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 #ifndef CONFIG_SPARSEMEM_VMEMMAP
95 static 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
)
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
);
118 __free_pages_bootmem(page
, 0);
123 static void register_page_bootmem_info_section(unsigned long start_pfn
)
125 unsigned long *usemap
, mapsize
, section_nr
, i
;
126 struct mem_section
*ms
;
127 struct page
*page
, *memmap
;
129 section_nr
= pfn_to_section_nr(start_pfn
);
130 ms
= __nr_to_section(section_nr
);
132 /* Get section's memmap address */
133 memmap
= sparse_decode_mem_map(ms
->section_mem_map
, section_nr
);
136 * Get page for the memmap's phys address
137 * XXX: need more consideration for sparse_vmemmap...
139 page
= virt_to_page(memmap
);
140 mapsize
= sizeof(struct page
) * PAGES_PER_SECTION
;
141 mapsize
= PAGE_ALIGN(mapsize
) >> PAGE_SHIFT
;
143 /* remember memmap's page */
144 for (i
= 0; i
< mapsize
; i
++, page
++)
145 get_page_bootmem(section_nr
, page
, SECTION_INFO
);
147 usemap
= __nr_to_section(section_nr
)->pageblock_flags
;
148 page
= virt_to_page(usemap
);
150 mapsize
= PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT
;
152 for (i
= 0; i
< mapsize
; i
++, page
++)
153 get_page_bootmem(section_nr
, page
, MIX_SECTION_INFO
);
157 void register_page_bootmem_info_node(struct pglist_data
*pgdat
)
159 unsigned long i
, pfn
, end_pfn
, nr_pages
;
160 int node
= pgdat
->node_id
;
164 nr_pages
= PAGE_ALIGN(sizeof(struct pglist_data
)) >> PAGE_SHIFT
;
165 page
= virt_to_page(pgdat
);
167 for (i
= 0; i
< nr_pages
; i
++, page
++)
168 get_page_bootmem(node
, page
, NODE_INFO
);
170 zone
= &pgdat
->node_zones
[0];
171 for (; zone
< pgdat
->node_zones
+ MAX_NR_ZONES
- 1; zone
++) {
172 if (zone
->wait_table
) {
173 nr_pages
= zone
->wait_table_hash_nr_entries
174 * sizeof(wait_queue_head_t
);
175 nr_pages
= PAGE_ALIGN(nr_pages
) >> PAGE_SHIFT
;
176 page
= virt_to_page(zone
->wait_table
);
178 for (i
= 0; i
< nr_pages
; i
++, page
++)
179 get_page_bootmem(node
, page
, NODE_INFO
);
183 pfn
= pgdat
->node_start_pfn
;
184 end_pfn
= pfn
+ pgdat
->node_spanned_pages
;
186 /* register_section info */
187 for (; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
189 * Some platforms can assign the same pfn to multiple nodes - on
190 * node0 as well as nodeN. To avoid registering a pfn against
191 * multiple nodes we check that this pfn does not already
192 * reside in some other node.
194 if (pfn_valid(pfn
) && (pfn_to_nid(pfn
) == node
))
195 register_page_bootmem_info_section(pfn
);
198 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
200 static void grow_zone_span(struct zone
*zone
, unsigned long start_pfn
,
201 unsigned long end_pfn
)
203 unsigned long old_zone_end_pfn
;
205 zone_span_writelock(zone
);
207 old_zone_end_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
208 if (!zone
->spanned_pages
|| start_pfn
< zone
->zone_start_pfn
)
209 zone
->zone_start_pfn
= start_pfn
;
211 zone
->spanned_pages
= max(old_zone_end_pfn
, end_pfn
) -
212 zone
->zone_start_pfn
;
214 zone_span_writeunlock(zone
);
217 static void resize_zone(struct zone
*zone
, unsigned long start_pfn
,
218 unsigned long end_pfn
)
220 zone_span_writelock(zone
);
222 zone
->zone_start_pfn
= start_pfn
;
223 zone
->spanned_pages
= end_pfn
- start_pfn
;
225 zone_span_writeunlock(zone
);
228 static void fix_zone_id(struct zone
*zone
, unsigned long start_pfn
,
229 unsigned long end_pfn
)
231 enum zone_type zid
= zone_idx(zone
);
232 int nid
= zone
->zone_pgdat
->node_id
;
235 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
236 set_page_links(pfn_to_page(pfn
), zid
, nid
, pfn
);
239 static int move_pfn_range_left(struct zone
*z1
, struct zone
*z2
,
240 unsigned long start_pfn
, unsigned long end_pfn
)
244 pgdat_resize_lock(z1
->zone_pgdat
, &flags
);
246 /* can't move pfns which are higher than @z2 */
247 if (end_pfn
> z2
->zone_start_pfn
+ z2
->spanned_pages
)
249 /* the move out part mast at the left most of @z2 */
250 if (start_pfn
> z2
->zone_start_pfn
)
252 /* must included/overlap */
253 if (end_pfn
<= z2
->zone_start_pfn
)
256 resize_zone(z1
, z1
->zone_start_pfn
, end_pfn
);
257 resize_zone(z2
, end_pfn
, z2
->zone_start_pfn
+ z2
->spanned_pages
);
259 pgdat_resize_unlock(z1
->zone_pgdat
, &flags
);
261 fix_zone_id(z1
, start_pfn
, end_pfn
);
265 pgdat_resize_unlock(z1
->zone_pgdat
, &flags
);
269 static int move_pfn_range_right(struct zone
*z1
, struct zone
*z2
,
270 unsigned long start_pfn
, unsigned long end_pfn
)
274 pgdat_resize_lock(z1
->zone_pgdat
, &flags
);
276 /* can't move pfns which are lower than @z1 */
277 if (z1
->zone_start_pfn
> start_pfn
)
279 /* the move out part mast at the right most of @z1 */
280 if (z1
->zone_start_pfn
+ z1
->spanned_pages
> end_pfn
)
282 /* must included/overlap */
283 if (start_pfn
>= z1
->zone_start_pfn
+ z1
->spanned_pages
)
286 resize_zone(z1
, z1
->zone_start_pfn
, start_pfn
);
287 resize_zone(z2
, start_pfn
, z2
->zone_start_pfn
+ z2
->spanned_pages
);
289 pgdat_resize_unlock(z1
->zone_pgdat
, &flags
);
291 fix_zone_id(z2
, start_pfn
, end_pfn
);
295 pgdat_resize_unlock(z1
->zone_pgdat
, &flags
);
299 static void grow_pgdat_span(struct pglist_data
*pgdat
, unsigned long start_pfn
,
300 unsigned long end_pfn
)
302 unsigned long old_pgdat_end_pfn
=
303 pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
305 if (!pgdat
->node_spanned_pages
|| start_pfn
< pgdat
->node_start_pfn
)
306 pgdat
->node_start_pfn
= start_pfn
;
308 pgdat
->node_spanned_pages
= max(old_pgdat_end_pfn
, end_pfn
) -
309 pgdat
->node_start_pfn
;
312 static int __meminit
__add_zone(struct zone
*zone
, unsigned long phys_start_pfn
)
314 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
315 int nr_pages
= PAGES_PER_SECTION
;
316 int nid
= pgdat
->node_id
;
320 zone_type
= zone
- pgdat
->node_zones
;
321 if (!zone
->wait_table
) {
324 ret
= init_currently_empty_zone(zone
, phys_start_pfn
,
325 nr_pages
, MEMMAP_HOTPLUG
);
329 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
330 grow_zone_span(zone
, phys_start_pfn
, phys_start_pfn
+ nr_pages
);
331 grow_pgdat_span(zone
->zone_pgdat
, phys_start_pfn
,
332 phys_start_pfn
+ nr_pages
);
333 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
334 memmap_init_zone(nr_pages
, nid
, zone_type
,
335 phys_start_pfn
, MEMMAP_HOTPLUG
);
339 static int __meminit
__add_section(int nid
, struct zone
*zone
,
340 unsigned long phys_start_pfn
)
342 int nr_pages
= PAGES_PER_SECTION
;
345 if (pfn_valid(phys_start_pfn
))
348 ret
= sparse_add_one_section(zone
, phys_start_pfn
, nr_pages
);
353 ret
= __add_zone(zone
, phys_start_pfn
);
358 return register_new_memory(nid
, __pfn_to_section(phys_start_pfn
));
361 #ifdef CONFIG_SPARSEMEM_VMEMMAP
362 static int __remove_section(struct zone
*zone
, struct mem_section
*ms
)
365 * XXX: Freeing memmap with vmemmap is not implement yet.
366 * This should be removed later.
371 static int __remove_section(struct zone
*zone
, struct mem_section
*ms
)
374 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
377 if (!valid_section(ms
))
380 ret
= unregister_memory_section(ms
);
384 pgdat_resize_lock(pgdat
, &flags
);
385 sparse_remove_one_section(zone
, ms
);
386 pgdat_resize_unlock(pgdat
, &flags
);
392 * Reasonably generic function for adding memory. It is
393 * expected that archs that support memory hotplug will
394 * call this function after deciding the zone to which to
397 int __ref
__add_pages(int nid
, struct zone
*zone
, unsigned long phys_start_pfn
,
398 unsigned long nr_pages
)
402 int start_sec
, end_sec
;
403 /* during initialize mem_map, align hot-added range to section */
404 start_sec
= pfn_to_section_nr(phys_start_pfn
);
405 end_sec
= pfn_to_section_nr(phys_start_pfn
+ nr_pages
- 1);
407 for (i
= start_sec
; i
<= end_sec
; i
++) {
408 err
= __add_section(nid
, zone
, i
<< PFN_SECTION_SHIFT
);
411 * EEXIST is finally dealt with by ioresource collision
412 * check. see add_memory() => register_memory_resource()
413 * Warning will be printed if there is collision.
415 if (err
&& (err
!= -EEXIST
))
422 EXPORT_SYMBOL_GPL(__add_pages
);
425 * __remove_pages() - remove sections of pages from a zone
426 * @zone: zone from which pages need to be removed
427 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
428 * @nr_pages: number of pages to remove (must be multiple of section size)
430 * Generic helper function to remove section mappings and sysfs entries
431 * for the section of the memory we are removing. Caller needs to make
432 * sure that pages are marked reserved and zones are adjust properly by
433 * calling offline_pages().
435 int __remove_pages(struct zone
*zone
, unsigned long phys_start_pfn
,
436 unsigned long nr_pages
)
438 unsigned long i
, ret
= 0;
439 int sections_to_remove
;
442 * We can only remove entire sections
444 BUG_ON(phys_start_pfn
& ~PAGE_SECTION_MASK
);
445 BUG_ON(nr_pages
% PAGES_PER_SECTION
);
447 release_mem_region(phys_start_pfn
<< PAGE_SHIFT
, nr_pages
* PAGE_SIZE
);
449 sections_to_remove
= nr_pages
/ PAGES_PER_SECTION
;
450 for (i
= 0; i
< sections_to_remove
; i
++) {
451 unsigned long pfn
= phys_start_pfn
+ i
*PAGES_PER_SECTION
;
452 ret
= __remove_section(zone
, __pfn_to_section(pfn
));
458 EXPORT_SYMBOL_GPL(__remove_pages
);
460 int set_online_page_callback(online_page_callback_t callback
)
464 lock_memory_hotplug();
466 if (online_page_callback
== generic_online_page
) {
467 online_page_callback
= callback
;
471 unlock_memory_hotplug();
475 EXPORT_SYMBOL_GPL(set_online_page_callback
);
477 int restore_online_page_callback(online_page_callback_t callback
)
481 lock_memory_hotplug();
483 if (online_page_callback
== callback
) {
484 online_page_callback
= generic_online_page
;
488 unlock_memory_hotplug();
492 EXPORT_SYMBOL_GPL(restore_online_page_callback
);
494 void __online_page_set_limits(struct page
*page
)
496 unsigned long pfn
= page_to_pfn(page
);
498 if (pfn
>= num_physpages
)
499 num_physpages
= pfn
+ 1;
501 EXPORT_SYMBOL_GPL(__online_page_set_limits
);
503 void __online_page_increment_counters(struct page
*page
)
507 #ifdef CONFIG_HIGHMEM
508 if (PageHighMem(page
))
512 EXPORT_SYMBOL_GPL(__online_page_increment_counters
);
514 void __online_page_free(struct page
*page
)
516 ClearPageReserved(page
);
517 init_page_count(page
);
520 EXPORT_SYMBOL_GPL(__online_page_free
);
522 static void generic_online_page(struct page
*page
)
524 __online_page_set_limits(page
);
525 __online_page_increment_counters(page
);
526 __online_page_free(page
);
529 static int online_pages_range(unsigned long start_pfn
, unsigned long nr_pages
,
533 unsigned long onlined_pages
= *(unsigned long *)arg
;
535 if (PageReserved(pfn_to_page(start_pfn
)))
536 for (i
= 0; i
< nr_pages
; i
++) {
537 page
= pfn_to_page(start_pfn
+ i
);
538 (*online_page_callback
)(page
);
541 *(unsigned long *)arg
= onlined_pages
;
545 /* check which state of node_states will be changed when online memory */
546 static void node_states_check_changes_online(unsigned long nr_pages
,
547 struct zone
*zone
, struct memory_notify
*arg
)
549 int nid
= zone_to_nid(zone
);
550 enum zone_type zone_last
= ZONE_NORMAL
;
553 * If we have HIGHMEM, node_states[N_NORMAL_MEMORY] contains nodes
554 * which have 0...ZONE_NORMAL, set zone_last to ZONE_NORMAL.
556 * If we don't have HIGHMEM, node_states[N_NORMAL_MEMORY] contains nodes
557 * which have 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
559 if (N_HIGH_MEMORY
== N_NORMAL_MEMORY
)
560 zone_last
= ZONE_MOVABLE
;
563 * if the memory to be online is in a zone of 0...zone_last, and
564 * the zones of 0...zone_last don't have memory before online, we will
565 * need to set the node to node_states[N_NORMAL_MEMORY] after
566 * the memory is online.
568 if (zone_idx(zone
) <= zone_last
&& !node_state(nid
, N_NORMAL_MEMORY
))
569 arg
->status_change_nid_normal
= nid
;
571 arg
->status_change_nid_normal
= -1;
574 * if the node don't have memory befor online, we will need to
575 * set the node to node_states[N_HIGH_MEMORY] after the memory
578 if (!node_state(nid
, N_HIGH_MEMORY
))
579 arg
->status_change_nid
= nid
;
581 arg
->status_change_nid
= -1;
584 static void node_states_set_node(int node
, struct memory_notify
*arg
)
586 if (arg
->status_change_nid_normal
>= 0)
587 node_set_state(node
, N_NORMAL_MEMORY
);
589 node_set_state(node
, N_HIGH_MEMORY
);
593 int __ref
online_pages(unsigned long pfn
, unsigned long nr_pages
, int online_type
)
595 unsigned long onlined_pages
= 0;
597 int need_zonelists_rebuild
= 0;
600 struct memory_notify arg
;
602 lock_memory_hotplug();
604 * This doesn't need a lock to do pfn_to_page().
605 * The section can't be removed here because of the
606 * memory_block->state_mutex.
608 zone
= page_zone(pfn_to_page(pfn
));
610 if (online_type
== ONLINE_KERNEL
&& zone_idx(zone
) == ZONE_MOVABLE
) {
611 if (move_pfn_range_left(zone
- 1, zone
, pfn
, pfn
+ nr_pages
)) {
612 unlock_memory_hotplug();
616 if (online_type
== ONLINE_MOVABLE
&& zone_idx(zone
) == ZONE_MOVABLE
- 1) {
617 if (move_pfn_range_right(zone
, zone
+ 1, pfn
, pfn
+ nr_pages
)) {
618 unlock_memory_hotplug();
623 /* Previous code may changed the zone of the pfn range */
624 zone
= page_zone(pfn_to_page(pfn
));
627 arg
.nr_pages
= nr_pages
;
628 node_states_check_changes_online(nr_pages
, zone
, &arg
);
630 nid
= page_to_nid(pfn_to_page(pfn
));
632 ret
= memory_notify(MEM_GOING_ONLINE
, &arg
);
633 ret
= notifier_to_errno(ret
);
635 memory_notify(MEM_CANCEL_ONLINE
, &arg
);
636 unlock_memory_hotplug();
640 * If this zone is not populated, then it is not in zonelist.
641 * This means the page allocator ignores this zone.
642 * So, zonelist must be updated after online.
644 mutex_lock(&zonelists_mutex
);
645 if (!populated_zone(zone
)) {
646 need_zonelists_rebuild
= 1;
647 build_all_zonelists(NULL
, zone
);
650 ret
= walk_system_ram_range(pfn
, nr_pages
, &onlined_pages
,
653 if (need_zonelists_rebuild
)
654 zone_pcp_reset(zone
);
655 mutex_unlock(&zonelists_mutex
);
656 printk(KERN_DEBUG
"online_pages [mem %#010llx-%#010llx] failed\n",
657 (unsigned long long) pfn
<< PAGE_SHIFT
,
658 (((unsigned long long) pfn
+ nr_pages
)
660 memory_notify(MEM_CANCEL_ONLINE
, &arg
);
661 unlock_memory_hotplug();
665 zone
->present_pages
+= onlined_pages
;
666 zone
->zone_pgdat
->node_present_pages
+= onlined_pages
;
668 node_states_set_node(zone_to_nid(zone
), &arg
);
669 if (need_zonelists_rebuild
)
670 build_all_zonelists(NULL
, NULL
);
672 zone_pcp_update(zone
);
675 mutex_unlock(&zonelists_mutex
);
677 init_per_zone_wmark_min();
680 kswapd_run(zone_to_nid(zone
));
682 vm_total_pages
= nr_free_pagecache_pages();
684 writeback_set_ratelimit();
687 memory_notify(MEM_ONLINE
, &arg
);
688 unlock_memory_hotplug();
692 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
694 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
695 static pg_data_t __ref
*hotadd_new_pgdat(int nid
, u64 start
)
697 struct pglist_data
*pgdat
;
698 unsigned long zones_size
[MAX_NR_ZONES
] = {0};
699 unsigned long zholes_size
[MAX_NR_ZONES
] = {0};
700 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
702 pgdat
= arch_alloc_nodedata(nid
);
706 arch_refresh_nodedata(nid
, pgdat
);
708 /* we can use NODE_DATA(nid) from here */
710 /* init node's zones as empty zones, we don't have any present pages.*/
711 free_area_init_node(nid
, zones_size
, start_pfn
, zholes_size
);
714 * The node we allocated has no zone fallback lists. For avoiding
715 * to access not-initialized zonelist, build here.
717 mutex_lock(&zonelists_mutex
);
718 build_all_zonelists(pgdat
, NULL
);
719 mutex_unlock(&zonelists_mutex
);
724 static void rollback_node_hotadd(int nid
, pg_data_t
*pgdat
)
726 arch_refresh_nodedata(nid
, NULL
);
727 arch_free_nodedata(pgdat
);
733 * called by cpu_up() to online a node without onlined memory.
735 int mem_online_node(int nid
)
740 lock_memory_hotplug();
741 pgdat
= hotadd_new_pgdat(nid
, 0);
746 node_set_online(nid
);
747 ret
= register_one_node(nid
);
751 unlock_memory_hotplug();
755 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
756 int __ref
add_memory(int nid
, u64 start
, u64 size
)
758 pg_data_t
*pgdat
= NULL
;
760 struct resource
*res
;
763 lock_memory_hotplug();
765 res
= register_memory_resource(start
, size
);
770 if (!node_online(nid
)) {
771 pgdat
= hotadd_new_pgdat(nid
, start
);
778 /* call arch's memory hotadd */
779 ret
= arch_add_memory(nid
, start
, size
);
784 /* we online node here. we can't roll back from here. */
785 node_set_online(nid
);
788 ret
= register_one_node(nid
);
790 * If sysfs file of new node can't create, cpu on the node
791 * can't be hot-added. There is no rollback way now.
792 * So, check by BUG_ON() to catch it reluctantly..
797 /* create new memmap entry */
798 firmware_map_add_hotplug(start
, start
+ size
, "System RAM");
803 /* rollback pgdat allocation and others */
805 rollback_node_hotadd(nid
, pgdat
);
807 release_memory_resource(res
);
810 unlock_memory_hotplug();
813 EXPORT_SYMBOL_GPL(add_memory
);
815 #ifdef CONFIG_MEMORY_HOTREMOVE
817 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
818 * set and the size of the free page is given by page_order(). Using this,
819 * the function determines if the pageblock contains only free pages.
820 * Due to buddy contraints, a free page at least the size of a pageblock will
821 * be located at the start of the pageblock
823 static inline int pageblock_free(struct page
*page
)
825 return PageBuddy(page
) && page_order(page
) >= pageblock_order
;
828 /* Return the start of the next active pageblock after a given page */
829 static struct page
*next_active_pageblock(struct page
*page
)
831 /* Ensure the starting page is pageblock-aligned */
832 BUG_ON(page_to_pfn(page
) & (pageblock_nr_pages
- 1));
834 /* If the entire pageblock is free, move to the end of free page */
835 if (pageblock_free(page
)) {
837 /* be careful. we don't have locks, page_order can be changed.*/
838 order
= page_order(page
);
839 if ((order
< MAX_ORDER
) && (order
>= pageblock_order
))
840 return page
+ (1 << order
);
843 return page
+ pageblock_nr_pages
;
846 /* Checks if this range of memory is likely to be hot-removable. */
847 int is_mem_section_removable(unsigned long start_pfn
, unsigned long nr_pages
)
849 struct page
*page
= pfn_to_page(start_pfn
);
850 struct page
*end_page
= page
+ nr_pages
;
852 /* Check the starting page of each pageblock within the range */
853 for (; page
< end_page
; page
= next_active_pageblock(page
)) {
854 if (!is_pageblock_removable_nolock(page
))
859 /* All pageblocks in the memory block are likely to be hot-removable */
864 * Confirm all pages in a range [start, end) is belongs to the same zone.
866 static int test_pages_in_a_zone(unsigned long start_pfn
, unsigned long end_pfn
)
869 struct zone
*zone
= NULL
;
872 for (pfn
= start_pfn
;
874 pfn
+= MAX_ORDER_NR_PAGES
) {
876 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
877 while ((i
< MAX_ORDER_NR_PAGES
) && !pfn_valid_within(pfn
+ i
))
879 if (i
== MAX_ORDER_NR_PAGES
)
881 page
= pfn_to_page(pfn
+ i
);
882 if (zone
&& page_zone(page
) != zone
)
884 zone
= page_zone(page
);
890 * Scanning pfn is much easier than scanning lru list.
891 * Scan pfn from start to end and Find LRU page.
893 static unsigned long scan_lru_pages(unsigned long start
, unsigned long end
)
897 for (pfn
= start
; pfn
< end
; pfn
++) {
898 if (pfn_valid(pfn
)) {
899 page
= pfn_to_page(pfn
);
907 #define NR_OFFLINE_AT_ONCE_PAGES (256)
909 do_migrate_range(unsigned long start_pfn
, unsigned long end_pfn
)
913 int move_pages
= NR_OFFLINE_AT_ONCE_PAGES
;
918 for (pfn
= start_pfn
; pfn
< end_pfn
&& move_pages
> 0; pfn
++) {
921 page
= pfn_to_page(pfn
);
922 if (!get_page_unless_zero(page
))
925 * We can skip free pages. And we can only deal with pages on
928 ret
= isolate_lru_page(page
);
929 if (!ret
) { /* Success */
931 list_add_tail(&page
->lru
, &source
);
933 inc_zone_page_state(page
, NR_ISOLATED_ANON
+
934 page_is_file_cache(page
));
937 #ifdef CONFIG_DEBUG_VM
938 printk(KERN_ALERT
"removing pfn %lx from LRU failed\n",
943 /* Because we don't have big zone->lock. we should
944 check this again here. */
945 if (page_count(page
)) {
952 if (!list_empty(&source
)) {
954 putback_lru_pages(&source
);
959 * alloc_migrate_target should be improooooved!!
960 * migrate_pages returns # of failed pages.
962 ret
= migrate_pages(&source
, alloc_migrate_target
, 0,
965 putback_lru_pages(&source
);
972 * remove from free_area[] and mark all as Reserved.
975 offline_isolated_pages_cb(unsigned long start
, unsigned long nr_pages
,
978 __offline_isolated_pages(start
, start
+ nr_pages
);
983 offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
985 walk_system_ram_range(start_pfn
, end_pfn
- start_pfn
, NULL
,
986 offline_isolated_pages_cb
);
990 * Check all pages in range, recoreded as memory resource, are isolated.
993 check_pages_isolated_cb(unsigned long start_pfn
, unsigned long nr_pages
,
997 long offlined
= *(long *)data
;
998 ret
= test_pages_isolated(start_pfn
, start_pfn
+ nr_pages
, true);
1001 *(long *)data
+= offlined
;
1006 check_pages_isolated(unsigned long start_pfn
, unsigned long end_pfn
)
1011 ret
= walk_system_ram_range(start_pfn
, end_pfn
- start_pfn
, &offlined
,
1012 check_pages_isolated_cb
);
1014 offlined
= (long)ret
;
1018 /* check which state of node_states will be changed when offline memory */
1019 static void node_states_check_changes_offline(unsigned long nr_pages
,
1020 struct zone
*zone
, struct memory_notify
*arg
)
1022 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1023 unsigned long present_pages
= 0;
1024 enum zone_type zt
, zone_last
= ZONE_NORMAL
;
1027 * If we have HIGHMEM, node_states[N_NORMAL_MEMORY] contains nodes
1028 * which have 0...ZONE_NORMAL, set zone_last to ZONE_NORMAL.
1030 * If we don't have HIGHMEM, node_states[N_NORMAL_MEMORY] contains nodes
1031 * which have 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1033 if (N_HIGH_MEMORY
== N_NORMAL_MEMORY
)
1034 zone_last
= ZONE_MOVABLE
;
1037 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1038 * If the memory to be offline is in a zone of 0...zone_last,
1039 * and it is the last present memory, 0...zone_last will
1040 * become empty after offline , thus we can determind we will
1041 * need to clear the node from node_states[N_NORMAL_MEMORY].
1043 for (zt
= 0; zt
<= zone_last
; zt
++)
1044 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1045 if (zone_idx(zone
) <= zone_last
&& nr_pages
>= present_pages
)
1046 arg
->status_change_nid_normal
= zone_to_nid(zone
);
1048 arg
->status_change_nid_normal
= -1;
1051 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1053 zone_last
= ZONE_MOVABLE
;
1056 * check whether node_states[N_HIGH_MEMORY] will be changed
1057 * If we try to offline the last present @nr_pages from the node,
1058 * we can determind we will need to clear the node from
1059 * node_states[N_HIGH_MEMORY].
1061 for (; zt
<= zone_last
; zt
++)
1062 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1063 if (nr_pages
>= present_pages
)
1064 arg
->status_change_nid
= zone_to_nid(zone
);
1066 arg
->status_change_nid
= -1;
1069 static void node_states_clear_node(int node
, struct memory_notify
*arg
)
1071 if (arg
->status_change_nid_normal
>= 0)
1072 node_clear_state(node
, N_NORMAL_MEMORY
);
1074 if ((N_HIGH_MEMORY
!= N_NORMAL_MEMORY
) &&
1075 (arg
->status_change_nid
>= 0))
1076 node_clear_state(node
, N_HIGH_MEMORY
);
1079 static int __ref
__offline_pages(unsigned long start_pfn
,
1080 unsigned long end_pfn
, unsigned long timeout
)
1082 unsigned long pfn
, nr_pages
, expire
;
1083 long offlined_pages
;
1084 int ret
, drain
, retry_max
, node
;
1086 struct memory_notify arg
;
1088 BUG_ON(start_pfn
>= end_pfn
);
1089 /* at least, alignment against pageblock is necessary */
1090 if (!IS_ALIGNED(start_pfn
, pageblock_nr_pages
))
1092 if (!IS_ALIGNED(end_pfn
, pageblock_nr_pages
))
1094 /* This makes hotplug much easier...and readable.
1095 we assume this for now. .*/
1096 if (!test_pages_in_a_zone(start_pfn
, end_pfn
))
1099 lock_memory_hotplug();
1101 zone
= page_zone(pfn_to_page(start_pfn
));
1102 node
= zone_to_nid(zone
);
1103 nr_pages
= end_pfn
- start_pfn
;
1105 /* set above range as isolated */
1106 ret
= start_isolate_page_range(start_pfn
, end_pfn
,
1107 MIGRATE_MOVABLE
, true);
1111 arg
.start_pfn
= start_pfn
;
1112 arg
.nr_pages
= nr_pages
;
1113 node_states_check_changes_offline(nr_pages
, zone
, &arg
);
1115 ret
= memory_notify(MEM_GOING_OFFLINE
, &arg
);
1116 ret
= notifier_to_errno(ret
);
1118 goto failed_removal
;
1121 expire
= jiffies
+ timeout
;
1125 /* start memory hot removal */
1127 if (time_after(jiffies
, expire
))
1128 goto failed_removal
;
1130 if (signal_pending(current
))
1131 goto failed_removal
;
1134 lru_add_drain_all();
1139 pfn
= scan_lru_pages(start_pfn
, end_pfn
);
1140 if (pfn
) { /* We have page on LRU */
1141 ret
= do_migrate_range(pfn
, end_pfn
);
1147 if (--retry_max
== 0)
1148 goto failed_removal
;
1154 /* drain all zone's lru pagevec, this is asyncronous... */
1155 lru_add_drain_all();
1157 /* drain pcp pages , this is synchrouns. */
1160 offlined_pages
= check_pages_isolated(start_pfn
, end_pfn
);
1161 if (offlined_pages
< 0) {
1163 goto failed_removal
;
1165 printk(KERN_INFO
"Offlined Pages %ld\n", offlined_pages
);
1166 /* Ok, all of our target is islaoted.
1167 We cannot do rollback at this point. */
1168 offline_isolated_pages(start_pfn
, end_pfn
);
1169 /* reset pagetype flags and makes migrate type to be MOVABLE */
1170 undo_isolate_page_range(start_pfn
, end_pfn
, MIGRATE_MOVABLE
);
1171 /* removal success */
1172 zone
->present_pages
-= offlined_pages
;
1173 zone
->zone_pgdat
->node_present_pages
-= offlined_pages
;
1174 totalram_pages
-= offlined_pages
;
1176 init_per_zone_wmark_min();
1178 if (!populated_zone(zone
)) {
1179 zone_pcp_reset(zone
);
1180 mutex_lock(&zonelists_mutex
);
1181 build_all_zonelists(NULL
, NULL
);
1182 mutex_unlock(&zonelists_mutex
);
1184 zone_pcp_update(zone
);
1186 node_states_clear_node(node
, &arg
);
1187 if (arg
.status_change_nid
>= 0)
1190 vm_total_pages
= nr_free_pagecache_pages();
1191 writeback_set_ratelimit();
1193 memory_notify(MEM_OFFLINE
, &arg
);
1194 unlock_memory_hotplug();
1198 printk(KERN_INFO
"memory offlining [mem %#010llx-%#010llx] failed\n",
1199 (unsigned long long) start_pfn
<< PAGE_SHIFT
,
1200 ((unsigned long long) end_pfn
<< PAGE_SHIFT
) - 1);
1201 memory_notify(MEM_CANCEL_OFFLINE
, &arg
);
1202 /* pushback to free area */
1203 undo_isolate_page_range(start_pfn
, end_pfn
, MIGRATE_MOVABLE
);
1206 unlock_memory_hotplug();
1210 int offline_pages(unsigned long start_pfn
, unsigned long nr_pages
)
1212 return __offline_pages(start_pfn
, start_pfn
+ nr_pages
, 120 * HZ
);
1215 int remove_memory(u64 start
, u64 size
)
1217 struct memory_block
*mem
= NULL
;
1218 struct mem_section
*section
;
1219 unsigned long start_pfn
, end_pfn
;
1220 unsigned long pfn
, section_nr
;
1223 start_pfn
= PFN_DOWN(start
);
1224 end_pfn
= start_pfn
+ PFN_DOWN(size
);
1226 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
1227 section_nr
= pfn_to_section_nr(pfn
);
1228 if (!present_section_nr(section_nr
))
1231 section
= __nr_to_section(section_nr
);
1232 /* same memblock? */
1234 if ((section_nr
>= mem
->start_section_nr
) &&
1235 (section_nr
<= mem
->end_section_nr
))
1238 mem
= find_memory_block_hinted(section
, mem
);
1242 ret
= offline_memory_block(mem
);
1244 kobject_put(&mem
->dev
.kobj
);
1250 kobject_put(&mem
->dev
.kobj
);
1255 int offline_pages(unsigned long start_pfn
, unsigned long nr_pages
)
1259 int remove_memory(u64 start
, u64 size
)
1263 #endif /* CONFIG_MEMORY_HOTREMOVE */
1264 EXPORT_SYMBOL_GPL(remove_memory
);