2 * linux/mm/memory_hotplug.c
7 #include <linux/stddef.h>
9 #include <linux/swap.h>
10 #include <linux/interrupt.h>
11 #include <linux/pagemap.h>
12 #include <linux/bootmem.h>
13 #include <linux/compiler.h>
14 #include <linux/export.h>
15 #include <linux/pagevec.h>
16 #include <linux/writeback.h>
17 #include <linux/slab.h>
18 #include <linux/sysctl.h>
19 #include <linux/cpu.h>
20 #include <linux/memory.h>
21 #include <linux/memory_hotplug.h>
22 #include <linux/highmem.h>
23 #include <linux/vmalloc.h>
24 #include <linux/ioport.h>
25 #include <linux/delay.h>
26 #include <linux/migrate.h>
27 #include <linux/page-isolation.h>
28 #include <linux/pfn.h>
29 #include <linux/suspend.h>
30 #include <linux/mm_inline.h>
31 #include <linux/firmware-map.h>
32 #include <linux/stop_machine.h>
34 #include <asm/tlbflush.h>
39 * online_page_callback contains pointer to current page onlining function.
40 * Initially it is generic_online_page(). If it is required it could be
41 * changed by calling set_online_page_callback() for callback registration
42 * and restore_online_page_callback() for generic callback restore.
45 static void generic_online_page(struct page
*page
);
47 static online_page_callback_t online_page_callback
= generic_online_page
;
49 DEFINE_MUTEX(mem_hotplug_mutex
);
51 void lock_memory_hotplug(void)
53 mutex_lock(&mem_hotplug_mutex
);
56 void unlock_memory_hotplug(void)
58 mutex_unlock(&mem_hotplug_mutex
);
62 /* add this memory to iomem resource */
63 static struct resource
*register_memory_resource(u64 start
, u64 size
)
66 res
= kzalloc(sizeof(struct resource
), GFP_KERNEL
);
69 res
->name
= "System RAM";
71 res
->end
= start
+ size
- 1;
72 res
->flags
= IORESOURCE_MEM
| IORESOURCE_BUSY
;
73 if (request_resource(&iomem_resource
, res
) < 0) {
74 pr_debug("System RAM resource %pR cannot be added\n", res
);
81 static void release_memory_resource(struct resource
*res
)
85 release_resource(res
);
90 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
91 void get_page_bootmem(unsigned long info
, struct page
*page
,
94 page
->lru
.next
= (struct list_head
*) type
;
96 set_page_private(page
, info
);
97 atomic_inc(&page
->_count
);
100 void put_page_bootmem(struct page
*page
)
104 type
= (unsigned long) page
->lru
.next
;
105 BUG_ON(type
< MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE
||
106 type
> MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE
);
108 if (atomic_dec_return(&page
->_count
) == 1) {
109 ClearPagePrivate(page
);
110 set_page_private(page
, 0);
111 INIT_LIST_HEAD(&page
->lru
);
112 free_reserved_page(page
);
116 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
117 #ifndef CONFIG_SPARSEMEM_VMEMMAP
118 static void register_page_bootmem_info_section(unsigned long start_pfn
)
120 unsigned long *usemap
, mapsize
, section_nr
, i
;
121 struct mem_section
*ms
;
122 struct page
*page
, *memmap
;
124 section_nr
= pfn_to_section_nr(start_pfn
);
125 ms
= __nr_to_section(section_nr
);
127 /* Get section's memmap address */
128 memmap
= sparse_decode_mem_map(ms
->section_mem_map
, section_nr
);
131 * Get page for the memmap's phys address
132 * XXX: need more consideration for sparse_vmemmap...
134 page
= virt_to_page(memmap
);
135 mapsize
= sizeof(struct page
) * PAGES_PER_SECTION
;
136 mapsize
= PAGE_ALIGN(mapsize
) >> PAGE_SHIFT
;
138 /* remember memmap's page */
139 for (i
= 0; i
< mapsize
; i
++, page
++)
140 get_page_bootmem(section_nr
, page
, SECTION_INFO
);
142 usemap
= __nr_to_section(section_nr
)->pageblock_flags
;
143 page
= virt_to_page(usemap
);
145 mapsize
= PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT
;
147 for (i
= 0; i
< mapsize
; i
++, page
++)
148 get_page_bootmem(section_nr
, page
, MIX_SECTION_INFO
);
151 #else /* CONFIG_SPARSEMEM_VMEMMAP */
152 static void register_page_bootmem_info_section(unsigned long start_pfn
)
154 unsigned long *usemap
, mapsize
, section_nr
, i
;
155 struct mem_section
*ms
;
156 struct page
*page
, *memmap
;
158 if (!pfn_valid(start_pfn
))
161 section_nr
= pfn_to_section_nr(start_pfn
);
162 ms
= __nr_to_section(section_nr
);
164 memmap
= sparse_decode_mem_map(ms
->section_mem_map
, section_nr
);
166 register_page_bootmem_memmap(section_nr
, memmap
, PAGES_PER_SECTION
);
168 usemap
= __nr_to_section(section_nr
)->pageblock_flags
;
169 page
= virt_to_page(usemap
);
171 mapsize
= PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT
;
173 for (i
= 0; i
< mapsize
; i
++, page
++)
174 get_page_bootmem(section_nr
, page
, MIX_SECTION_INFO
);
176 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
178 void register_page_bootmem_info_node(struct pglist_data
*pgdat
)
180 unsigned long i
, pfn
, end_pfn
, nr_pages
;
181 int node
= pgdat
->node_id
;
185 nr_pages
= PAGE_ALIGN(sizeof(struct pglist_data
)) >> PAGE_SHIFT
;
186 page
= virt_to_page(pgdat
);
188 for (i
= 0; i
< nr_pages
; i
++, page
++)
189 get_page_bootmem(node
, page
, NODE_INFO
);
191 zone
= &pgdat
->node_zones
[0];
192 for (; zone
< pgdat
->node_zones
+ MAX_NR_ZONES
- 1; zone
++) {
193 if (zone
->wait_table
) {
194 nr_pages
= zone
->wait_table_hash_nr_entries
195 * sizeof(wait_queue_head_t
);
196 nr_pages
= PAGE_ALIGN(nr_pages
) >> PAGE_SHIFT
;
197 page
= virt_to_page(zone
->wait_table
);
199 for (i
= 0; i
< nr_pages
; i
++, page
++)
200 get_page_bootmem(node
, page
, NODE_INFO
);
204 pfn
= pgdat
->node_start_pfn
;
205 end_pfn
= pgdat_end_pfn(pgdat
);
207 /* register section info */
208 for (; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
210 * Some platforms can assign the same pfn to multiple nodes - on
211 * node0 as well as nodeN. To avoid registering a pfn against
212 * multiple nodes we check that this pfn does not already
213 * reside in some other nodes.
215 if (pfn_valid(pfn
) && (pfn_to_nid(pfn
) == node
))
216 register_page_bootmem_info_section(pfn
);
219 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
221 static void grow_zone_span(struct zone
*zone
, unsigned long start_pfn
,
222 unsigned long end_pfn
)
224 unsigned long old_zone_end_pfn
;
226 zone_span_writelock(zone
);
228 old_zone_end_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
229 if (!zone
->spanned_pages
|| start_pfn
< zone
->zone_start_pfn
)
230 zone
->zone_start_pfn
= start_pfn
;
232 zone
->spanned_pages
= max(old_zone_end_pfn
, end_pfn
) -
233 zone
->zone_start_pfn
;
235 zone_span_writeunlock(zone
);
238 static void resize_zone(struct zone
*zone
, unsigned long start_pfn
,
239 unsigned long end_pfn
)
241 zone_span_writelock(zone
);
243 if (end_pfn
- start_pfn
) {
244 zone
->zone_start_pfn
= start_pfn
;
245 zone
->spanned_pages
= end_pfn
- start_pfn
;
248 * make it consist as free_area_init_core(),
249 * if spanned_pages = 0, then keep start_pfn = 0
251 zone
->zone_start_pfn
= 0;
252 zone
->spanned_pages
= 0;
255 zone_span_writeunlock(zone
);
258 static void fix_zone_id(struct zone
*zone
, unsigned long start_pfn
,
259 unsigned long end_pfn
)
261 enum zone_type zid
= zone_idx(zone
);
262 int nid
= zone
->zone_pgdat
->node_id
;
265 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
266 set_page_links(pfn_to_page(pfn
), zid
, nid
, pfn
);
269 /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
270 * alloc_bootmem_node_nopanic() */
271 static int __ref
ensure_zone_is_initialized(struct zone
*zone
,
272 unsigned long start_pfn
, unsigned long num_pages
)
274 if (!zone_is_initialized(zone
))
275 return init_currently_empty_zone(zone
, start_pfn
, num_pages
,
280 static int __meminit
move_pfn_range_left(struct zone
*z1
, struct zone
*z2
,
281 unsigned long start_pfn
, unsigned long end_pfn
)
285 unsigned long z1_start_pfn
;
287 ret
= ensure_zone_is_initialized(z1
, start_pfn
, end_pfn
- start_pfn
);
291 pgdat_resize_lock(z1
->zone_pgdat
, &flags
);
293 /* can't move pfns which are higher than @z2 */
294 if (end_pfn
> zone_end_pfn(z2
))
296 /* the move out part must be at the left most of @z2 */
297 if (start_pfn
> z2
->zone_start_pfn
)
299 /* must included/overlap */
300 if (end_pfn
<= z2
->zone_start_pfn
)
303 /* use start_pfn for z1's start_pfn if z1 is empty */
304 if (z1
->spanned_pages
)
305 z1_start_pfn
= z1
->zone_start_pfn
;
307 z1_start_pfn
= start_pfn
;
309 resize_zone(z1
, z1_start_pfn
, end_pfn
);
310 resize_zone(z2
, end_pfn
, zone_end_pfn(z2
));
312 pgdat_resize_unlock(z1
->zone_pgdat
, &flags
);
314 fix_zone_id(z1
, start_pfn
, end_pfn
);
318 pgdat_resize_unlock(z1
->zone_pgdat
, &flags
);
322 static int __meminit
move_pfn_range_right(struct zone
*z1
, struct zone
*z2
,
323 unsigned long start_pfn
, unsigned long end_pfn
)
327 unsigned long z2_end_pfn
;
329 ret
= ensure_zone_is_initialized(z2
, start_pfn
, end_pfn
- start_pfn
);
333 pgdat_resize_lock(z1
->zone_pgdat
, &flags
);
335 /* can't move pfns which are lower than @z1 */
336 if (z1
->zone_start_pfn
> start_pfn
)
338 /* the move out part mast at the right most of @z1 */
339 if (zone_end_pfn(z1
) > end_pfn
)
341 /* must included/overlap */
342 if (start_pfn
>= zone_end_pfn(z1
))
345 /* use end_pfn for z2's end_pfn if z2 is empty */
346 if (z2
->spanned_pages
)
347 z2_end_pfn
= zone_end_pfn(z2
);
349 z2_end_pfn
= end_pfn
;
351 resize_zone(z1
, z1
->zone_start_pfn
, start_pfn
);
352 resize_zone(z2
, start_pfn
, z2_end_pfn
);
354 pgdat_resize_unlock(z1
->zone_pgdat
, &flags
);
356 fix_zone_id(z2
, start_pfn
, end_pfn
);
360 pgdat_resize_unlock(z1
->zone_pgdat
, &flags
);
364 static void grow_pgdat_span(struct pglist_data
*pgdat
, unsigned long start_pfn
,
365 unsigned long end_pfn
)
367 unsigned long old_pgdat_end_pfn
=
368 pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
370 if (!pgdat
->node_spanned_pages
|| start_pfn
< pgdat
->node_start_pfn
)
371 pgdat
->node_start_pfn
= start_pfn
;
373 pgdat
->node_spanned_pages
= max(old_pgdat_end_pfn
, end_pfn
) -
374 pgdat
->node_start_pfn
;
377 static int __meminit
__add_zone(struct zone
*zone
, unsigned long phys_start_pfn
)
379 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
380 int nr_pages
= PAGES_PER_SECTION
;
381 int nid
= pgdat
->node_id
;
386 zone_type
= zone
- pgdat
->node_zones
;
387 ret
= ensure_zone_is_initialized(zone
, phys_start_pfn
, nr_pages
);
391 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
392 grow_zone_span(zone
, phys_start_pfn
, phys_start_pfn
+ nr_pages
);
393 grow_pgdat_span(zone
->zone_pgdat
, phys_start_pfn
,
394 phys_start_pfn
+ nr_pages
);
395 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
396 memmap_init_zone(nr_pages
, nid
, zone_type
,
397 phys_start_pfn
, MEMMAP_HOTPLUG
);
401 static int __meminit
__add_section(int nid
, struct zone
*zone
,
402 unsigned long phys_start_pfn
)
404 int nr_pages
= PAGES_PER_SECTION
;
407 if (pfn_valid(phys_start_pfn
))
410 ret
= sparse_add_one_section(zone
, phys_start_pfn
, nr_pages
);
415 ret
= __add_zone(zone
, phys_start_pfn
);
420 return register_new_memory(nid
, __pfn_to_section(phys_start_pfn
));
424 * Reasonably generic function for adding memory. It is
425 * expected that archs that support memory hotplug will
426 * call this function after deciding the zone to which to
429 int __ref
__add_pages(int nid
, struct zone
*zone
, unsigned long phys_start_pfn
,
430 unsigned long nr_pages
)
434 int start_sec
, end_sec
;
435 /* during initialize mem_map, align hot-added range to section */
436 start_sec
= pfn_to_section_nr(phys_start_pfn
);
437 end_sec
= pfn_to_section_nr(phys_start_pfn
+ nr_pages
- 1);
439 for (i
= start_sec
; i
<= end_sec
; i
++) {
440 err
= __add_section(nid
, zone
, i
<< PFN_SECTION_SHIFT
);
443 * EEXIST is finally dealt with by ioresource collision
444 * check. see add_memory() => register_memory_resource()
445 * Warning will be printed if there is collision.
447 if (err
&& (err
!= -EEXIST
))
454 EXPORT_SYMBOL_GPL(__add_pages
);
456 #ifdef CONFIG_MEMORY_HOTREMOVE
457 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
458 static int find_smallest_section_pfn(int nid
, struct zone
*zone
,
459 unsigned long start_pfn
,
460 unsigned long end_pfn
)
462 struct mem_section
*ms
;
464 for (; start_pfn
< end_pfn
; start_pfn
+= PAGES_PER_SECTION
) {
465 ms
= __pfn_to_section(start_pfn
);
467 if (unlikely(!valid_section(ms
)))
470 if (unlikely(pfn_to_nid(start_pfn
) != nid
))
473 if (zone
&& zone
!= page_zone(pfn_to_page(start_pfn
)))
482 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
483 static int find_biggest_section_pfn(int nid
, struct zone
*zone
,
484 unsigned long start_pfn
,
485 unsigned long end_pfn
)
487 struct mem_section
*ms
;
490 /* pfn is the end pfn of a memory section. */
492 for (; pfn
>= start_pfn
; pfn
-= PAGES_PER_SECTION
) {
493 ms
= __pfn_to_section(pfn
);
495 if (unlikely(!valid_section(ms
)))
498 if (unlikely(pfn_to_nid(pfn
) != nid
))
501 if (zone
&& zone
!= page_zone(pfn_to_page(pfn
)))
510 static void shrink_zone_span(struct zone
*zone
, unsigned long start_pfn
,
511 unsigned long end_pfn
)
513 unsigned long zone_start_pfn
= zone
->zone_start_pfn
;
514 unsigned long zone_end_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
516 struct mem_section
*ms
;
517 int nid
= zone_to_nid(zone
);
519 zone_span_writelock(zone
);
520 if (zone_start_pfn
== start_pfn
) {
522 * If the section is smallest section in the zone, it need
523 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
524 * In this case, we find second smallest valid mem_section
525 * for shrinking zone.
527 pfn
= find_smallest_section_pfn(nid
, zone
, end_pfn
,
530 zone
->zone_start_pfn
= pfn
;
531 zone
->spanned_pages
= zone_end_pfn
- pfn
;
533 } else if (zone_end_pfn
== end_pfn
) {
535 * If the section is biggest section in the zone, it need
536 * shrink zone->spanned_pages.
537 * In this case, we find second biggest valid mem_section for
540 pfn
= find_biggest_section_pfn(nid
, zone
, zone_start_pfn
,
543 zone
->spanned_pages
= pfn
- zone_start_pfn
+ 1;
547 * The section is not biggest or smallest mem_section in the zone, it
548 * only creates a hole in the zone. So in this case, we need not
549 * change the zone. But perhaps, the zone has only hole data. Thus
550 * it check the zone has only hole or not.
552 pfn
= zone_start_pfn
;
553 for (; pfn
< zone_end_pfn
; pfn
+= PAGES_PER_SECTION
) {
554 ms
= __pfn_to_section(pfn
);
556 if (unlikely(!valid_section(ms
)))
559 if (page_zone(pfn_to_page(pfn
)) != zone
)
562 /* If the section is current section, it continues the loop */
563 if (start_pfn
== pfn
)
566 /* If we find valid section, we have nothing to do */
567 zone_span_writeunlock(zone
);
571 /* The zone has no valid section */
572 zone
->zone_start_pfn
= 0;
573 zone
->spanned_pages
= 0;
574 zone_span_writeunlock(zone
);
577 static void shrink_pgdat_span(struct pglist_data
*pgdat
,
578 unsigned long start_pfn
, unsigned long end_pfn
)
580 unsigned long pgdat_start_pfn
= pgdat
->node_start_pfn
;
581 unsigned long pgdat_end_pfn
=
582 pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
584 struct mem_section
*ms
;
585 int nid
= pgdat
->node_id
;
587 if (pgdat_start_pfn
== start_pfn
) {
589 * If the section is smallest section in the pgdat, it need
590 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
591 * In this case, we find second smallest valid mem_section
592 * for shrinking zone.
594 pfn
= find_smallest_section_pfn(nid
, NULL
, end_pfn
,
597 pgdat
->node_start_pfn
= pfn
;
598 pgdat
->node_spanned_pages
= pgdat_end_pfn
- pfn
;
600 } else if (pgdat_end_pfn
== end_pfn
) {
602 * If the section is biggest section in the pgdat, it need
603 * shrink pgdat->node_spanned_pages.
604 * In this case, we find second biggest valid mem_section for
607 pfn
= find_biggest_section_pfn(nid
, NULL
, pgdat_start_pfn
,
610 pgdat
->node_spanned_pages
= pfn
- pgdat_start_pfn
+ 1;
614 * If the section is not biggest or smallest mem_section in the pgdat,
615 * it only creates a hole in the pgdat. So in this case, we need not
617 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
618 * has only hole or not.
620 pfn
= pgdat_start_pfn
;
621 for (; pfn
< pgdat_end_pfn
; pfn
+= PAGES_PER_SECTION
) {
622 ms
= __pfn_to_section(pfn
);
624 if (unlikely(!valid_section(ms
)))
627 if (pfn_to_nid(pfn
) != nid
)
630 /* If the section is current section, it continues the loop */
631 if (start_pfn
== pfn
)
634 /* If we find valid section, we have nothing to do */
638 /* The pgdat has no valid section */
639 pgdat
->node_start_pfn
= 0;
640 pgdat
->node_spanned_pages
= 0;
643 static void __remove_zone(struct zone
*zone
, unsigned long start_pfn
)
645 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
646 int nr_pages
= PAGES_PER_SECTION
;
650 zone_type
= zone
- pgdat
->node_zones
;
652 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
653 shrink_zone_span(zone
, start_pfn
, start_pfn
+ nr_pages
);
654 shrink_pgdat_span(pgdat
, start_pfn
, start_pfn
+ nr_pages
);
655 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
658 static int __remove_section(struct zone
*zone
, struct mem_section
*ms
)
660 unsigned long start_pfn
;
664 if (!valid_section(ms
))
667 ret
= unregister_memory_section(ms
);
671 scn_nr
= __section_nr(ms
);
672 start_pfn
= section_nr_to_pfn(scn_nr
);
673 __remove_zone(zone
, start_pfn
);
675 sparse_remove_one_section(zone
, ms
);
680 * __remove_pages() - remove sections of pages from a zone
681 * @zone: zone from which pages need to be removed
682 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
683 * @nr_pages: number of pages to remove (must be multiple of section size)
685 * Generic helper function to remove section mappings and sysfs entries
686 * for the section of the memory we are removing. Caller needs to make
687 * sure that pages are marked reserved and zones are adjust properly by
688 * calling offline_pages().
690 int __remove_pages(struct zone
*zone
, unsigned long phys_start_pfn
,
691 unsigned long nr_pages
)
694 int sections_to_remove
;
695 resource_size_t start
, size
;
699 * We can only remove entire sections
701 BUG_ON(phys_start_pfn
& ~PAGE_SECTION_MASK
);
702 BUG_ON(nr_pages
% PAGES_PER_SECTION
);
704 start
= phys_start_pfn
<< PAGE_SHIFT
;
705 size
= nr_pages
* PAGE_SIZE
;
706 ret
= release_mem_region_adjustable(&iomem_resource
, start
, size
);
708 resource_size_t endres
= start
+ size
- 1;
710 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
711 &start
, &endres
, ret
);
714 sections_to_remove
= nr_pages
/ PAGES_PER_SECTION
;
715 for (i
= 0; i
< sections_to_remove
; i
++) {
716 unsigned long pfn
= phys_start_pfn
+ i
*PAGES_PER_SECTION
;
717 ret
= __remove_section(zone
, __pfn_to_section(pfn
));
723 EXPORT_SYMBOL_GPL(__remove_pages
);
724 #endif /* CONFIG_MEMORY_HOTREMOVE */
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
)
763 EXPORT_SYMBOL_GPL(__online_page_set_limits
);
765 void __online_page_increment_counters(struct page
*page
)
767 adjust_managed_page_count(page
, 1);
769 EXPORT_SYMBOL_GPL(__online_page_increment_counters
);
771 void __online_page_free(struct page
*page
)
773 __free_reserved_page(page
);
775 EXPORT_SYMBOL_GPL(__online_page_free
);
777 static void generic_online_page(struct page
*page
)
779 __online_page_set_limits(page
);
780 __online_page_increment_counters(page
);
781 __online_page_free(page
);
784 static int online_pages_range(unsigned long start_pfn
, unsigned long nr_pages
,
788 unsigned long onlined_pages
= *(unsigned long *)arg
;
790 if (PageReserved(pfn_to_page(start_pfn
)))
791 for (i
= 0; i
< nr_pages
; i
++) {
792 page
= pfn_to_page(start_pfn
+ i
);
793 (*online_page_callback
)(page
);
796 *(unsigned long *)arg
= onlined_pages
;
800 #ifdef CONFIG_MOVABLE_NODE
802 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
805 static bool can_online_high_movable(struct zone
*zone
)
809 #else /* CONFIG_MOVABLE_NODE */
810 /* ensure every online node has NORMAL memory */
811 static bool can_online_high_movable(struct zone
*zone
)
813 return node_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
815 #endif /* CONFIG_MOVABLE_NODE */
817 /* check which state of node_states will be changed when online memory */
818 static void node_states_check_changes_online(unsigned long nr_pages
,
819 struct zone
*zone
, struct memory_notify
*arg
)
821 int nid
= zone_to_nid(zone
);
822 enum zone_type zone_last
= ZONE_NORMAL
;
825 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
826 * contains nodes which have zones of 0...ZONE_NORMAL,
827 * set zone_last to ZONE_NORMAL.
829 * If we don't have HIGHMEM nor movable node,
830 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
831 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
833 if (N_MEMORY
== N_NORMAL_MEMORY
)
834 zone_last
= ZONE_MOVABLE
;
837 * if the memory to be online is in a zone of 0...zone_last, and
838 * the zones of 0...zone_last don't have memory before online, we will
839 * need to set the node to node_states[N_NORMAL_MEMORY] after
840 * the memory is online.
842 if (zone_idx(zone
) <= zone_last
&& !node_state(nid
, N_NORMAL_MEMORY
))
843 arg
->status_change_nid_normal
= nid
;
845 arg
->status_change_nid_normal
= -1;
847 #ifdef CONFIG_HIGHMEM
849 * If we have movable node, node_states[N_HIGH_MEMORY]
850 * contains nodes which have zones of 0...ZONE_HIGHMEM,
851 * set zone_last to ZONE_HIGHMEM.
853 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
854 * contains nodes which have zones of 0...ZONE_MOVABLE,
855 * set zone_last to ZONE_MOVABLE.
857 zone_last
= ZONE_HIGHMEM
;
858 if (N_MEMORY
== N_HIGH_MEMORY
)
859 zone_last
= ZONE_MOVABLE
;
861 if (zone_idx(zone
) <= zone_last
&& !node_state(nid
, N_HIGH_MEMORY
))
862 arg
->status_change_nid_high
= nid
;
864 arg
->status_change_nid_high
= -1;
866 arg
->status_change_nid_high
= arg
->status_change_nid_normal
;
870 * if the node don't have memory befor online, we will need to
871 * set the node to node_states[N_MEMORY] after the memory
874 if (!node_state(nid
, N_MEMORY
))
875 arg
->status_change_nid
= nid
;
877 arg
->status_change_nid
= -1;
880 static void node_states_set_node(int node
, struct memory_notify
*arg
)
882 if (arg
->status_change_nid_normal
>= 0)
883 node_set_state(node
, N_NORMAL_MEMORY
);
885 if (arg
->status_change_nid_high
>= 0)
886 node_set_state(node
, N_HIGH_MEMORY
);
888 node_set_state(node
, N_MEMORY
);
892 int __ref
online_pages(unsigned long pfn
, unsigned long nr_pages
, int online_type
)
895 unsigned long onlined_pages
= 0;
897 int need_zonelists_rebuild
= 0;
900 struct memory_notify arg
;
902 lock_memory_hotplug();
904 * This doesn't need a lock to do pfn_to_page().
905 * The section can't be removed here because of the
906 * memory_block->state_mutex.
908 zone
= page_zone(pfn_to_page(pfn
));
910 if ((zone_idx(zone
) > ZONE_NORMAL
|| online_type
== ONLINE_MOVABLE
) &&
911 !can_online_high_movable(zone
)) {
912 unlock_memory_hotplug();
916 if (online_type
== ONLINE_KERNEL
&& zone_idx(zone
) == ZONE_MOVABLE
) {
917 if (move_pfn_range_left(zone
- 1, zone
, pfn
, pfn
+ nr_pages
)) {
918 unlock_memory_hotplug();
922 if (online_type
== ONLINE_MOVABLE
&& zone_idx(zone
) == ZONE_MOVABLE
- 1) {
923 if (move_pfn_range_right(zone
, zone
+ 1, pfn
, pfn
+ nr_pages
)) {
924 unlock_memory_hotplug();
929 /* Previous code may changed the zone of the pfn range */
930 zone
= page_zone(pfn_to_page(pfn
));
933 arg
.nr_pages
= nr_pages
;
934 node_states_check_changes_online(nr_pages
, zone
, &arg
);
936 nid
= page_to_nid(pfn_to_page(pfn
));
938 ret
= memory_notify(MEM_GOING_ONLINE
, &arg
);
939 ret
= notifier_to_errno(ret
);
941 memory_notify(MEM_CANCEL_ONLINE
, &arg
);
942 unlock_memory_hotplug();
946 * If this zone is not populated, then it is not in zonelist.
947 * This means the page allocator ignores this zone.
948 * So, zonelist must be updated after online.
950 mutex_lock(&zonelists_mutex
);
951 if (!populated_zone(zone
)) {
952 need_zonelists_rebuild
= 1;
953 build_all_zonelists(NULL
, zone
);
956 ret
= walk_system_ram_range(pfn
, nr_pages
, &onlined_pages
,
959 if (need_zonelists_rebuild
)
960 zone_pcp_reset(zone
);
961 mutex_unlock(&zonelists_mutex
);
962 printk(KERN_DEBUG
"online_pages [mem %#010llx-%#010llx] failed\n",
963 (unsigned long long) pfn
<< PAGE_SHIFT
,
964 (((unsigned long long) pfn
+ nr_pages
)
966 memory_notify(MEM_CANCEL_ONLINE
, &arg
);
967 unlock_memory_hotplug();
971 zone
->present_pages
+= onlined_pages
;
973 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
974 zone
->zone_pgdat
->node_present_pages
+= onlined_pages
;
975 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
978 node_states_set_node(zone_to_nid(zone
), &arg
);
979 if (need_zonelists_rebuild
)
980 build_all_zonelists(NULL
, NULL
);
982 zone_pcp_update(zone
);
985 mutex_unlock(&zonelists_mutex
);
987 init_per_zone_wmark_min();
990 kswapd_run(zone_to_nid(zone
));
992 vm_total_pages
= nr_free_pagecache_pages();
994 writeback_set_ratelimit();
997 memory_notify(MEM_ONLINE
, &arg
);
998 unlock_memory_hotplug();
1002 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1004 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1005 static pg_data_t __ref
*hotadd_new_pgdat(int nid
, u64 start
)
1007 struct pglist_data
*pgdat
;
1008 unsigned long zones_size
[MAX_NR_ZONES
] = {0};
1009 unsigned long zholes_size
[MAX_NR_ZONES
] = {0};
1010 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
1012 pgdat
= NODE_DATA(nid
);
1014 pgdat
= arch_alloc_nodedata(nid
);
1018 arch_refresh_nodedata(nid
, pgdat
);
1021 /* we can use NODE_DATA(nid) from here */
1023 /* init node's zones as empty zones, we don't have any present pages.*/
1024 free_area_init_node(nid
, zones_size
, start_pfn
, zholes_size
);
1027 * The node we allocated has no zone fallback lists. For avoiding
1028 * to access not-initialized zonelist, build here.
1030 mutex_lock(&zonelists_mutex
);
1031 build_all_zonelists(pgdat
, NULL
);
1032 mutex_unlock(&zonelists_mutex
);
1037 static void rollback_node_hotadd(int nid
, pg_data_t
*pgdat
)
1039 arch_refresh_nodedata(nid
, NULL
);
1040 arch_free_nodedata(pgdat
);
1046 * called by cpu_up() to online a node without onlined memory.
1048 int mem_online_node(int nid
)
1053 lock_memory_hotplug();
1054 pgdat
= hotadd_new_pgdat(nid
, 0);
1059 node_set_online(nid
);
1060 ret
= register_one_node(nid
);
1064 unlock_memory_hotplug();
1068 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1069 int __ref
add_memory(int nid
, u64 start
, u64 size
)
1071 pg_data_t
*pgdat
= NULL
;
1074 struct resource
*res
;
1077 lock_memory_hotplug();
1079 res
= register_memory_resource(start
, size
);
1084 { /* Stupid hack to suppress address-never-null warning */
1085 void *p
= NODE_DATA(nid
);
1088 new_node
= !node_online(nid
);
1090 pgdat
= hotadd_new_pgdat(nid
, start
);
1096 /* call arch's memory hotadd */
1097 ret
= arch_add_memory(nid
, start
, size
);
1102 /* we online node here. we can't roll back from here. */
1103 node_set_online(nid
);
1106 ret
= register_one_node(nid
);
1108 * If sysfs file of new node can't create, cpu on the node
1109 * can't be hot-added. There is no rollback way now.
1110 * So, check by BUG_ON() to catch it reluctantly..
1115 /* create new memmap entry */
1116 firmware_map_add_hotplug(start
, start
+ size
, "System RAM");
1121 /* rollback pgdat allocation and others */
1123 rollback_node_hotadd(nid
, pgdat
);
1124 release_memory_resource(res
);
1127 unlock_memory_hotplug();
1130 EXPORT_SYMBOL_GPL(add_memory
);
1132 #ifdef CONFIG_MEMORY_HOTREMOVE
1134 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1135 * set and the size of the free page is given by page_order(). Using this,
1136 * the function determines if the pageblock contains only free pages.
1137 * Due to buddy contraints, a free page at least the size of a pageblock will
1138 * be located at the start of the pageblock
1140 static inline int pageblock_free(struct page
*page
)
1142 return PageBuddy(page
) && page_order(page
) >= pageblock_order
;
1145 /* Return the start of the next active pageblock after a given page */
1146 static struct page
*next_active_pageblock(struct page
*page
)
1148 /* Ensure the starting page is pageblock-aligned */
1149 BUG_ON(page_to_pfn(page
) & (pageblock_nr_pages
- 1));
1151 /* If the entire pageblock is free, move to the end of free page */
1152 if (pageblock_free(page
)) {
1154 /* be careful. we don't have locks, page_order can be changed.*/
1155 order
= page_order(page
);
1156 if ((order
< MAX_ORDER
) && (order
>= pageblock_order
))
1157 return page
+ (1 << order
);
1160 return page
+ pageblock_nr_pages
;
1163 /* Checks if this range of memory is likely to be hot-removable. */
1164 int is_mem_section_removable(unsigned long start_pfn
, unsigned long nr_pages
)
1166 struct page
*page
= pfn_to_page(start_pfn
);
1167 struct page
*end_page
= page
+ nr_pages
;
1169 /* Check the starting page of each pageblock within the range */
1170 for (; page
< end_page
; page
= next_active_pageblock(page
)) {
1171 if (!is_pageblock_removable_nolock(page
))
1176 /* All pageblocks in the memory block are likely to be hot-removable */
1181 * Confirm all pages in a range [start, end) is belongs to the same zone.
1183 static int test_pages_in_a_zone(unsigned long start_pfn
, unsigned long end_pfn
)
1186 struct zone
*zone
= NULL
;
1189 for (pfn
= start_pfn
;
1191 pfn
+= MAX_ORDER_NR_PAGES
) {
1193 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1194 while ((i
< MAX_ORDER_NR_PAGES
) && !pfn_valid_within(pfn
+ i
))
1196 if (i
== MAX_ORDER_NR_PAGES
)
1198 page
= pfn_to_page(pfn
+ i
);
1199 if (zone
&& page_zone(page
) != zone
)
1201 zone
= page_zone(page
);
1207 * Scanning pfn is much easier than scanning lru list.
1208 * Scan pfn from start to end and Find LRU page.
1210 static unsigned long scan_lru_pages(unsigned long start
, unsigned long end
)
1214 for (pfn
= start
; pfn
< end
; pfn
++) {
1215 if (pfn_valid(pfn
)) {
1216 page
= pfn_to_page(pfn
);
1224 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1226 do_migrate_range(unsigned long start_pfn
, unsigned long end_pfn
)
1230 int move_pages
= NR_OFFLINE_AT_ONCE_PAGES
;
1231 int not_managed
= 0;
1235 for (pfn
= start_pfn
; pfn
< end_pfn
&& move_pages
> 0; pfn
++) {
1236 if (!pfn_valid(pfn
))
1238 page
= pfn_to_page(pfn
);
1239 if (!get_page_unless_zero(page
))
1242 * We can skip free pages. And we can only deal with pages on
1245 ret
= isolate_lru_page(page
);
1246 if (!ret
) { /* Success */
1248 list_add_tail(&page
->lru
, &source
);
1250 inc_zone_page_state(page
, NR_ISOLATED_ANON
+
1251 page_is_file_cache(page
));
1254 #ifdef CONFIG_DEBUG_VM
1255 printk(KERN_ALERT
"removing pfn %lx from LRU failed\n",
1260 /* Because we don't have big zone->lock. we should
1261 check this again here. */
1262 if (page_count(page
)) {
1269 if (!list_empty(&source
)) {
1271 putback_lru_pages(&source
);
1276 * alloc_migrate_target should be improooooved!!
1277 * migrate_pages returns # of failed pages.
1279 ret
= migrate_pages(&source
, alloc_migrate_target
, 0,
1280 MIGRATE_SYNC
, MR_MEMORY_HOTPLUG
);
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
;
1467 unsigned long flags
;
1469 struct memory_notify arg
;
1471 BUG_ON(start_pfn
>= end_pfn
);
1472 /* at least, alignment against pageblock is necessary */
1473 if (!IS_ALIGNED(start_pfn
, pageblock_nr_pages
))
1475 if (!IS_ALIGNED(end_pfn
, pageblock_nr_pages
))
1477 /* This makes hotplug much easier...and readable.
1478 we assume this for now. .*/
1479 if (!test_pages_in_a_zone(start_pfn
, end_pfn
))
1482 lock_memory_hotplug();
1484 zone
= page_zone(pfn_to_page(start_pfn
));
1485 node
= zone_to_nid(zone
);
1486 nr_pages
= end_pfn
- start_pfn
;
1489 if (zone_idx(zone
) <= ZONE_NORMAL
&& !can_offline_normal(zone
, nr_pages
))
1492 /* set above range as isolated */
1493 ret
= start_isolate_page_range(start_pfn
, end_pfn
,
1494 MIGRATE_MOVABLE
, true);
1498 arg
.start_pfn
= start_pfn
;
1499 arg
.nr_pages
= nr_pages
;
1500 node_states_check_changes_offline(nr_pages
, zone
, &arg
);
1502 ret
= memory_notify(MEM_GOING_OFFLINE
, &arg
);
1503 ret
= notifier_to_errno(ret
);
1505 goto failed_removal
;
1508 expire
= jiffies
+ timeout
;
1512 /* start memory hot removal */
1514 if (time_after(jiffies
, expire
))
1515 goto failed_removal
;
1517 if (signal_pending(current
))
1518 goto failed_removal
;
1521 lru_add_drain_all();
1526 pfn
= scan_lru_pages(start_pfn
, end_pfn
);
1527 if (pfn
) { /* We have page on LRU */
1528 ret
= do_migrate_range(pfn
, end_pfn
);
1534 if (--retry_max
== 0)
1535 goto failed_removal
;
1541 /* drain all zone's lru pagevec, this is asynchronous... */
1542 lru_add_drain_all();
1544 /* drain pcp pages, this is synchronous. */
1547 offlined_pages
= check_pages_isolated(start_pfn
, end_pfn
);
1548 if (offlined_pages
< 0) {
1550 goto failed_removal
;
1552 printk(KERN_INFO
"Offlined Pages %ld\n", offlined_pages
);
1553 /* Ok, all of our target is isolated.
1554 We cannot do rollback at this point. */
1555 offline_isolated_pages(start_pfn
, end_pfn
);
1556 /* reset pagetype flags and makes migrate type to be MOVABLE */
1557 undo_isolate_page_range(start_pfn
, end_pfn
, MIGRATE_MOVABLE
);
1558 /* removal success */
1559 adjust_managed_page_count(pfn_to_page(start_pfn
), -offlined_pages
);
1560 zone
->present_pages
-= offlined_pages
;
1562 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
1563 zone
->zone_pgdat
->node_present_pages
-= offlined_pages
;
1564 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
1566 init_per_zone_wmark_min();
1568 if (!populated_zone(zone
)) {
1569 zone_pcp_reset(zone
);
1570 mutex_lock(&zonelists_mutex
);
1571 build_all_zonelists(NULL
, NULL
);
1572 mutex_unlock(&zonelists_mutex
);
1574 zone_pcp_update(zone
);
1576 node_states_clear_node(node
, &arg
);
1577 if (arg
.status_change_nid
>= 0)
1580 vm_total_pages
= nr_free_pagecache_pages();
1581 writeback_set_ratelimit();
1583 memory_notify(MEM_OFFLINE
, &arg
);
1584 unlock_memory_hotplug();
1588 printk(KERN_INFO
"memory offlining [mem %#010llx-%#010llx] failed\n",
1589 (unsigned long long) start_pfn
<< PAGE_SHIFT
,
1590 ((unsigned long long) end_pfn
<< PAGE_SHIFT
) - 1);
1591 memory_notify(MEM_CANCEL_OFFLINE
, &arg
);
1592 /* pushback to free area */
1593 undo_isolate_page_range(start_pfn
, end_pfn
, MIGRATE_MOVABLE
);
1596 unlock_memory_hotplug();
1600 int offline_pages(unsigned long start_pfn
, unsigned long nr_pages
)
1602 return __offline_pages(start_pfn
, start_pfn
+ nr_pages
, 120 * HZ
);
1604 #endif /* CONFIG_MEMORY_HOTREMOVE */
1607 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1608 * @start_pfn: start pfn of the memory range
1609 * @end_pfn: end pfn of the memory range
1610 * @arg: argument passed to func
1611 * @func: callback for each memory section walked
1613 * This function walks through all present mem sections in range
1614 * [start_pfn, end_pfn) and call func on each mem section.
1616 * Returns the return value of func.
1618 int walk_memory_range(unsigned long start_pfn
, unsigned long end_pfn
,
1619 void *arg
, int (*func
)(struct memory_block
*, void *))
1621 struct memory_block
*mem
= NULL
;
1622 struct mem_section
*section
;
1623 unsigned long pfn
, section_nr
;
1626 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
1627 section_nr
= pfn_to_section_nr(pfn
);
1628 if (!present_section_nr(section_nr
))
1631 section
= __nr_to_section(section_nr
);
1632 /* same memblock? */
1634 if ((section_nr
>= mem
->start_section_nr
) &&
1635 (section_nr
<= mem
->end_section_nr
))
1638 mem
= find_memory_block_hinted(section
, mem
);
1642 ret
= func(mem
, arg
);
1644 kobject_put(&mem
->dev
.kobj
);
1650 kobject_put(&mem
->dev
.kobj
);
1655 #ifdef CONFIG_MEMORY_HOTREMOVE
1656 static int is_memblock_offlined_cb(struct memory_block
*mem
, void *arg
)
1658 int ret
= !is_memblock_offlined(mem
);
1660 if (unlikely(ret
)) {
1661 phys_addr_t beginpa
, endpa
;
1663 beginpa
= PFN_PHYS(section_nr_to_pfn(mem
->start_section_nr
));
1664 endpa
= PFN_PHYS(section_nr_to_pfn(mem
->end_section_nr
+ 1))-1;
1665 pr_warn("removing memory fails, because memory "
1666 "[%pa-%pa] is onlined\n",
1673 static int check_cpu_on_node(void *data
)
1675 struct pglist_data
*pgdat
= data
;
1678 for_each_present_cpu(cpu
) {
1679 if (cpu_to_node(cpu
) == pgdat
->node_id
)
1681 * the cpu on this node isn't removed, and we can't
1682 * offline this node.
1690 static void unmap_cpu_on_node(void *data
)
1692 #ifdef CONFIG_ACPI_NUMA
1693 struct pglist_data
*pgdat
= data
;
1696 for_each_possible_cpu(cpu
)
1697 if (cpu_to_node(cpu
) == pgdat
->node_id
)
1698 numa_clear_node(cpu
);
1702 static int check_and_unmap_cpu_on_node(void *data
)
1704 int ret
= check_cpu_on_node(data
);
1710 * the node will be offlined when we come here, so we can clear
1711 * the cpu_to_node() now.
1714 unmap_cpu_on_node(data
);
1718 /* offline the node if all memory sections of this node are removed */
1719 void try_offline_node(int nid
)
1721 pg_data_t
*pgdat
= NODE_DATA(nid
);
1722 unsigned long start_pfn
= pgdat
->node_start_pfn
;
1723 unsigned long end_pfn
= start_pfn
+ pgdat
->node_spanned_pages
;
1725 struct page
*pgdat_page
= virt_to_page(pgdat
);
1728 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
1729 unsigned long section_nr
= pfn_to_section_nr(pfn
);
1731 if (!present_section_nr(section_nr
))
1734 if (pfn_to_nid(pfn
) != nid
)
1738 * some memory sections of this node are not removed, and we
1739 * can't offline node now.
1744 if (stop_machine(check_and_unmap_cpu_on_node
, pgdat
, NULL
))
1748 * all memory/cpu of this node are removed, we can offline this
1751 node_set_offline(nid
);
1752 unregister_one_node(nid
);
1754 if (!PageSlab(pgdat_page
) && !PageCompound(pgdat_page
))
1755 /* node data is allocated from boot memory */
1758 /* free waittable in each zone */
1759 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
1760 struct zone
*zone
= pgdat
->node_zones
+ i
;
1763 * wait_table may be allocated from boot memory,
1764 * here only free if it's allocated by vmalloc.
1766 if (is_vmalloc_addr(zone
->wait_table
))
1767 vfree(zone
->wait_table
);
1771 * Since there is no way to guarentee the address of pgdat/zone is not
1772 * on stack of any kernel threads or used by other kernel objects
1773 * without reference counting or other symchronizing method, do not
1774 * reset node_data and free pgdat here. Just reset it to 0 and reuse
1775 * the memory when the node is online again.
1777 memset(pgdat
, 0, sizeof(*pgdat
));
1779 EXPORT_SYMBOL(try_offline_node
);
1781 void __ref
remove_memory(int nid
, u64 start
, u64 size
)
1785 lock_memory_hotplug();
1788 * All memory blocks must be offlined before removing memory. Check
1789 * whether all memory blocks in question are offline and trigger a BUG()
1790 * if this is not the case.
1792 ret
= walk_memory_range(PFN_DOWN(start
), PFN_UP(start
+ size
- 1), NULL
,
1793 is_memblock_offlined_cb
);
1795 unlock_memory_hotplug();
1799 /* remove memmap entry */
1800 firmware_map_remove(start
, start
+ size
, "System RAM");
1802 arch_remove_memory(start
, size
);
1804 try_offline_node(nid
);
1806 unlock_memory_hotplug();
1808 EXPORT_SYMBOL_GPL(remove_memory
);
1809 #endif /* CONFIG_MEMORY_HOTREMOVE */