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 static int __remove_section(struct zone
*zone
, struct mem_section
*ms
)
439 if (!valid_section(ms
))
442 ret
= unregister_memory_section(ms
);
446 sparse_remove_one_section(zone
, ms
);
451 * Reasonably generic function for adding memory. It is
452 * expected that archs that support memory hotplug will
453 * call this function after deciding the zone to which to
456 int __ref
__add_pages(int nid
, struct zone
*zone
, unsigned long phys_start_pfn
,
457 unsigned long nr_pages
)
461 int start_sec
, end_sec
;
462 /* during initialize mem_map, align hot-added range to section */
463 start_sec
= pfn_to_section_nr(phys_start_pfn
);
464 end_sec
= pfn_to_section_nr(phys_start_pfn
+ nr_pages
- 1);
466 for (i
= start_sec
; i
<= end_sec
; i
++) {
467 err
= __add_section(nid
, zone
, i
<< PFN_SECTION_SHIFT
);
470 * EEXIST is finally dealt with by ioresource collision
471 * check. see add_memory() => register_memory_resource()
472 * Warning will be printed if there is collision.
474 if (err
&& (err
!= -EEXIST
))
481 EXPORT_SYMBOL_GPL(__add_pages
);
484 * __remove_pages() - remove sections of pages from a zone
485 * @zone: zone from which pages need to be removed
486 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
487 * @nr_pages: number of pages to remove (must be multiple of section size)
489 * Generic helper function to remove section mappings and sysfs entries
490 * for the section of the memory we are removing. Caller needs to make
491 * sure that pages are marked reserved and zones are adjust properly by
492 * calling offline_pages().
494 int __remove_pages(struct zone
*zone
, unsigned long phys_start_pfn
,
495 unsigned long nr_pages
)
497 unsigned long i
, ret
= 0;
498 int sections_to_remove
;
501 * We can only remove entire sections
503 BUG_ON(phys_start_pfn
& ~PAGE_SECTION_MASK
);
504 BUG_ON(nr_pages
% PAGES_PER_SECTION
);
506 release_mem_region(phys_start_pfn
<< PAGE_SHIFT
, nr_pages
* PAGE_SIZE
);
508 sections_to_remove
= nr_pages
/ PAGES_PER_SECTION
;
509 for (i
= 0; i
< sections_to_remove
; i
++) {
510 unsigned long pfn
= phys_start_pfn
+ i
*PAGES_PER_SECTION
;
511 ret
= __remove_section(zone
, __pfn_to_section(pfn
));
517 EXPORT_SYMBOL_GPL(__remove_pages
);
519 int set_online_page_callback(online_page_callback_t callback
)
523 lock_memory_hotplug();
525 if (online_page_callback
== generic_online_page
) {
526 online_page_callback
= callback
;
530 unlock_memory_hotplug();
534 EXPORT_SYMBOL_GPL(set_online_page_callback
);
536 int restore_online_page_callback(online_page_callback_t callback
)
540 lock_memory_hotplug();
542 if (online_page_callback
== callback
) {
543 online_page_callback
= generic_online_page
;
547 unlock_memory_hotplug();
551 EXPORT_SYMBOL_GPL(restore_online_page_callback
);
553 void __online_page_set_limits(struct page
*page
)
555 unsigned long pfn
= page_to_pfn(page
);
557 if (pfn
>= num_physpages
)
558 num_physpages
= pfn
+ 1;
560 EXPORT_SYMBOL_GPL(__online_page_set_limits
);
562 void __online_page_increment_counters(struct page
*page
)
566 #ifdef CONFIG_HIGHMEM
567 if (PageHighMem(page
))
571 EXPORT_SYMBOL_GPL(__online_page_increment_counters
);
573 void __online_page_free(struct page
*page
)
575 ClearPageReserved(page
);
576 init_page_count(page
);
579 EXPORT_SYMBOL_GPL(__online_page_free
);
581 static void generic_online_page(struct page
*page
)
583 __online_page_set_limits(page
);
584 __online_page_increment_counters(page
);
585 __online_page_free(page
);
588 static int online_pages_range(unsigned long start_pfn
, unsigned long nr_pages
,
592 unsigned long onlined_pages
= *(unsigned long *)arg
;
594 if (PageReserved(pfn_to_page(start_pfn
)))
595 for (i
= 0; i
< nr_pages
; i
++) {
596 page
= pfn_to_page(start_pfn
+ i
);
597 (*online_page_callback
)(page
);
600 *(unsigned long *)arg
= onlined_pages
;
604 #ifdef CONFIG_MOVABLE_NODE
606 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
609 static bool can_online_high_movable(struct zone
*zone
)
613 #else /* CONFIG_MOVABLE_NODE */
614 /* ensure every online node has NORMAL memory */
615 static bool can_online_high_movable(struct zone
*zone
)
617 return node_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
619 #endif /* CONFIG_MOVABLE_NODE */
621 /* check which state of node_states will be changed when online memory */
622 static void node_states_check_changes_online(unsigned long nr_pages
,
623 struct zone
*zone
, struct memory_notify
*arg
)
625 int nid
= zone_to_nid(zone
);
626 enum zone_type zone_last
= ZONE_NORMAL
;
629 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
630 * contains nodes which have zones of 0...ZONE_NORMAL,
631 * set zone_last to ZONE_NORMAL.
633 * If we don't have HIGHMEM nor movable node,
634 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
635 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
637 if (N_MEMORY
== N_NORMAL_MEMORY
)
638 zone_last
= ZONE_MOVABLE
;
641 * if the memory to be online is in a zone of 0...zone_last, and
642 * the zones of 0...zone_last don't have memory before online, we will
643 * need to set the node to node_states[N_NORMAL_MEMORY] after
644 * the memory is online.
646 if (zone_idx(zone
) <= zone_last
&& !node_state(nid
, N_NORMAL_MEMORY
))
647 arg
->status_change_nid_normal
= nid
;
649 arg
->status_change_nid_normal
= -1;
651 #ifdef CONFIG_HIGHMEM
653 * If we have movable node, node_states[N_HIGH_MEMORY]
654 * contains nodes which have zones of 0...ZONE_HIGHMEM,
655 * set zone_last to ZONE_HIGHMEM.
657 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
658 * contains nodes which have zones of 0...ZONE_MOVABLE,
659 * set zone_last to ZONE_MOVABLE.
661 zone_last
= ZONE_HIGHMEM
;
662 if (N_MEMORY
== N_HIGH_MEMORY
)
663 zone_last
= ZONE_MOVABLE
;
665 if (zone_idx(zone
) <= zone_last
&& !node_state(nid
, N_HIGH_MEMORY
))
666 arg
->status_change_nid_high
= nid
;
668 arg
->status_change_nid_high
= -1;
670 arg
->status_change_nid_high
= arg
->status_change_nid_normal
;
674 * if the node don't have memory befor online, we will need to
675 * set the node to node_states[N_MEMORY] after the memory
678 if (!node_state(nid
, N_MEMORY
))
679 arg
->status_change_nid
= nid
;
681 arg
->status_change_nid
= -1;
684 static void node_states_set_node(int node
, struct memory_notify
*arg
)
686 if (arg
->status_change_nid_normal
>= 0)
687 node_set_state(node
, N_NORMAL_MEMORY
);
689 if (arg
->status_change_nid_high
>= 0)
690 node_set_state(node
, N_HIGH_MEMORY
);
692 node_set_state(node
, N_MEMORY
);
696 int __ref
online_pages(unsigned long pfn
, unsigned long nr_pages
, int online_type
)
698 unsigned long onlined_pages
= 0;
700 int need_zonelists_rebuild
= 0;
703 struct memory_notify arg
;
705 lock_memory_hotplug();
707 * This doesn't need a lock to do pfn_to_page().
708 * The section can't be removed here because of the
709 * memory_block->state_mutex.
711 zone
= page_zone(pfn_to_page(pfn
));
713 if ((zone_idx(zone
) > ZONE_NORMAL
|| online_type
== ONLINE_MOVABLE
) &&
714 !can_online_high_movable(zone
)) {
715 unlock_memory_hotplug();
719 if (online_type
== ONLINE_KERNEL
&& zone_idx(zone
) == ZONE_MOVABLE
) {
720 if (move_pfn_range_left(zone
- 1, zone
, pfn
, pfn
+ nr_pages
)) {
721 unlock_memory_hotplug();
725 if (online_type
== ONLINE_MOVABLE
&& zone_idx(zone
) == ZONE_MOVABLE
- 1) {
726 if (move_pfn_range_right(zone
, zone
+ 1, pfn
, pfn
+ nr_pages
)) {
727 unlock_memory_hotplug();
732 /* Previous code may changed the zone of the pfn range */
733 zone
= page_zone(pfn_to_page(pfn
));
736 arg
.nr_pages
= nr_pages
;
737 node_states_check_changes_online(nr_pages
, zone
, &arg
);
739 nid
= page_to_nid(pfn_to_page(pfn
));
741 ret
= memory_notify(MEM_GOING_ONLINE
, &arg
);
742 ret
= notifier_to_errno(ret
);
744 memory_notify(MEM_CANCEL_ONLINE
, &arg
);
745 unlock_memory_hotplug();
749 * If this zone is not populated, then it is not in zonelist.
750 * This means the page allocator ignores this zone.
751 * So, zonelist must be updated after online.
753 mutex_lock(&zonelists_mutex
);
754 if (!populated_zone(zone
)) {
755 need_zonelists_rebuild
= 1;
756 build_all_zonelists(NULL
, zone
);
759 ret
= walk_system_ram_range(pfn
, nr_pages
, &onlined_pages
,
762 if (need_zonelists_rebuild
)
763 zone_pcp_reset(zone
);
764 mutex_unlock(&zonelists_mutex
);
765 printk(KERN_DEBUG
"online_pages [mem %#010llx-%#010llx] failed\n",
766 (unsigned long long) pfn
<< PAGE_SHIFT
,
767 (((unsigned long long) pfn
+ nr_pages
)
769 memory_notify(MEM_CANCEL_ONLINE
, &arg
);
770 unlock_memory_hotplug();
774 zone
->managed_pages
+= onlined_pages
;
775 zone
->present_pages
+= onlined_pages
;
776 zone
->zone_pgdat
->node_present_pages
+= onlined_pages
;
778 node_states_set_node(zone_to_nid(zone
), &arg
);
779 if (need_zonelists_rebuild
)
780 build_all_zonelists(NULL
, NULL
);
782 zone_pcp_update(zone
);
785 mutex_unlock(&zonelists_mutex
);
787 init_per_zone_wmark_min();
790 kswapd_run(zone_to_nid(zone
));
792 vm_total_pages
= nr_free_pagecache_pages();
794 writeback_set_ratelimit();
797 memory_notify(MEM_ONLINE
, &arg
);
798 unlock_memory_hotplug();
802 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
804 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
805 static pg_data_t __ref
*hotadd_new_pgdat(int nid
, u64 start
)
807 struct pglist_data
*pgdat
;
808 unsigned long zones_size
[MAX_NR_ZONES
] = {0};
809 unsigned long zholes_size
[MAX_NR_ZONES
] = {0};
810 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
812 pgdat
= arch_alloc_nodedata(nid
);
816 arch_refresh_nodedata(nid
, pgdat
);
818 /* we can use NODE_DATA(nid) from here */
820 /* init node's zones as empty zones, we don't have any present pages.*/
821 free_area_init_node(nid
, zones_size
, start_pfn
, zholes_size
);
824 * The node we allocated has no zone fallback lists. For avoiding
825 * to access not-initialized zonelist, build here.
827 mutex_lock(&zonelists_mutex
);
828 build_all_zonelists(pgdat
, NULL
);
829 mutex_unlock(&zonelists_mutex
);
834 static void rollback_node_hotadd(int nid
, pg_data_t
*pgdat
)
836 arch_refresh_nodedata(nid
, NULL
);
837 arch_free_nodedata(pgdat
);
843 * called by cpu_up() to online a node without onlined memory.
845 int mem_online_node(int nid
)
850 lock_memory_hotplug();
851 pgdat
= hotadd_new_pgdat(nid
, 0);
856 node_set_online(nid
);
857 ret
= register_one_node(nid
);
861 unlock_memory_hotplug();
865 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
866 int __ref
add_memory(int nid
, u64 start
, u64 size
)
868 pg_data_t
*pgdat
= NULL
;
870 struct resource
*res
;
873 lock_memory_hotplug();
875 res
= register_memory_resource(start
, size
);
880 if (!node_online(nid
)) {
881 pgdat
= hotadd_new_pgdat(nid
, start
);
888 /* call arch's memory hotadd */
889 ret
= arch_add_memory(nid
, start
, size
);
894 /* we online node here. we can't roll back from here. */
895 node_set_online(nid
);
898 ret
= register_one_node(nid
);
900 * If sysfs file of new node can't create, cpu on the node
901 * can't be hot-added. There is no rollback way now.
902 * So, check by BUG_ON() to catch it reluctantly..
907 /* create new memmap entry */
908 firmware_map_add_hotplug(start
, start
+ size
, "System RAM");
913 /* rollback pgdat allocation and others */
915 rollback_node_hotadd(nid
, pgdat
);
916 release_memory_resource(res
);
919 unlock_memory_hotplug();
922 EXPORT_SYMBOL_GPL(add_memory
);
924 #ifdef CONFIG_MEMORY_HOTREMOVE
926 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
927 * set and the size of the free page is given by page_order(). Using this,
928 * the function determines if the pageblock contains only free pages.
929 * Due to buddy contraints, a free page at least the size of a pageblock will
930 * be located at the start of the pageblock
932 static inline int pageblock_free(struct page
*page
)
934 return PageBuddy(page
) && page_order(page
) >= pageblock_order
;
937 /* Return the start of the next active pageblock after a given page */
938 static struct page
*next_active_pageblock(struct page
*page
)
940 /* Ensure the starting page is pageblock-aligned */
941 BUG_ON(page_to_pfn(page
) & (pageblock_nr_pages
- 1));
943 /* If the entire pageblock is free, move to the end of free page */
944 if (pageblock_free(page
)) {
946 /* be careful. we don't have locks, page_order can be changed.*/
947 order
= page_order(page
);
948 if ((order
< MAX_ORDER
) && (order
>= pageblock_order
))
949 return page
+ (1 << order
);
952 return page
+ pageblock_nr_pages
;
955 /* Checks if this range of memory is likely to be hot-removable. */
956 int is_mem_section_removable(unsigned long start_pfn
, unsigned long nr_pages
)
958 struct page
*page
= pfn_to_page(start_pfn
);
959 struct page
*end_page
= page
+ nr_pages
;
961 /* Check the starting page of each pageblock within the range */
962 for (; page
< end_page
; page
= next_active_pageblock(page
)) {
963 if (!is_pageblock_removable_nolock(page
))
968 /* All pageblocks in the memory block are likely to be hot-removable */
973 * Confirm all pages in a range [start, end) is belongs to the same zone.
975 static int test_pages_in_a_zone(unsigned long start_pfn
, unsigned long end_pfn
)
978 struct zone
*zone
= NULL
;
981 for (pfn
= start_pfn
;
983 pfn
+= MAX_ORDER_NR_PAGES
) {
985 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
986 while ((i
< MAX_ORDER_NR_PAGES
) && !pfn_valid_within(pfn
+ i
))
988 if (i
== MAX_ORDER_NR_PAGES
)
990 page
= pfn_to_page(pfn
+ i
);
991 if (zone
&& page_zone(page
) != zone
)
993 zone
= page_zone(page
);
999 * Scanning pfn is much easier than scanning lru list.
1000 * Scan pfn from start to end and Find LRU page.
1002 static unsigned long scan_lru_pages(unsigned long start
, unsigned long end
)
1006 for (pfn
= start
; pfn
< end
; pfn
++) {
1007 if (pfn_valid(pfn
)) {
1008 page
= pfn_to_page(pfn
);
1016 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1018 do_migrate_range(unsigned long start_pfn
, unsigned long end_pfn
)
1022 int move_pages
= NR_OFFLINE_AT_ONCE_PAGES
;
1023 int not_managed
= 0;
1027 for (pfn
= start_pfn
; pfn
< end_pfn
&& move_pages
> 0; pfn
++) {
1028 if (!pfn_valid(pfn
))
1030 page
= pfn_to_page(pfn
);
1031 if (!get_page_unless_zero(page
))
1034 * We can skip free pages. And we can only deal with pages on
1037 ret
= isolate_lru_page(page
);
1038 if (!ret
) { /* Success */
1040 list_add_tail(&page
->lru
, &source
);
1042 inc_zone_page_state(page
, NR_ISOLATED_ANON
+
1043 page_is_file_cache(page
));
1046 #ifdef CONFIG_DEBUG_VM
1047 printk(KERN_ALERT
"removing pfn %lx from LRU failed\n",
1052 /* Because we don't have big zone->lock. we should
1053 check this again here. */
1054 if (page_count(page
)) {
1061 if (!list_empty(&source
)) {
1063 putback_lru_pages(&source
);
1068 * alloc_migrate_target should be improooooved!!
1069 * migrate_pages returns # of failed pages.
1071 ret
= migrate_pages(&source
, alloc_migrate_target
, 0,
1075 putback_lru_pages(&source
);
1082 * remove from free_area[] and mark all as Reserved.
1085 offline_isolated_pages_cb(unsigned long start
, unsigned long nr_pages
,
1088 __offline_isolated_pages(start
, start
+ nr_pages
);
1093 offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
1095 walk_system_ram_range(start_pfn
, end_pfn
- start_pfn
, NULL
,
1096 offline_isolated_pages_cb
);
1100 * Check all pages in range, recoreded as memory resource, are isolated.
1103 check_pages_isolated_cb(unsigned long start_pfn
, unsigned long nr_pages
,
1107 long offlined
= *(long *)data
;
1108 ret
= test_pages_isolated(start_pfn
, start_pfn
+ nr_pages
, true);
1109 offlined
= nr_pages
;
1111 *(long *)data
+= offlined
;
1116 check_pages_isolated(unsigned long start_pfn
, unsigned long end_pfn
)
1121 ret
= walk_system_ram_range(start_pfn
, end_pfn
- start_pfn
, &offlined
,
1122 check_pages_isolated_cb
);
1124 offlined
= (long)ret
;
1128 #ifdef CONFIG_MOVABLE_NODE
1130 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1133 static bool can_offline_normal(struct zone
*zone
, unsigned long nr_pages
)
1137 #else /* CONFIG_MOVABLE_NODE */
1138 /* ensure the node has NORMAL memory if it is still online */
1139 static bool can_offline_normal(struct zone
*zone
, unsigned long nr_pages
)
1141 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1142 unsigned long present_pages
= 0;
1145 for (zt
= 0; zt
<= ZONE_NORMAL
; zt
++)
1146 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1148 if (present_pages
> nr_pages
)
1152 for (; zt
<= ZONE_MOVABLE
; zt
++)
1153 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1156 * we can't offline the last normal memory until all
1157 * higher memory is offlined.
1159 return present_pages
== 0;
1161 #endif /* CONFIG_MOVABLE_NODE */
1163 /* check which state of node_states will be changed when offline memory */
1164 static void node_states_check_changes_offline(unsigned long nr_pages
,
1165 struct zone
*zone
, struct memory_notify
*arg
)
1167 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1168 unsigned long present_pages
= 0;
1169 enum zone_type zt
, zone_last
= ZONE_NORMAL
;
1172 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1173 * contains nodes which have zones of 0...ZONE_NORMAL,
1174 * set zone_last to ZONE_NORMAL.
1176 * If we don't have HIGHMEM nor movable node,
1177 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1178 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1180 if (N_MEMORY
== N_NORMAL_MEMORY
)
1181 zone_last
= ZONE_MOVABLE
;
1184 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1185 * If the memory to be offline is in a zone of 0...zone_last,
1186 * and it is the last present memory, 0...zone_last will
1187 * become empty after offline , thus we can determind we will
1188 * need to clear the node from node_states[N_NORMAL_MEMORY].
1190 for (zt
= 0; zt
<= zone_last
; zt
++)
1191 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1192 if (zone_idx(zone
) <= zone_last
&& nr_pages
>= present_pages
)
1193 arg
->status_change_nid_normal
= zone_to_nid(zone
);
1195 arg
->status_change_nid_normal
= -1;
1197 #ifdef CONFIG_HIGHMEM
1199 * If we have movable node, node_states[N_HIGH_MEMORY]
1200 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1201 * set zone_last to ZONE_HIGHMEM.
1203 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1204 * contains nodes which have zones of 0...ZONE_MOVABLE,
1205 * set zone_last to ZONE_MOVABLE.
1207 zone_last
= ZONE_HIGHMEM
;
1208 if (N_MEMORY
== N_HIGH_MEMORY
)
1209 zone_last
= ZONE_MOVABLE
;
1211 for (; zt
<= zone_last
; zt
++)
1212 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1213 if (zone_idx(zone
) <= zone_last
&& nr_pages
>= present_pages
)
1214 arg
->status_change_nid_high
= zone_to_nid(zone
);
1216 arg
->status_change_nid_high
= -1;
1218 arg
->status_change_nid_high
= arg
->status_change_nid_normal
;
1222 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1224 zone_last
= ZONE_MOVABLE
;
1227 * check whether node_states[N_HIGH_MEMORY] will be changed
1228 * If we try to offline the last present @nr_pages from the node,
1229 * we can determind we will need to clear the node from
1230 * node_states[N_HIGH_MEMORY].
1232 for (; zt
<= zone_last
; zt
++)
1233 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1234 if (nr_pages
>= present_pages
)
1235 arg
->status_change_nid
= zone_to_nid(zone
);
1237 arg
->status_change_nid
= -1;
1240 static void node_states_clear_node(int node
, struct memory_notify
*arg
)
1242 if (arg
->status_change_nid_normal
>= 0)
1243 node_clear_state(node
, N_NORMAL_MEMORY
);
1245 if ((N_MEMORY
!= N_NORMAL_MEMORY
) &&
1246 (arg
->status_change_nid_high
>= 0))
1247 node_clear_state(node
, N_HIGH_MEMORY
);
1249 if ((N_MEMORY
!= N_HIGH_MEMORY
) &&
1250 (arg
->status_change_nid
>= 0))
1251 node_clear_state(node
, N_MEMORY
);
1254 static int __ref
__offline_pages(unsigned long start_pfn
,
1255 unsigned long end_pfn
, unsigned long timeout
)
1257 unsigned long pfn
, nr_pages
, expire
;
1258 long offlined_pages
;
1259 int ret
, drain
, retry_max
, node
;
1261 struct memory_notify arg
;
1263 BUG_ON(start_pfn
>= end_pfn
);
1264 /* at least, alignment against pageblock is necessary */
1265 if (!IS_ALIGNED(start_pfn
, pageblock_nr_pages
))
1267 if (!IS_ALIGNED(end_pfn
, pageblock_nr_pages
))
1269 /* This makes hotplug much easier...and readable.
1270 we assume this for now. .*/
1271 if (!test_pages_in_a_zone(start_pfn
, end_pfn
))
1274 lock_memory_hotplug();
1276 zone
= page_zone(pfn_to_page(start_pfn
));
1277 node
= zone_to_nid(zone
);
1278 nr_pages
= end_pfn
- start_pfn
;
1281 if (zone_idx(zone
) <= ZONE_NORMAL
&& !can_offline_normal(zone
, nr_pages
))
1284 /* set above range as isolated */
1285 ret
= start_isolate_page_range(start_pfn
, end_pfn
,
1286 MIGRATE_MOVABLE
, true);
1290 arg
.start_pfn
= start_pfn
;
1291 arg
.nr_pages
= nr_pages
;
1292 node_states_check_changes_offline(nr_pages
, zone
, &arg
);
1294 ret
= memory_notify(MEM_GOING_OFFLINE
, &arg
);
1295 ret
= notifier_to_errno(ret
);
1297 goto failed_removal
;
1300 expire
= jiffies
+ timeout
;
1304 /* start memory hot removal */
1306 if (time_after(jiffies
, expire
))
1307 goto failed_removal
;
1309 if (signal_pending(current
))
1310 goto failed_removal
;
1313 lru_add_drain_all();
1318 pfn
= scan_lru_pages(start_pfn
, end_pfn
);
1319 if (pfn
) { /* We have page on LRU */
1320 ret
= do_migrate_range(pfn
, end_pfn
);
1326 if (--retry_max
== 0)
1327 goto failed_removal
;
1333 /* drain all zone's lru pagevec, this is asynchronous... */
1334 lru_add_drain_all();
1336 /* drain pcp pages, this is synchronous. */
1339 offlined_pages
= check_pages_isolated(start_pfn
, end_pfn
);
1340 if (offlined_pages
< 0) {
1342 goto failed_removal
;
1344 printk(KERN_INFO
"Offlined Pages %ld\n", offlined_pages
);
1345 /* Ok, all of our target is isolated.
1346 We cannot do rollback at this point. */
1347 offline_isolated_pages(start_pfn
, end_pfn
);
1348 /* reset pagetype flags and makes migrate type to be MOVABLE */
1349 undo_isolate_page_range(start_pfn
, end_pfn
, MIGRATE_MOVABLE
);
1350 /* removal success */
1351 zone
->managed_pages
-= offlined_pages
;
1352 zone
->present_pages
-= offlined_pages
;
1353 zone
->zone_pgdat
->node_present_pages
-= offlined_pages
;
1354 totalram_pages
-= offlined_pages
;
1356 init_per_zone_wmark_min();
1358 if (!populated_zone(zone
)) {
1359 zone_pcp_reset(zone
);
1360 mutex_lock(&zonelists_mutex
);
1361 build_all_zonelists(NULL
, NULL
);
1362 mutex_unlock(&zonelists_mutex
);
1364 zone_pcp_update(zone
);
1366 node_states_clear_node(node
, &arg
);
1367 if (arg
.status_change_nid
>= 0)
1370 vm_total_pages
= nr_free_pagecache_pages();
1371 writeback_set_ratelimit();
1373 memory_notify(MEM_OFFLINE
, &arg
);
1374 unlock_memory_hotplug();
1378 printk(KERN_INFO
"memory offlining [mem %#010llx-%#010llx] failed\n",
1379 (unsigned long long) start_pfn
<< PAGE_SHIFT
,
1380 ((unsigned long long) end_pfn
<< PAGE_SHIFT
) - 1);
1381 memory_notify(MEM_CANCEL_OFFLINE
, &arg
);
1382 /* pushback to free area */
1383 undo_isolate_page_range(start_pfn
, end_pfn
, MIGRATE_MOVABLE
);
1386 unlock_memory_hotplug();
1390 int offline_pages(unsigned long start_pfn
, unsigned long nr_pages
)
1392 return __offline_pages(start_pfn
, start_pfn
+ nr_pages
, 120 * HZ
);
1396 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1397 * @start_pfn: start pfn of the memory range
1398 * @end_pfn: end pft of the memory range
1399 * @arg: argument passed to func
1400 * @func: callback for each memory section walked
1402 * This function walks through all present mem sections in range
1403 * [start_pfn, end_pfn) and call func on each mem section.
1405 * Returns the return value of func.
1407 static int walk_memory_range(unsigned long start_pfn
, unsigned long end_pfn
,
1408 void *arg
, int (*func
)(struct memory_block
*, void *))
1410 struct memory_block
*mem
= NULL
;
1411 struct mem_section
*section
;
1412 unsigned long pfn
, section_nr
;
1415 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
1416 section_nr
= pfn_to_section_nr(pfn
);
1417 if (!present_section_nr(section_nr
))
1420 section
= __nr_to_section(section_nr
);
1421 /* same memblock? */
1423 if ((section_nr
>= mem
->start_section_nr
) &&
1424 (section_nr
<= mem
->end_section_nr
))
1427 mem
= find_memory_block_hinted(section
, mem
);
1431 ret
= func(mem
, arg
);
1433 kobject_put(&mem
->dev
.kobj
);
1439 kobject_put(&mem
->dev
.kobj
);
1445 * offline_memory_block_cb - callback function for offlining memory block
1446 * @mem: the memory block to be offlined
1447 * @arg: buffer to hold error msg
1449 * Always return 0, and put the error msg in arg if any.
1451 static int offline_memory_block_cb(struct memory_block
*mem
, void *arg
)
1454 int error
= offline_memory_block(mem
);
1456 if (error
!= 0 && *ret
== 0)
1462 static int is_memblock_offlined_cb(struct memory_block
*mem
, void *arg
)
1464 int ret
= !is_memblock_offlined(mem
);
1467 pr_warn("removing memory fails, because memory "
1468 "[%#010llx-%#010llx] is onlined\n",
1469 PFN_PHYS(section_nr_to_pfn(mem
->start_section_nr
)),
1470 PFN_PHYS(section_nr_to_pfn(mem
->end_section_nr
+ 1))-1);
1475 int __ref
remove_memory(u64 start
, u64 size
)
1477 unsigned long start_pfn
, end_pfn
;
1481 start_pfn
= PFN_DOWN(start
);
1482 end_pfn
= start_pfn
+ PFN_DOWN(size
);
1485 * When CONFIG_MEMCG is on, one memory block may be used by other
1486 * blocks to store page cgroup when onlining pages. But we don't know
1487 * in what order pages are onlined. So we iterate twice to offline
1489 * 1st iterate: offline every non primary memory block.
1490 * 2nd iterate: offline primary (i.e. first added) memory block.
1493 walk_memory_range(start_pfn
, end_pfn
, &ret
,
1494 offline_memory_block_cb
);
1504 lock_memory_hotplug();
1507 * we have offlined all memory blocks like this:
1508 * 1. lock memory hotplug
1509 * 2. offline a memory block
1510 * 3. unlock memory hotplug
1512 * repeat step1-3 to offline the memory block. All memory blocks
1513 * must be offlined before removing memory. But we don't hold the
1514 * lock in the whole operation. So we should check whether all
1515 * memory blocks are offlined.
1518 ret
= walk_memory_range(start_pfn
, end_pfn
, NULL
,
1519 is_memblock_offlined_cb
);
1521 unlock_memory_hotplug();
1525 /* remove memmap entry */
1526 firmware_map_remove(start
, start
+ size
, "System RAM");
1528 arch_remove_memory(start
, size
);
1530 unlock_memory_hotplug();
1535 int offline_pages(unsigned long start_pfn
, unsigned long nr_pages
)
1539 int remove_memory(u64 start
, u64 size
)
1543 #endif /* CONFIG_MEMORY_HOTREMOVE */
1544 EXPORT_SYMBOL_GPL(remove_memory
);