_resource()
679 * Warning will be printed if there is collision.
681 if (err
&& (err
!= -EEXIST
))
688 EXPORT_SYMBOL_GPL(__add_pages
);
691 * __remove_pages() - remove sections of pages from a zone
692 * @zone: zone from which pages need to be removed
693 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
694 * @nr_pages: number of pages to remove (must be multiple of section size)
696 * Generic helper function to remove section mappings and sysfs entries
697 * for the section of the memory we are removing. Caller needs to make
698 * sure that pages are marked reserved and zones are adjust properly by
699 * calling offline_pages().
701 int __remove_pages(struct zone
*zone
, unsigned long phys_start_pfn
,
702 unsigned long nr_pages
)
704 unsigned long i
, ret
= 0;
705 int sections_to_remove
;
708 * We can only remove entire sections
710 BUG_ON(phys_start_pfn
& ~PAGE_SECTION_MASK
);
711 BUG_ON(nr_pages
% PAGES_PER_SECTION
);
713 release_mem_region(phys_start_pfn
<< PAGE_SHIFT
, nr_pages
* PAGE_SIZE
);
715 sections_to_remove
= nr_pages
/ PAGES_PER_SECTION
;
716 for (i
= 0; i
< sections_to_remove
; i
++) {
717 unsigned long pfn
= phys_start_pfn
+ i
*PAGES_PER_SECTION
;
718 ret
= __remove_section(zone
, __pfn_to_section(pfn
));
724 EXPORT_SYMBOL_GPL(__remove_pages
);
726 int set_online_page_callback(online_page_callback_t callback
)
730 lock_memory_hotplug();
732 if (online_page_callback
== generic_online_page
) {
733 online_page_callback
= callback
;
737 unlock_memory_hotplug();
741 EXPORT_SYMBOL_GPL(set_online_page_callback
);
743 int restore_online_page_callback(online_page_callback_t callback
)
747 lock_memory_hotplug();
749 if (online_page_callback
== callback
) {
750 online_page_callback
= generic_online_page
;
754 unlock_memory_hotplug();
758 EXPORT_SYMBOL_GPL(restore_online_page_callback
);
760 void __online_page_set_limits(struct page
*page
)
762 unsigned long pfn
= page_to_pfn(page
);
764 if (pfn
>= num_physpages
)
765 num_physpages
= pfn
+ 1;
767 EXPORT_SYMBOL_GPL(__online_page_set_limits
);
769 void __online_page_increment_counters(struct page
*page
)
773 #ifdef CONFIG_HIGHMEM
774 if (PageHighMem(page
))
778 EXPORT_SYMBOL_GPL(__online_page_increment_counters
);
780 void __online_page_free(struct page
*page
)
782 ClearPageReserved(page
);
783 init_page_count(page
);
786 EXPORT_SYMBOL_GPL(__online_page_free
);
788 static void generic_online_page(struct page
*page
)
790 __online_page_set_limits(page
);
791 __online_page_increment_counters(page
);
792 __online_page_free(page
);
795 static int online_pages_range(unsigned long start_pfn
, unsigned long nr_pages
,
799 unsigned long onlined_pages
= *(unsigned long *)arg
;
801 if (PageReserved(pfn_to_page(start_pfn
)))
802 for (i
= 0; i
< nr_pages
; i
++) {
803 page
= pfn_to_page(start_pfn
+ i
);
804 (*online_page_callback
)(page
);
807 *(unsigned long *)arg
= onlined_pages
;
811 #ifdef CONFIG_MOVABLE_NODE
813 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
816 static bool can_online_high_movable(struct zone
*zone
)
820 #else /* CONFIG_MOVABLE_NODE */
821 /* ensure every online node has NORMAL memory */
822 static bool can_online_high_movable(struct zone
*zone
)
824 return node_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
826 #endif /* CONFIG_MOVABLE_NODE */
828 /* check which state of node_states will be changed when online memory */
829 static void node_states_check_changes_online(unsigned long nr_pages
,
830 struct zone
*zone
, struct memory_notify
*arg
)
832 int nid
= zone_to_nid(zone
);
833 enum zone_type zone_last
= ZONE_NORMAL
;
836 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
837 * contains nodes which have zones of 0...ZONE_NORMAL,
838 * set zone_last to ZONE_NORMAL.
840 * If we don't have HIGHMEM nor movable node,
841 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
842 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
844 if (N_MEMORY
== N_NORMAL_MEMORY
)
845 zone_last
= ZONE_MOVABLE
;
848 * if the memory to be online is in a zone of 0...zone_last, and
849 * the zones of 0...zone_last don't have memory before online, we will
850 * need to set the node to node_states[N_NORMAL_MEMORY] after
851 * the memory is online.
853 if (zone_idx(zone
) <= zone_last
&& !node_state(nid
, N_NORMAL_MEMORY
))
854 arg
->status_change_nid_normal
= nid
;
856 arg
->status_change_nid_normal
= -1;
858 #ifdef CONFIG_HIGHMEM
860 * If we have movable node, node_states[N_HIGH_MEMORY]
861 * contains nodes which have zones of 0...ZONE_HIGHMEM,
862 * set zone_last to ZONE_HIGHMEM.
864 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
865 * contains nodes which have zones of 0...ZONE_MOVABLE,
866 * set zone_last to ZONE_MOVABLE.
868 zone_last
= ZONE_HIGHMEM
;
869 if (N_MEMORY
== N_HIGH_MEMORY
)
870 zone_last
= ZONE_MOVABLE
;
872 if (zone_idx(zone
) <= zone_last
&& !node_state(nid
, N_HIGH_MEMORY
))
873 arg
->status_change_nid_high
= nid
;
875 arg
->status_change_nid_high
= -1;
877 arg
->status_change_nid_high
= arg
->status_change_nid_normal
;
881 * if the node don't have memory befor online, we will need to
882 * set the node to node_states[N_MEMORY] after the memory
885 if (!node_state(nid
, N_MEMORY
))
886 arg
->status_change_nid
= nid
;
888 arg
->status_change_nid
= -1;
891 static void node_states_set_node(int node
, struct memory_notify
*arg
)
893 if (arg
->status_change_nid_normal
>= 0)
894 node_set_state(node
, N_NORMAL_MEMORY
);
896 if (arg
->status_change_nid_high
>= 0)
897 node_set_state(node
, N_HIGH_MEMORY
);
899 node_set_state(node
, N_MEMORY
);
903 int __ref
online_pages(unsigned long pfn
, unsigned long nr_pages
, int online_type
)
905 unsigned long onlined_pages
= 0;
907 int need_zonelists_rebuild
= 0;
910 struct memory_notify arg
;
912 lock_memory_hotplug();
914 * This doesn't need a lock to do pfn_to_page().
915 * The section can't be removed here because of the
916 * memory_block->state_mutex.
918 zone
= page_zone(pfn_to_page(pfn
));
920 if ((zone_idx(zone
) > ZONE_NORMAL
|| online_type
== ONLINE_MOVABLE
) &&
921 !can_online_high_movable(zone
)) {
922 unlock_memory_hotplug();
926 if (online_type
== ONLINE_KERNEL
&& zone_idx(zone
) == ZONE_MOVABLE
) {
927 if (move_pfn_range_left(zone
- 1, zone
, pfn
, pfn
+ nr_pages
)) {
928 unlock_memory_hotplug();
932 if (online_type
== ONLINE_MOVABLE
&& zone_idx(zone
) == ZONE_MOVABLE
- 1) {
933 if (move_pfn_range_right(zone
, zone
+ 1, pfn
, pfn
+ nr_pages
)) {
934 unlock_memory_hotplug();
939 /* Previous code may changed the zone of the pfn range */
940 zone
= page_zone(pfn_to_page(pfn
));
943 arg
.nr_pages
= nr_pages
;
944 node_states_check_changes_online(nr_pages
, zone
, &arg
);
946 nid
= page_to_nid(pfn_to_page(pfn
));
948 ret
= memory_notify(MEM_GOING_ONLINE
, &arg
);
949 ret
= notifier_to_errno(ret
);
951 memory_notify(MEM_CANCEL_ONLINE
, &arg
);
952 unlock_memory_hotplug();
956 * If this zone is not populated, then it is not in zonelist.
957 * This means the page allocator ignores this zone.
958 * So, zonelist must be updated after online.
960 mutex_lock(&zonelists_mutex
);
961 if (!populated_zone(zone
)) {
962 need_zonelists_rebuild
= 1;
963 build_all_zonelists(NULL
, zone
);
966 ret
= walk_system_ram_range(pfn
, nr_pages
, &onlined_pages
,
969 if (need_zonelists_rebuild
)
970 zone_pcp_reset(zone
);
971 mutex_unlock(&zonelists_mutex
);
972 printk(KERN_DEBUG
"online_pages [mem %#010llx-%#010llx] failed\n",
973 (unsigned long long) pfn
<< PAGE_SHIFT
,
974 (((unsigned long long) pfn
+ nr_pages
)
976 memory_notify(MEM_CANCEL_ONLINE
, &arg
);
977 unlock_memory_hotplug();
981 zone
->managed_pages
+= onlined_pages
;
982 zone
->present_pages
+= onlined_pages
;
983 zone
->zone_pgdat
->node_present_pages
+= onlined_pages
;
985 node_states_set_node(zone_to_nid(zone
), &arg
);
986 if (need_zonelists_rebuild
)
987 build_all_zonelists(NULL
, NULL
);
989 zone_pcp_update(zone
);
992 mutex_unlock(&zonelists_mutex
);
994 init_per_zone_wmark_min();
997 kswapd_run(zone_to_nid(zone
));
999 vm_total_pages
= nr_free_pagecache_pages();
1001 writeback_set_ratelimit();
1004 memory_notify(MEM_ONLINE
, &arg
);
1005 unlock_memory_hotplug();
1009 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1011 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1012 static pg_data_t __ref
*hotadd_new_pgdat(int nid
, u64 start
)
1014 struct pglist_data
*pgdat
;
1015 unsigned long zones_size
[MAX_NR_ZONES
] = {0};
1016 unsigned long zholes_size
[MAX_NR_ZONES
] = {0};
1017 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
1019 pgdat
= arch_alloc_nodedata(nid
);
1023 arch_refresh_nodedata(nid
, pgdat
);
1025 /* we can use NODE_DATA(nid) from here */
1027 /* init node's zones as empty zones, we don't have any present pages.*/
1028 free_area_init_node(nid
, zones_size
, start_pfn
, zholes_size
);
1031 * The node we allocated has no zone fallback lists. For avoiding
1032 * to access not-initialized zonelist, build here.
1034 mutex_lock(&zonelists_mutex
);
1035 build_all_zonelists(pgdat
, NULL
);
1036 mutex_unlock(&zonelists_mutex
);
1041 static void rollback_node_hotadd(int nid
, pg_data_t
*pgdat
)
1043 arch_refresh_nodedata(nid
, NULL
);
1044 arch_free_nodedata(pgdat
);
1050 * called by cpu_up() to online a node without onlined memory.
1052 int mem_online_node(int nid
)
1057 lock_memory_hotplug();
1058 pgdat
= hotadd_new_pgdat(nid
, 0);
1063 node_set_online(nid
);
1064 ret
= register_one_node(nid
);
1068 unlock_memory_hotplug();
1072 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1073 int __ref
add_memory(int nid
, u64 start
, u64 size
)
1075 pg_data_t
*pgdat
= NULL
;
1077 struct resource
*res
;
1080 lock_memory_hotplug();
1082 res
= register_memory_resource(start
, size
);
1087 if (!node_online(nid
)) {
1088 pgdat
= hotadd_new_pgdat(nid
, start
);
1095 /* call arch's memory hotadd */
1096 ret
= arch_add_memory(nid
, start
, size
);
1101 /* we online node here. we can't roll back from here. */
1102 node_set_online(nid
);
1105 ret
= register_one_node(nid
);
1107 * If sysfs file of new node can't create, cpu on the node
1108 * can't be hot-added. There is no rollback way now.
1109 * So, check by BUG_ON() to catch it reluctantly..
1114 /* create new memmap entry */
1115 firmware_map_add_hotplug(start
, start
+ size
, "System RAM");
1120 /* rollback pgdat allocation and others */
1122 rollback_node_hotadd(nid
, pgdat
);
1123 release_memory_resource(res
);
1126 unlock_memory_hotplug();
1129 EXPORT_SYMBOL_GPL(add_memory
);
1131 #ifdef CONFIG_MEMORY_HOTREMOVE
1133 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1134 * set and the size of the free page is given by page_order(). Using this,
1135 * the function determines if the pageblock contains only free pages.
1136 * Due to buddy contraints, a free page at least the size of a pageblock will
1137 * be located at the start of the pageblock
1139 static inline int pageblock_free(struct page
*page
)
1141 return PageBuddy(page
) && page_order(page
) >= pageblock_order
;
1144 /* Return the start of the next active pageblock after a given page */
1145 static struct page
*next_active_pageblock(struct page
*page
)
1147 /* Ensure the starting page is pageblock-aligned */
1148 BUG_ON(page_to_pfn(page
) & (pageblock_nr_pages
- 1));
1150 /* If the entire pageblock is free, move to the end of free page */
1151 if (pageblock_free(page
)) {
1153 /* be careful. we don't have locks, page_order can be changed.*/
1154 order
= page_order(page
);
1155 if ((order
< MAX_ORDER
) && (order
>= pageblock_order
))
1156 return page
+ (1 << order
);
1159 return page
+ pageblock_nr_pages
;
1162 /* Checks if this range of memory is likely to be hot-removable. */
1163 int is_mem_section_removable(unsigned long start_pfn
, unsigned long nr_pages
)
1165 struct page
*page
= pfn_to_page(start_pfn
);
1166 struct page
*end_page
= page
+ nr_pages
;
1168 /* Check the starting page of each pageblock within the range */
1169 for (; page
< end_page
; page
= next_active_pageblock(page
)) {
1170 if (!is_pageblock_removable_nolock(page
))
1175 /* All pageblocks in the memory block are likely to be hot-removable */
1180 * Confirm all pages in a range [start, end) is belongs to the same zone.
1182 static int test_pages_in_a_zone(unsigned long start_pfn
, unsigned long end_pfn
)
1185 struct zone
*zone
= NULL
;
1188 for (pfn
= start_pfn
;
1190 pfn
+= MAX_ORDER_NR_PAGES
) {
1192 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1193 while ((i
< MAX_ORDER_NR_PAGES
) && !pfn_valid_within(pfn
+ i
))
1195 if (i
== MAX_ORDER_NR_PAGES
)
1197 page
= pfn_to_page(pfn
+ i
);
1198 if (zone
&& page_zone(page
) != zone
)
1200 zone
= page_zone(page
);
1206 * Scanning pfn is much easier than scanning lru list.
1207 * Scan pfn from start to end and Find LRU page.
1209 static unsigned long scan_lru_pages(unsigned long start
, unsigned long end
)
1213 for (pfn
= start
; pfn
< end
; pfn
++) {
1214 if (pfn_valid(pfn
)) {
1215 page
= pfn_to_page(pfn
);
1223 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1225 do_migrate_range(unsigned long start_pfn
, unsigned long end_pfn
)
1229 int move_pages
= NR_OFFLINE_AT_ONCE_PAGES
;
1230 int not_managed
= 0;
1234 for (pfn
= start_pfn
; pfn
< end_pfn
&& move_pages
> 0; pfn
++) {
1235 if (!pfn_valid(pfn
))
1237 page
= pfn_to_page(pfn
);
1238 if (!get_page_unless_zero(page
))
1241 * We can skip free pages. And we can only deal with pages on
1244 ret
= isolate_lru_page(page
);
1245 if (!ret
) { /* Success */
1247 list_add_tail(&page
->lru
, &source
);
1249 inc_zone_page_state(page
, NR_ISOLATED_ANON
+
1250 page_is_file_cache(page
));
1253 #ifdef CONFIG_DEBUG_VM
1254 printk(KERN_ALERT
"removing pfn %lx from LRU failed\n",
1259 /* Because we don't have big zone->lock. we should
1260 check this again here. */
1261 if (page_count(page
)) {
1268 if (!list_empty(&source
)) {
1270 putback_lru_pages(&source
);
1275 * alloc_migrate_target should be improooooved!!
1276 * migrate_pages returns # of failed pages.
1278 ret
= migrate_pages(&source
, alloc_migrate_target
, 0,
1282 putback_lru_pages(&source
);
1289 * remove from free_area[] and mark all as Reserved.
1292 offline_isolated_pages_cb(unsigned long start
, unsigned long nr_pages
,
1295 __offline_isolated_pages(start
, start
+ nr_pages
);
1300 offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
1302 walk_system_ram_range(start_pfn
, end_pfn
- start_pfn
, NULL
,
1303 offline_isolated_pages_cb
);
1307 * Check all pages in range, recoreded as memory resource, are isolated.
1310 check_pages_isolated_cb(unsigned long start_pfn
, unsigned long nr_pages
,
1314 long offlined
= *(long *)data
;
1315 ret
= test_pages_isolated(start_pfn
, start_pfn
+ nr_pages
, true);
1316 offlined
= nr_pages
;
1318 *(long *)data
+= offlined
;
1323 check_pages_isolated(unsigned long start_pfn
, unsigned long end_pfn
)
1328 ret
= walk_system_ram_range(start_pfn
, end_pfn
- start_pfn
, &offlined
,
1329 check_pages_isolated_cb
);
1331 offlined
= (long)ret
;
1335 #ifdef CONFIG_MOVABLE_NODE
1337 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1340 static bool can_offline_normal(struct zone
*zone
, unsigned long nr_pages
)
1344 #else /* CONFIG_MOVABLE_NODE */
1345 /* ensure the node has NORMAL memory if it is still online */
1346 static bool can_offline_normal(struct zone
*zone
, unsigned long nr_pages
)
1348 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1349 unsigned long present_pages
= 0;
1352 for (zt
= 0; zt
<= ZONE_NORMAL
; zt
++)
1353 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1355 if (present_pages
> nr_pages
)
1359 for (; zt
<= ZONE_MOVABLE
; zt
++)
1360 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1363 * we can't offline the last normal memory until all
1364 * higher memory is offlined.
1366 return present_pages
== 0;
1368 #endif /* CONFIG_MOVABLE_NODE */
1370 /* check which state of node_states will be changed when offline memory */
1371 static void node_states_check_changes_offline(unsigned long nr_pages
,
1372 struct zone
*zone
, struct memory_notify
*arg
)
1374 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1375 unsigned long present_pages
= 0;
1376 enum zone_type zt
, zone_last
= ZONE_NORMAL
;
1379 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1380 * contains nodes which have zones of 0...ZONE_NORMAL,
1381 * set zone_last to ZONE_NORMAL.
1383 * If we don't have HIGHMEM nor movable node,
1384 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1385 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1387 if (N_MEMORY
== N_NORMAL_MEMORY
)
1388 zone_last
= ZONE_MOVABLE
;
1391 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1392 * If the memory to be offline is in a zone of 0...zone_last,
1393 * and it is the last present memory, 0...zone_last will
1394 * become empty after offline , thus we can determind we will
1395 * need to clear the node from node_states[N_NORMAL_MEMORY].
1397 for (zt
= 0; zt
<= zone_last
; zt
++)
1398 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1399 if (zone_idx(zone
) <= zone_last
&& nr_pages
>= present_pages
)
1400 arg
->status_change_nid_normal
= zone_to_nid(zone
);
1402 arg
->status_change_nid_normal
= -1;
1404 #ifdef CONFIG_HIGHMEM
1406 * If we have movable node, node_states[N_HIGH_MEMORY]
1407 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1408 * set zone_last to ZONE_HIGHMEM.
1410 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1411 * contains nodes which have zones of 0...ZONE_MOVABLE,
1412 * set zone_last to ZONE_MOVABLE.
1414 zone_last
= ZONE_HIGHMEM
;
1415 if (N_MEMORY
== N_HIGH_MEMORY
)
1416 zone_last
= ZONE_MOVABLE
;
1418 for (; zt
<= zone_last
; zt
++)
1419 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1420 if (zone_idx(zone
) <= zone_last
&& nr_pages
>= present_pages
)
1421 arg
->status_change_nid_high
= zone_to_nid(zone
);
1423 arg
->status_change_nid_high
= -1;
1425 arg
->status_change_nid_high
= arg
->status_change_nid_normal
;
1429 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1431 zone_last
= ZONE_MOVABLE
;
1434 * check whether node_states[N_HIGH_MEMORY] will be changed
1435 * If we try to offline the last present @nr_pages from the node,
1436 * we can determind we will need to clear the node from
1437 * node_states[N_HIGH_MEMORY].
1439 for (; zt
<= zone_last
; zt
++)
1440 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1441 if (nr_pages
>= present_pages
)
1442 arg
->status_change_nid
= zone_to_nid(zone
);
1444 arg
->status_change_nid
= -1;
1447 static void node_states_clear_node(int node
, struct memory_notify
*arg
)
1449 if (arg
->status_change_nid_normal
>= 0)
1450 node_clear_state(node
, N_NORMAL_MEMORY
);
1452 if ((N_MEMORY
!= N_NORMAL_MEMORY
) &&
1453 (arg
->status_change_nid_high
>= 0))
1454 node_clear_state(node
, N_HIGH_MEMORY
);
1456 if ((N_MEMORY
!= N_HIGH_MEMORY
) &&
1457 (arg
->status_change_nid
>= 0))
1458 node_clear_state(node
, N_MEMORY
);
1461 static int __ref
__offline_pages(unsigned long start_pfn
,
1462 unsigned long end_pfn
, unsigned long timeout
)
1464 unsigned long pfn
, nr_pages
, expire
;
1465 long offlined_pages
;
1466 int ret
, drain
, retry_max
, node
;
1468 struct memory_notify arg
;
1470 BUG_ON(start_pfn
>= end_pfn
);
1471 /* at least, alignment against pageblock is necessary */
1472 if (!IS_ALIGNED(start_pfn
, pageblock_nr_pages
))
1474 if (!IS_ALIGNED(end_pfn
, pageblock_nr_pages
))
1476 /* This makes hotplug much easier...and readable.
1477 we assume this for now. .*/
1478 if (!test_pages_in_a_zone(start_pfn
, end_pfn
))
1481 lock_memory_hotplug();
1483 zone
= page_zone(pfn_to_page(start_pfn
));
1484 node
= zone_to_nid(zone
);
1485 nr_pages
= end_pfn
- start_pfn
;
1488 if (zone_idx(zone
) <= ZONE_NORMAL
&& !can_offline_normal(zone
, nr_pages
))
1491 /* set above range as isolated */
1492 ret
= start_isolate_page_range(start_pfn
, end_pfn
,
1493 MIGRATE_MOVABLE
, true);
1497 arg
.start_pfn
= start_pfn
;
1498 arg
.nr_pages
= nr_pages
;
1499 node_states_check_changes_offline(nr_pages
, zone
, &arg
);
1501 ret
= memory_notify(MEM_GOING_OFFLINE
, &arg
);
1502 ret
= notifier_to_errno(ret
);
1504 goto failed_removal
;
1507 expire
= jiffies
+ timeout
;
1511 /* start memory hot removal */
1513 if (time_after(jiffies
, expire
))
1514 goto failed_removal
;
1516 if (signal_pending(current
))
1517 goto failed_removal
;
1520 lru_add_drain_all();
1525 pfn
= scan_lru_pages(start_pfn
, end_pfn
);
1526 if (pfn
) { /* We have page on LRU */
1527 ret
= do_migrate_range(pfn
, end_pfn
);
1533 if (--retry_max
== 0)
1534 goto failed_removal
;
1540 /* drain all zone's lru pagevec, this is asynchronous... */
1541 lru_add_drain_all();
1543 /* drain pcp pages, this is synchronous. */
1546 offlined_pages
= check_pages_isolated(start_pfn
, end_pfn
);
1547 if (offlined_pages
< 0) {
1549 goto failed_removal
;
1551 printk(KERN_INFO
"Offlined Pages %ld\n", offlined_pages
);
1552 /* Ok, all of our target is isolated.
1553 We cannot do rollback at this point. */
1554 offline_isolated_pages(start_pfn
, end_pfn
);
1555 /* reset pagetype flags and makes migrate type to be MOVABLE */
1556 undo_isolate_page_range(start_pfn
, end_pfn
, MIGRATE_MOVABLE
);
1557 /* removal success */
1558 zone
->managed_pages
-= offlined_pages
;
1559 zone
->present_pages
-= offlined_pages
;
1560 zone
->zone_pgdat
->node_present_pages
-= offlined_pages
;
1561 totalram_pages
-= offlined_pages
;
1563 init_per_zone_wmark_min();
1565 if (!populated_zone(zone
)) {
1566 zone_pcp_reset(zone
);
1567 mutex_lock(&zonelists_mutex
);
1568 build_all_zonelists(NULL
, NULL
);
1569 mutex_unlock(&zonelists_mutex
);
1571 zone_pcp_update(zone
);
1573 node_states_clear_node(node
, &arg
);
1574 if (arg
.status_change_nid
>= 0)
1577 vm_total_pages
= nr_free_pagecache_pages();
1578 writeback_set_ratelimit();
1580 memory_notify(MEM_OFFLINE
, &arg
);
1581 unlock_memory_hotplug();
1585 printk(KERN_INFO
"memory offlining [mem %#010llx-%#010llx] failed\n",
1586 (unsigned long long) start_pfn
<< PAGE_SHIFT
,
1587 ((unsigned long long) end_pfn
<< PAGE_SHIFT
) - 1);
1588 memory_notify(MEM_CANCEL_OFFLINE
, &arg
);
1589 /* pushback to free area */
1590 undo_isolate_page_range(start_pfn
, end_pfn
, MIGRATE_MOVABLE
);
1593 unlock_memory_hotplug();
1597 int offline_pages(unsigned long start_pfn
, unsigned long nr_pages
)
1599 return __offline_pages(start_pfn
, start_pfn
+ nr_pages
, 120 * HZ
);
1603 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1604 * @start_pfn: start pfn of the memory range
1605 * @end_pfn: end pft of the memory range
1606 * @arg: argument passed to func
1607 * @func: callback for each memory section walked
1609 * This function walks through all present mem sections in range
1610 * [start_pfn, end_pfn) and call func on each mem section.
1612 * Returns the return value of func.
1614 static int walk_memory_range(unsigned long start_pfn
, unsigned long end_pfn
,
1615 void *arg
, int (*func
)(struct memory_block
*, void *))
1617 struct memory_block
*mem
= NULL
;
1618 struct mem_section
*section
;
1619 unsigned long pfn
, section_nr
;
1622 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
1623 section_nr
= pfn_to_section_nr(pfn
);
1624 if (!present_section_nr(section_nr
))
1627 section
= __nr_to_section(section_nr
);
1628 /* same memblock? */
1630 if ((section_nr
>= mem
->start_section_nr
) &&
1631 (section_nr
<= mem
->end_section_nr
))
1634 mem
= find_memory_block_hinted(section
, mem
);
1638 ret
= func(mem
, arg
);
1640 kobject_put(&mem
->dev
.kobj
);
1646 kobject_put(&mem
->dev
.kobj
);
1652 * offline_memory_block_cb - callback function for offlining memory block
1653 * @mem: the memory block to be offlined
1654 * @arg: buffer to hold error msg
1656 * Always return 0, and put the error msg in arg if any.
1658 static int offline_memory_block_cb(struct memory_block
*mem
, void *arg
)
1661 int error
= offline_memory_block(mem
);
1663 if (error
!= 0 && *ret
== 0)
1669 static int is_memblock_offlined_cb(struct memory_block
*mem
, void *arg
)
1671 int ret
= !is_memblock_offlined(mem
);
1674 pr_warn("removing memory fails, because memory "
1675 "[%#010llx-%#010llx] is onlined\n",
1676 PFN_PHYS(section_nr_to_pfn(mem
->start_section_nr
)),
1677 PFN_PHYS(section_nr_to_pfn(mem
->end_section_nr
+ 1))-1);
1682 int __ref
remove_memory(u64 start
, u64 size
)
1684 unsigned long start_pfn
, end_pfn
;
1688 start_pfn
= PFN_DOWN(start
);
1689 end_pfn
= start_pfn
+ PFN_DOWN(size
);
1692 * When CONFIG_MEMCG is on, one memory block may be used by other
1693 * blocks to store page cgroup when onlining pages. But we don't know
1694 * in what order pages are onlined. So we iterate twice to offline
1696 * 1st iterate: offline every non primary memory block.
1697 * 2nd iterate: offline primary (i.e. first added) memory block.
1700 walk_memory_range(start_pfn
, end_pfn
, &ret
,
1701 offline_memory_block_cb
);
1711 lock_memory_hotplug();
1714 * we have offlined all memory blocks like this:
1715 * 1. lock memory hotplug
1716 * 2. offline a memory block
1717 * 3. unlock memory hotplug
1719 * repeat step1-3 to offline the memory block. All memory blocks
1720 * must be offlined before removing memory. But we don't hold the
1721 * lock in the whole operation. So we should check whether all
1722 * memory blocks are offlined.
1725 ret
= walk_memory_range(start_pfn
, end_pfn
, NULL
,
1726 is_memblock_offlined_cb
);
1728 unlock_memory_hotplug();
1732 /* remove memmap entry */
1733 firmware_map_remove(start
, start
+ size
, "System RAM");
1735 arch_remove_memory(start
, size
);
1737 unlock_memory_hotplug();
1742 int offline_pages(unsigned long start_pfn
, unsigned long nr_pages
)
1746 int remove_memory(u64 start
, u64 size
)
1750 #endif /* CONFIG_MEMORY_HOTREMOVE */
1751 EXPORT_SYMBOL_GPL(remove_memory
);