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/compiler.h>
13 #include <linux/export.h>
14 #include <linux/pagevec.h>
15 #include <linux/writeback.h>
16 #include <linux/slab.h>
17 #include <linux/sysctl.h>
18 #include <linux/cpu.h>
19 #include <linux/memory.h>
20 #include <linux/memory_hotplug.h>
21 #include <linux/highmem.h>
22 #include <linux/vmalloc.h>
23 #include <linux/ioport.h>
24 #include <linux/delay.h>
25 #include <linux/migrate.h>
26 #include <linux/page-isolation.h>
27 #include <linux/pfn.h>
28 #include <linux/suspend.h>
29 #include <linux/mm_inline.h>
30 #include <linux/firmware-map.h>
31 #include <linux/stop_machine.h>
32 #include <linux/hugetlb.h>
33 #include <linux/memblock.h>
34 #include <linux/bootmem.h>
36 #include <asm/tlbflush.h>
41 * online_page_callback contains pointer to current page onlining function.
42 * Initially it is generic_online_page(). If it is required it could be
43 * changed by calling set_online_page_callback() for callback registration
44 * and restore_online_page_callback() for generic callback restore.
47 static void generic_online_page(struct page
*page
);
49 static online_page_callback_t online_page_callback
= generic_online_page
;
50 static DEFINE_MUTEX(online_page_callback_lock
);
52 /* The same as the cpu_hotplug lock, but for memory hotplug. */
54 struct task_struct
*active_writer
;
55 struct mutex lock
; /* Synchronizes accesses to refcount, */
57 * Also blocks the new readers during
58 * an ongoing mem hotplug operation.
62 #ifdef CONFIG_DEBUG_LOCK_ALLOC
63 struct lockdep_map dep_map
;
66 .active_writer
= NULL
,
67 .lock
= __MUTEX_INITIALIZER(mem_hotplug
.lock
),
69 #ifdef CONFIG_DEBUG_LOCK_ALLOC
70 .dep_map
= {.name
= "mem_hotplug.lock" },
74 /* Lockdep annotations for get/put_online_mems() and mem_hotplug_begin/end() */
75 #define memhp_lock_acquire_read() lock_map_acquire_read(&mem_hotplug.dep_map)
76 #define memhp_lock_acquire() lock_map_acquire(&mem_hotplug.dep_map)
77 #define memhp_lock_release() lock_map_release(&mem_hotplug.dep_map)
79 void get_online_mems(void)
82 if (mem_hotplug
.active_writer
== current
)
84 memhp_lock_acquire_read();
85 mutex_lock(&mem_hotplug
.lock
);
86 mem_hotplug
.refcount
++;
87 mutex_unlock(&mem_hotplug
.lock
);
91 void put_online_mems(void)
93 if (mem_hotplug
.active_writer
== current
)
95 mutex_lock(&mem_hotplug
.lock
);
97 if (WARN_ON(!mem_hotplug
.refcount
))
98 mem_hotplug
.refcount
++; /* try to fix things up */
100 if (!--mem_hotplug
.refcount
&& unlikely(mem_hotplug
.active_writer
))
101 wake_up_process(mem_hotplug
.active_writer
);
102 mutex_unlock(&mem_hotplug
.lock
);
103 memhp_lock_release();
107 void mem_hotplug_begin(void)
109 mem_hotplug
.active_writer
= current
;
111 memhp_lock_acquire();
113 mutex_lock(&mem_hotplug
.lock
);
114 if (likely(!mem_hotplug
.refcount
))
116 __set_current_state(TASK_UNINTERRUPTIBLE
);
117 mutex_unlock(&mem_hotplug
.lock
);
122 void mem_hotplug_done(void)
124 mem_hotplug
.active_writer
= NULL
;
125 mutex_unlock(&mem_hotplug
.lock
);
126 memhp_lock_release();
129 /* add this memory to iomem resource */
130 static struct resource
*register_memory_resource(u64 start
, u64 size
)
132 struct resource
*res
;
133 res
= kzalloc(sizeof(struct resource
), GFP_KERNEL
);
135 return ERR_PTR(-ENOMEM
);
137 res
->name
= "System RAM";
139 res
->end
= start
+ size
- 1;
140 res
->flags
= IORESOURCE_MEM
| IORESOURCE_BUSY
;
141 if (request_resource(&iomem_resource
, res
) < 0) {
142 pr_debug("System RAM resource %pR cannot be added\n", res
);
144 return ERR_PTR(-EEXIST
);
149 static void release_memory_resource(struct resource
*res
)
153 release_resource(res
);
158 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
159 void get_page_bootmem(unsigned long info
, struct page
*page
,
162 page
->lru
.next
= (struct list_head
*) type
;
163 SetPagePrivate(page
);
164 set_page_private(page
, info
);
165 atomic_inc(&page
->_count
);
168 void put_page_bootmem(struct page
*page
)
172 type
= (unsigned long) page
->lru
.next
;
173 BUG_ON(type
< MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE
||
174 type
> MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE
);
176 if (atomic_dec_return(&page
->_count
) == 1) {
177 ClearPagePrivate(page
);
178 set_page_private(page
, 0);
179 INIT_LIST_HEAD(&page
->lru
);
180 free_reserved_page(page
);
184 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
185 #ifndef CONFIG_SPARSEMEM_VMEMMAP
186 static void register_page_bootmem_info_section(unsigned long start_pfn
)
188 unsigned long *usemap
, mapsize
, section_nr
, i
;
189 struct mem_section
*ms
;
190 struct page
*page
, *memmap
;
192 section_nr
= pfn_to_section_nr(start_pfn
);
193 ms
= __nr_to_section(section_nr
);
195 /* Get section's memmap address */
196 memmap
= sparse_decode_mem_map(ms
->section_mem_map
, section_nr
);
199 * Get page for the memmap's phys address
200 * XXX: need more consideration for sparse_vmemmap...
202 page
= virt_to_page(memmap
);
203 mapsize
= sizeof(struct page
) * PAGES_PER_SECTION
;
204 mapsize
= PAGE_ALIGN(mapsize
) >> PAGE_SHIFT
;
206 /* remember memmap's page */
207 for (i
= 0; i
< mapsize
; i
++, page
++)
208 get_page_bootmem(section_nr
, page
, SECTION_INFO
);
210 usemap
= __nr_to_section(section_nr
)->pageblock_flags
;
211 page
= virt_to_page(usemap
);
213 mapsize
= PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT
;
215 for (i
= 0; i
< mapsize
; i
++, page
++)
216 get_page_bootmem(section_nr
, page
, MIX_SECTION_INFO
);
219 #else /* CONFIG_SPARSEMEM_VMEMMAP */
220 static void register_page_bootmem_info_section(unsigned long start_pfn
)
222 unsigned long *usemap
, mapsize
, section_nr
, i
;
223 struct mem_section
*ms
;
224 struct page
*page
, *memmap
;
226 if (!pfn_valid(start_pfn
))
229 section_nr
= pfn_to_section_nr(start_pfn
);
230 ms
= __nr_to_section(section_nr
);
232 memmap
= sparse_decode_mem_map(ms
->section_mem_map
, section_nr
);
234 register_page_bootmem_memmap(section_nr
, memmap
, PAGES_PER_SECTION
);
236 usemap
= __nr_to_section(section_nr
)->pageblock_flags
;
237 page
= virt_to_page(usemap
);
239 mapsize
= PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT
;
241 for (i
= 0; i
< mapsize
; i
++, page
++)
242 get_page_bootmem(section_nr
, page
, MIX_SECTION_INFO
);
244 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
246 void register_page_bootmem_info_node(struct pglist_data
*pgdat
)
248 unsigned long i
, pfn
, end_pfn
, nr_pages
;
249 int node
= pgdat
->node_id
;
253 nr_pages
= PAGE_ALIGN(sizeof(struct pglist_data
)) >> PAGE_SHIFT
;
254 page
= virt_to_page(pgdat
);
256 for (i
= 0; i
< nr_pages
; i
++, page
++)
257 get_page_bootmem(node
, page
, NODE_INFO
);
259 zone
= &pgdat
->node_zones
[0];
260 for (; zone
< pgdat
->node_zones
+ MAX_NR_ZONES
- 1; zone
++) {
261 if (zone_is_initialized(zone
)) {
262 nr_pages
= zone
->wait_table_hash_nr_entries
263 * sizeof(wait_queue_head_t
);
264 nr_pages
= PAGE_ALIGN(nr_pages
) >> PAGE_SHIFT
;
265 page
= virt_to_page(zone
->wait_table
);
267 for (i
= 0; i
< nr_pages
; i
++, page
++)
268 get_page_bootmem(node
, page
, NODE_INFO
);
272 pfn
= pgdat
->node_start_pfn
;
273 end_pfn
= pgdat_end_pfn(pgdat
);
275 /* register section info */
276 for (; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
278 * Some platforms can assign the same pfn to multiple nodes - on
279 * node0 as well as nodeN. To avoid registering a pfn against
280 * multiple nodes we check that this pfn does not already
281 * reside in some other nodes.
283 if (pfn_valid(pfn
) && (pfn_to_nid(pfn
) == node
))
284 register_page_bootmem_info_section(pfn
);
287 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
289 static void __meminit
grow_zone_span(struct zone
*zone
, unsigned long start_pfn
,
290 unsigned long end_pfn
)
292 unsigned long old_zone_end_pfn
;
294 zone_span_writelock(zone
);
296 old_zone_end_pfn
= zone_end_pfn(zone
);
297 if (zone_is_empty(zone
) || start_pfn
< zone
->zone_start_pfn
)
298 zone
->zone_start_pfn
= start_pfn
;
300 zone
->spanned_pages
= max(old_zone_end_pfn
, end_pfn
) -
301 zone
->zone_start_pfn
;
303 zone_span_writeunlock(zone
);
306 static void resize_zone(struct zone
*zone
, unsigned long start_pfn
,
307 unsigned long end_pfn
)
309 zone_span_writelock(zone
);
311 if (end_pfn
- start_pfn
) {
312 zone
->zone_start_pfn
= start_pfn
;
313 zone
->spanned_pages
= end_pfn
- start_pfn
;
316 * make it consist as free_area_init_core(),
317 * if spanned_pages = 0, then keep start_pfn = 0
319 zone
->zone_start_pfn
= 0;
320 zone
->spanned_pages
= 0;
323 zone_span_writeunlock(zone
);
326 static void fix_zone_id(struct zone
*zone
, unsigned long start_pfn
,
327 unsigned long end_pfn
)
329 enum zone_type zid
= zone_idx(zone
);
330 int nid
= zone
->zone_pgdat
->node_id
;
333 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
334 set_page_links(pfn_to_page(pfn
), zid
, nid
, pfn
);
337 /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
338 * alloc_bootmem_node_nopanic()/memblock_virt_alloc_node_nopanic() */
339 static int __ref
ensure_zone_is_initialized(struct zone
*zone
,
340 unsigned long start_pfn
, unsigned long num_pages
)
342 if (!zone_is_initialized(zone
))
343 return init_currently_empty_zone(zone
, start_pfn
, num_pages
);
348 static int __meminit
move_pfn_range_left(struct zone
*z1
, struct zone
*z2
,
349 unsigned long start_pfn
, unsigned long end_pfn
)
353 unsigned long z1_start_pfn
;
355 ret
= ensure_zone_is_initialized(z1
, start_pfn
, end_pfn
- start_pfn
);
359 pgdat_resize_lock(z1
->zone_pgdat
, &flags
);
361 /* can't move pfns which are higher than @z2 */
362 if (end_pfn
> zone_end_pfn(z2
))
364 /* the move out part must be at the left most of @z2 */
365 if (start_pfn
> z2
->zone_start_pfn
)
367 /* must included/overlap */
368 if (end_pfn
<= z2
->zone_start_pfn
)
371 /* use start_pfn for z1's start_pfn if z1 is empty */
372 if (!zone_is_empty(z1
))
373 z1_start_pfn
= z1
->zone_start_pfn
;
375 z1_start_pfn
= start_pfn
;
377 resize_zone(z1
, z1_start_pfn
, end_pfn
);
378 resize_zone(z2
, end_pfn
, zone_end_pfn(z2
));
380 pgdat_resize_unlock(z1
->zone_pgdat
, &flags
);
382 fix_zone_id(z1
, start_pfn
, end_pfn
);
386 pgdat_resize_unlock(z1
->zone_pgdat
, &flags
);
390 static int __meminit
move_pfn_range_right(struct zone
*z1
, struct zone
*z2
,
391 unsigned long start_pfn
, unsigned long end_pfn
)
395 unsigned long z2_end_pfn
;
397 ret
= ensure_zone_is_initialized(z2
, start_pfn
, end_pfn
- start_pfn
);
401 pgdat_resize_lock(z1
->zone_pgdat
, &flags
);
403 /* can't move pfns which are lower than @z1 */
404 if (z1
->zone_start_pfn
> start_pfn
)
406 /* the move out part mast at the right most of @z1 */
407 if (zone_end_pfn(z1
) > end_pfn
)
409 /* must included/overlap */
410 if (start_pfn
>= zone_end_pfn(z1
))
413 /* use end_pfn for z2's end_pfn if z2 is empty */
414 if (!zone_is_empty(z2
))
415 z2_end_pfn
= zone_end_pfn(z2
);
417 z2_end_pfn
= end_pfn
;
419 resize_zone(z1
, z1
->zone_start_pfn
, start_pfn
);
420 resize_zone(z2
, start_pfn
, z2_end_pfn
);
422 pgdat_resize_unlock(z1
->zone_pgdat
, &flags
);
424 fix_zone_id(z2
, start_pfn
, end_pfn
);
428 pgdat_resize_unlock(z1
->zone_pgdat
, &flags
);
432 static void __meminit
grow_pgdat_span(struct pglist_data
*pgdat
, unsigned long start_pfn
,
433 unsigned long end_pfn
)
435 unsigned long old_pgdat_end_pfn
= pgdat_end_pfn(pgdat
);
437 if (!pgdat
->node_spanned_pages
|| start_pfn
< pgdat
->node_start_pfn
)
438 pgdat
->node_start_pfn
= start_pfn
;
440 pgdat
->node_spanned_pages
= max(old_pgdat_end_pfn
, end_pfn
) -
441 pgdat
->node_start_pfn
;
444 static int __meminit
__add_zone(struct zone
*zone
, unsigned long phys_start_pfn
)
446 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
447 int nr_pages
= PAGES_PER_SECTION
;
448 int nid
= pgdat
->node_id
;
450 unsigned long flags
, pfn
;
453 zone_type
= zone
- pgdat
->node_zones
;
454 ret
= ensure_zone_is_initialized(zone
, phys_start_pfn
, nr_pages
);
458 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
459 grow_zone_span(zone
, phys_start_pfn
, phys_start_pfn
+ nr_pages
);
460 grow_pgdat_span(zone
->zone_pgdat
, phys_start_pfn
,
461 phys_start_pfn
+ nr_pages
);
462 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
463 memmap_init_zone(nr_pages
, nid
, zone_type
,
464 phys_start_pfn
, MEMMAP_HOTPLUG
);
466 /* online_page_range is called later and expects pages reserved */
467 for (pfn
= phys_start_pfn
; pfn
< phys_start_pfn
+ nr_pages
; pfn
++) {
471 SetPageReserved(pfn_to_page(pfn
));
476 static int __meminit
__add_section(int nid
, struct zone
*zone
,
477 unsigned long phys_start_pfn
)
481 if (pfn_valid(phys_start_pfn
))
484 ret
= sparse_add_one_section(zone
, phys_start_pfn
);
489 ret
= __add_zone(zone
, phys_start_pfn
);
494 return register_new_memory(nid
, __pfn_to_section(phys_start_pfn
));
498 * Reasonably generic function for adding memory. It is
499 * expected that archs that support memory hotplug will
500 * call this function after deciding the zone to which to
503 int __ref
__add_pages(int nid
, struct zone
*zone
, unsigned long phys_start_pfn
,
504 unsigned long nr_pages
)
508 int start_sec
, end_sec
;
509 /* during initialize mem_map, align hot-added range to section */
510 start_sec
= pfn_to_section_nr(phys_start_pfn
);
511 end_sec
= pfn_to_section_nr(phys_start_pfn
+ nr_pages
- 1);
513 for (i
= start_sec
; i
<= end_sec
; i
++) {
514 err
= __add_section(nid
, zone
, section_nr_to_pfn(i
));
517 * EEXIST is finally dealt with by ioresource collision
518 * check. see add_memory() => register_memory_resource()
519 * Warning will be printed if there is collision.
521 if (err
&& (err
!= -EEXIST
))
525 vmemmap_populate_print_last();
529 EXPORT_SYMBOL_GPL(__add_pages
);
531 #ifdef CONFIG_MEMORY_HOTREMOVE
532 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
533 static int find_smallest_section_pfn(int nid
, struct zone
*zone
,
534 unsigned long start_pfn
,
535 unsigned long end_pfn
)
537 struct mem_section
*ms
;
539 for (; start_pfn
< end_pfn
; start_pfn
+= PAGES_PER_SECTION
) {
540 ms
= __pfn_to_section(start_pfn
);
542 if (unlikely(!valid_section(ms
)))
545 if (unlikely(pfn_to_nid(start_pfn
) != nid
))
548 if (zone
&& zone
!= page_zone(pfn_to_page(start_pfn
)))
557 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
558 static int find_biggest_section_pfn(int nid
, struct zone
*zone
,
559 unsigned long start_pfn
,
560 unsigned long end_pfn
)
562 struct mem_section
*ms
;
565 /* pfn is the end pfn of a memory section. */
567 for (; pfn
>= start_pfn
; pfn
-= PAGES_PER_SECTION
) {
568 ms
= __pfn_to_section(pfn
);
570 if (unlikely(!valid_section(ms
)))
573 if (unlikely(pfn_to_nid(pfn
) != nid
))
576 if (zone
&& zone
!= page_zone(pfn_to_page(pfn
)))
585 static void shrink_zone_span(struct zone
*zone
, unsigned long start_pfn
,
586 unsigned long end_pfn
)
588 unsigned long zone_start_pfn
= zone
->zone_start_pfn
;
589 unsigned long z
= zone_end_pfn(zone
); /* zone_end_pfn namespace clash */
590 unsigned long zone_end_pfn
= z
;
592 struct mem_section
*ms
;
593 int nid
= zone_to_nid(zone
);
595 zone_span_writelock(zone
);
596 if (zone_start_pfn
== start_pfn
) {
598 * If the section is smallest section in the zone, it need
599 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
600 * In this case, we find second smallest valid mem_section
601 * for shrinking zone.
603 pfn
= find_smallest_section_pfn(nid
, zone
, end_pfn
,
606 zone
->zone_start_pfn
= pfn
;
607 zone
->spanned_pages
= zone_end_pfn
- pfn
;
609 } else if (zone_end_pfn
== end_pfn
) {
611 * If the section is biggest section in the zone, it need
612 * shrink zone->spanned_pages.
613 * In this case, we find second biggest valid mem_section for
616 pfn
= find_biggest_section_pfn(nid
, zone
, zone_start_pfn
,
619 zone
->spanned_pages
= pfn
- zone_start_pfn
+ 1;
623 * The section is not biggest or smallest mem_section in the zone, it
624 * only creates a hole in the zone. So in this case, we need not
625 * change the zone. But perhaps, the zone has only hole data. Thus
626 * it check the zone has only hole or not.
628 pfn
= zone_start_pfn
;
629 for (; pfn
< zone_end_pfn
; pfn
+= PAGES_PER_SECTION
) {
630 ms
= __pfn_to_section(pfn
);
632 if (unlikely(!valid_section(ms
)))
635 if (page_zone(pfn_to_page(pfn
)) != zone
)
638 /* If the section is current section, it continues the loop */
639 if (start_pfn
== pfn
)
642 /* If we find valid section, we have nothing to do */
643 zone_span_writeunlock(zone
);
647 /* The zone has no valid section */
648 zone
->zone_start_pfn
= 0;
649 zone
->spanned_pages
= 0;
650 zone_span_writeunlock(zone
);
653 static void shrink_pgdat_span(struct pglist_data
*pgdat
,
654 unsigned long start_pfn
, unsigned long end_pfn
)
656 unsigned long pgdat_start_pfn
= pgdat
->node_start_pfn
;
657 unsigned long p
= pgdat_end_pfn(pgdat
); /* pgdat_end_pfn namespace clash */
658 unsigned long pgdat_end_pfn
= p
;
660 struct mem_section
*ms
;
661 int nid
= pgdat
->node_id
;
663 if (pgdat_start_pfn
== start_pfn
) {
665 * If the section is smallest section in the pgdat, it need
666 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
667 * In this case, we find second smallest valid mem_section
668 * for shrinking zone.
670 pfn
= find_smallest_section_pfn(nid
, NULL
, end_pfn
,
673 pgdat
->node_start_pfn
= pfn
;
674 pgdat
->node_spanned_pages
= pgdat_end_pfn
- pfn
;
676 } else if (pgdat_end_pfn
== end_pfn
) {
678 * If the section is biggest section in the pgdat, it need
679 * shrink pgdat->node_spanned_pages.
680 * In this case, we find second biggest valid mem_section for
683 pfn
= find_biggest_section_pfn(nid
, NULL
, pgdat_start_pfn
,
686 pgdat
->node_spanned_pages
= pfn
- pgdat_start_pfn
+ 1;
690 * If the section is not biggest or smallest mem_section in the pgdat,
691 * it only creates a hole in the pgdat. So in this case, we need not
693 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
694 * has only hole or not.
696 pfn
= pgdat_start_pfn
;
697 for (; pfn
< pgdat_end_pfn
; pfn
+= PAGES_PER_SECTION
) {
698 ms
= __pfn_to_section(pfn
);
700 if (unlikely(!valid_section(ms
)))
703 if (pfn_to_nid(pfn
) != nid
)
706 /* If the section is current section, it continues the loop */
707 if (start_pfn
== pfn
)
710 /* If we find valid section, we have nothing to do */
714 /* The pgdat has no valid section */
715 pgdat
->node_start_pfn
= 0;
716 pgdat
->node_spanned_pages
= 0;
719 static void __remove_zone(struct zone
*zone
, unsigned long start_pfn
)
721 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
722 int nr_pages
= PAGES_PER_SECTION
;
726 zone_type
= zone
- pgdat
->node_zones
;
728 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
729 shrink_zone_span(zone
, start_pfn
, start_pfn
+ nr_pages
);
730 shrink_pgdat_span(pgdat
, start_pfn
, start_pfn
+ nr_pages
);
731 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
734 static int __remove_section(struct zone
*zone
, struct mem_section
*ms
)
736 unsigned long start_pfn
;
740 if (!valid_section(ms
))
743 ret
= unregister_memory_section(ms
);
747 scn_nr
= __section_nr(ms
);
748 start_pfn
= section_nr_to_pfn(scn_nr
);
749 __remove_zone(zone
, start_pfn
);
751 sparse_remove_one_section(zone
, ms
);
756 * __remove_pages() - remove sections of pages from a zone
757 * @zone: zone from which pages need to be removed
758 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
759 * @nr_pages: number of pages to remove (must be multiple of section size)
761 * Generic helper function to remove section mappings and sysfs entries
762 * for the section of the memory we are removing. Caller needs to make
763 * sure that pages are marked reserved and zones are adjust properly by
764 * calling offline_pages().
766 int __remove_pages(struct zone
*zone
, unsigned long phys_start_pfn
,
767 unsigned long nr_pages
)
770 int sections_to_remove
;
771 resource_size_t start
, size
;
775 * We can only remove entire sections
777 BUG_ON(phys_start_pfn
& ~PAGE_SECTION_MASK
);
778 BUG_ON(nr_pages
% PAGES_PER_SECTION
);
780 start
= phys_start_pfn
<< PAGE_SHIFT
;
781 size
= nr_pages
* PAGE_SIZE
;
783 /* in the ZONE_DEVICE case device driver owns the memory region */
784 if (!is_dev_zone(zone
))
785 ret
= release_mem_region_adjustable(&iomem_resource
, start
, size
);
787 resource_size_t endres
= start
+ size
- 1;
789 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
790 &start
, &endres
, ret
);
793 sections_to_remove
= nr_pages
/ PAGES_PER_SECTION
;
794 for (i
= 0; i
< sections_to_remove
; i
++) {
795 unsigned long pfn
= phys_start_pfn
+ i
*PAGES_PER_SECTION
;
796 ret
= __remove_section(zone
, __pfn_to_section(pfn
));
802 EXPORT_SYMBOL_GPL(__remove_pages
);
803 #endif /* CONFIG_MEMORY_HOTREMOVE */
805 int set_online_page_callback(online_page_callback_t callback
)
810 mutex_lock(&online_page_callback_lock
);
812 if (online_page_callback
== generic_online_page
) {
813 online_page_callback
= callback
;
817 mutex_unlock(&online_page_callback_lock
);
822 EXPORT_SYMBOL_GPL(set_online_page_callback
);
824 int restore_online_page_callback(online_page_callback_t callback
)
829 mutex_lock(&online_page_callback_lock
);
831 if (online_page_callback
== callback
) {
832 online_page_callback
= generic_online_page
;
836 mutex_unlock(&online_page_callback_lock
);
841 EXPORT_SYMBOL_GPL(restore_online_page_callback
);
843 void __online_page_set_limits(struct page
*page
)
846 EXPORT_SYMBOL_GPL(__online_page_set_limits
);
848 void __online_page_increment_counters(struct page
*page
)
850 adjust_managed_page_count(page
, 1);
852 EXPORT_SYMBOL_GPL(__online_page_increment_counters
);
854 void __online_page_free(struct page
*page
)
856 __free_reserved_page(page
);
858 EXPORT_SYMBOL_GPL(__online_page_free
);
860 static void generic_online_page(struct page
*page
)
862 __online_page_set_limits(page
);
863 __online_page_increment_counters(page
);
864 __online_page_free(page
);
867 static int online_pages_range(unsigned long start_pfn
, unsigned long nr_pages
,
871 unsigned long onlined_pages
= *(unsigned long *)arg
;
873 if (PageReserved(pfn_to_page(start_pfn
)))
874 for (i
= 0; i
< nr_pages
; i
++) {
875 page
= pfn_to_page(start_pfn
+ i
);
876 (*online_page_callback
)(page
);
879 *(unsigned long *)arg
= onlined_pages
;
883 #ifdef CONFIG_MOVABLE_NODE
885 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
888 static bool can_online_high_movable(struct zone
*zone
)
892 #else /* CONFIG_MOVABLE_NODE */
893 /* ensure every online node has NORMAL memory */
894 static bool can_online_high_movable(struct zone
*zone
)
896 return node_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
898 #endif /* CONFIG_MOVABLE_NODE */
900 /* check which state of node_states will be changed when online memory */
901 static void node_states_check_changes_online(unsigned long nr_pages
,
902 struct zone
*zone
, struct memory_notify
*arg
)
904 int nid
= zone_to_nid(zone
);
905 enum zone_type zone_last
= ZONE_NORMAL
;
908 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
909 * contains nodes which have zones of 0...ZONE_NORMAL,
910 * set zone_last to ZONE_NORMAL.
912 * If we don't have HIGHMEM nor movable node,
913 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
914 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
916 if (N_MEMORY
== N_NORMAL_MEMORY
)
917 zone_last
= ZONE_MOVABLE
;
920 * if the memory to be online is in a zone of 0...zone_last, and
921 * the zones of 0...zone_last don't have memory before online, we will
922 * need to set the node to node_states[N_NORMAL_MEMORY] after
923 * the memory is online.
925 if (zone_idx(zone
) <= zone_last
&& !node_state(nid
, N_NORMAL_MEMORY
))
926 arg
->status_change_nid_normal
= nid
;
928 arg
->status_change_nid_normal
= -1;
930 #ifdef CONFIG_HIGHMEM
932 * If we have movable node, node_states[N_HIGH_MEMORY]
933 * contains nodes which have zones of 0...ZONE_HIGHMEM,
934 * set zone_last to ZONE_HIGHMEM.
936 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
937 * contains nodes which have zones of 0...ZONE_MOVABLE,
938 * set zone_last to ZONE_MOVABLE.
940 zone_last
= ZONE_HIGHMEM
;
941 if (N_MEMORY
== N_HIGH_MEMORY
)
942 zone_last
= ZONE_MOVABLE
;
944 if (zone_idx(zone
) <= zone_last
&& !node_state(nid
, N_HIGH_MEMORY
))
945 arg
->status_change_nid_high
= nid
;
947 arg
->status_change_nid_high
= -1;
949 arg
->status_change_nid_high
= arg
->status_change_nid_normal
;
953 * if the node don't have memory befor online, we will need to
954 * set the node to node_states[N_MEMORY] after the memory
957 if (!node_state(nid
, N_MEMORY
))
958 arg
->status_change_nid
= nid
;
960 arg
->status_change_nid
= -1;
963 static void node_states_set_node(int node
, struct memory_notify
*arg
)
965 if (arg
->status_change_nid_normal
>= 0)
966 node_set_state(node
, N_NORMAL_MEMORY
);
968 if (arg
->status_change_nid_high
>= 0)
969 node_set_state(node
, N_HIGH_MEMORY
);
971 node_set_state(node
, N_MEMORY
);
975 /* Must be protected by mem_hotplug_begin() */
976 int __ref
online_pages(unsigned long pfn
, unsigned long nr_pages
, int online_type
)
979 unsigned long onlined_pages
= 0;
981 int need_zonelists_rebuild
= 0;
984 struct memory_notify arg
;
987 * This doesn't need a lock to do pfn_to_page().
988 * The section can't be removed here because of the
989 * memory_block->state_mutex.
991 zone
= page_zone(pfn_to_page(pfn
));
993 if ((zone_idx(zone
) > ZONE_NORMAL
||
994 online_type
== MMOP_ONLINE_MOVABLE
) &&
995 !can_online_high_movable(zone
))
998 if (online_type
== MMOP_ONLINE_KERNEL
&&
999 zone_idx(zone
) == ZONE_MOVABLE
) {
1000 if (move_pfn_range_left(zone
- 1, zone
, pfn
, pfn
+ nr_pages
))
1003 if (online_type
== MMOP_ONLINE_MOVABLE
&&
1004 zone_idx(zone
) == ZONE_MOVABLE
- 1) {
1005 if (move_pfn_range_right(zone
, zone
+ 1, pfn
, pfn
+ nr_pages
))
1009 /* Previous code may changed the zone of the pfn range */
1010 zone
= page_zone(pfn_to_page(pfn
));
1012 arg
.start_pfn
= pfn
;
1013 arg
.nr_pages
= nr_pages
;
1014 node_states_check_changes_online(nr_pages
, zone
, &arg
);
1016 nid
= pfn_to_nid(pfn
);
1018 ret
= memory_notify(MEM_GOING_ONLINE
, &arg
);
1019 ret
= notifier_to_errno(ret
);
1021 memory_notify(MEM_CANCEL_ONLINE
, &arg
);
1025 * If this zone is not populated, then it is not in zonelist.
1026 * This means the page allocator ignores this zone.
1027 * So, zonelist must be updated after online.
1029 mutex_lock(&zonelists_mutex
);
1030 if (!populated_zone(zone
)) {
1031 need_zonelists_rebuild
= 1;
1032 build_all_zonelists(NULL
, zone
);
1035 ret
= walk_system_ram_range(pfn
, nr_pages
, &onlined_pages
,
1036 online_pages_range
);
1038 if (need_zonelists_rebuild
)
1039 zone_pcp_reset(zone
);
1040 mutex_unlock(&zonelists_mutex
);
1041 printk(KERN_DEBUG
"online_pages [mem %#010llx-%#010llx] failed\n",
1042 (unsigned long long) pfn
<< PAGE_SHIFT
,
1043 (((unsigned long long) pfn
+ nr_pages
)
1044 << PAGE_SHIFT
) - 1);
1045 memory_notify(MEM_CANCEL_ONLINE
, &arg
);
1049 zone
->present_pages
+= onlined_pages
;
1051 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
1052 zone
->zone_pgdat
->node_present_pages
+= onlined_pages
;
1053 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
1055 if (onlined_pages
) {
1056 node_states_set_node(zone_to_nid(zone
), &arg
);
1057 if (need_zonelists_rebuild
)
1058 build_all_zonelists(NULL
, NULL
);
1060 zone_pcp_update(zone
);
1063 mutex_unlock(&zonelists_mutex
);
1065 init_per_zone_wmark_min();
1068 kswapd_run(zone_to_nid(zone
));
1070 vm_total_pages
= nr_free_pagecache_pages();
1072 writeback_set_ratelimit();
1075 memory_notify(MEM_ONLINE
, &arg
);
1078 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1080 static void reset_node_present_pages(pg_data_t
*pgdat
)
1084 for (z
= pgdat
->node_zones
; z
< pgdat
->node_zones
+ MAX_NR_ZONES
; z
++)
1085 z
->present_pages
= 0;
1087 pgdat
->node_present_pages
= 0;
1090 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1091 static pg_data_t __ref
*hotadd_new_pgdat(int nid
, u64 start
)
1093 struct pglist_data
*pgdat
;
1094 unsigned long zones_size
[MAX_NR_ZONES
] = {0};
1095 unsigned long zholes_size
[MAX_NR_ZONES
] = {0};
1096 unsigned long start_pfn
= PFN_DOWN(start
);
1098 pgdat
= NODE_DATA(nid
);
1100 pgdat
= arch_alloc_nodedata(nid
);
1104 arch_refresh_nodedata(nid
, pgdat
);
1106 /* Reset the nr_zones and classzone_idx to 0 before reuse */
1107 pgdat
->nr_zones
= 0;
1108 pgdat
->classzone_idx
= 0;
1111 /* we can use NODE_DATA(nid) from here */
1113 /* init node's zones as empty zones, we don't have any present pages.*/
1114 free_area_init_node(nid
, zones_size
, start_pfn
, zholes_size
);
1117 * The node we allocated has no zone fallback lists. For avoiding
1118 * to access not-initialized zonelist, build here.
1120 mutex_lock(&zonelists_mutex
);
1121 build_all_zonelists(pgdat
, NULL
);
1122 mutex_unlock(&zonelists_mutex
);
1125 * zone->managed_pages is set to an approximate value in
1126 * free_area_init_core(), which will cause
1127 * /sys/device/system/node/nodeX/meminfo has wrong data.
1128 * So reset it to 0 before any memory is onlined.
1130 reset_node_managed_pages(pgdat
);
1133 * When memory is hot-added, all the memory is in offline state. So
1134 * clear all zones' present_pages because they will be updated in
1135 * online_pages() and offline_pages().
1137 reset_node_present_pages(pgdat
);
1142 static void rollback_node_hotadd(int nid
, pg_data_t
*pgdat
)
1144 arch_refresh_nodedata(nid
, NULL
);
1145 arch_free_nodedata(pgdat
);
1151 * try_online_node - online a node if offlined
1153 * called by cpu_up() to online a node without onlined memory.
1155 int try_online_node(int nid
)
1160 if (node_online(nid
))
1163 mem_hotplug_begin();
1164 pgdat
= hotadd_new_pgdat(nid
, 0);
1166 pr_err("Cannot online node %d due to NULL pgdat\n", nid
);
1170 node_set_online(nid
);
1171 ret
= register_one_node(nid
);
1174 if (pgdat
->node_zonelists
->_zonerefs
->zone
== NULL
) {
1175 mutex_lock(&zonelists_mutex
);
1176 build_all_zonelists(NULL
, NULL
);
1177 mutex_unlock(&zonelists_mutex
);
1185 static int check_hotplug_memory_range(u64 start
, u64 size
)
1187 u64 start_pfn
= PFN_DOWN(start
);
1188 u64 nr_pages
= size
>> PAGE_SHIFT
;
1190 /* Memory range must be aligned with section */
1191 if ((start_pfn
& ~PAGE_SECTION_MASK
) ||
1192 (nr_pages
% PAGES_PER_SECTION
) || (!nr_pages
)) {
1193 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1194 (unsigned long long)start
,
1195 (unsigned long long)size
);
1203 * If movable zone has already been setup, newly added memory should be check.
1204 * If its address is higher than movable zone, it should be added as movable.
1205 * Without this check, movable zone may overlap with other zone.
1207 static int should_add_memory_movable(int nid
, u64 start
, u64 size
)
1209 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
1210 pg_data_t
*pgdat
= NODE_DATA(nid
);
1211 struct zone
*movable_zone
= pgdat
->node_zones
+ ZONE_MOVABLE
;
1213 if (zone_is_empty(movable_zone
))
1216 if (movable_zone
->zone_start_pfn
<= start_pfn
)
1222 int zone_for_memory(int nid
, u64 start
, u64 size
, int zone_default
,
1225 #ifdef CONFIG_ZONE_DEVICE
1229 if (should_add_memory_movable(nid
, start
, size
))
1230 return ZONE_MOVABLE
;
1232 return zone_default
;
1235 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1236 int __ref
add_memory_resource(int nid
, struct resource
*res
)
1239 pg_data_t
*pgdat
= NULL
;
1245 size
= resource_size(res
);
1247 ret
= check_hotplug_memory_range(start
, size
);
1251 { /* Stupid hack to suppress address-never-null warning */
1252 void *p
= NODE_DATA(nid
);
1256 mem_hotplug_begin();
1259 * Add new range to memblock so that when hotadd_new_pgdat() is called
1260 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1261 * this new range and calculate total pages correctly. The range will
1262 * be removed at hot-remove time.
1264 memblock_add_node(start
, size
, nid
);
1266 new_node
= !node_online(nid
);
1268 pgdat
= hotadd_new_pgdat(nid
, start
);
1274 /* call arch's memory hotadd */
1275 ret
= arch_add_memory(nid
, start
, size
, false);
1280 /* we online node here. we can't roll back from here. */
1281 node_set_online(nid
);
1284 ret
= register_one_node(nid
);
1286 * If sysfs file of new node can't create, cpu on the node
1287 * can't be hot-added. There is no rollback way now.
1288 * So, check by BUG_ON() to catch it reluctantly..
1293 /* create new memmap entry */
1294 firmware_map_add_hotplug(start
, start
+ size
, "System RAM");
1299 /* rollback pgdat allocation and others */
1301 rollback_node_hotadd(nid
, pgdat
);
1302 memblock_remove(start
, size
);
1308 EXPORT_SYMBOL_GPL(add_memory_resource
);
1310 int __ref
add_memory(int nid
, u64 start
, u64 size
)
1312 struct resource
*res
;
1315 res
= register_memory_resource(start
, size
);
1317 return PTR_ERR(res
);
1319 ret
= add_memory_resource(nid
, res
);
1321 release_memory_resource(res
);
1324 EXPORT_SYMBOL_GPL(add_memory
);
1326 #ifdef CONFIG_MEMORY_HOTREMOVE
1328 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1329 * set and the size of the free page is given by page_order(). Using this,
1330 * the function determines if the pageblock contains only free pages.
1331 * Due to buddy contraints, a free page at least the size of a pageblock will
1332 * be located at the start of the pageblock
1334 static inline int pageblock_free(struct page
*page
)
1336 return PageBuddy(page
) && page_order(page
) >= pageblock_order
;
1339 /* Return the start of the next active pageblock after a given page */
1340 static struct page
*next_active_pageblock(struct page
*page
)
1342 /* Ensure the starting page is pageblock-aligned */
1343 BUG_ON(page_to_pfn(page
) & (pageblock_nr_pages
- 1));
1345 /* If the entire pageblock is free, move to the end of free page */
1346 if (pageblock_free(page
)) {
1348 /* be careful. we don't have locks, page_order can be changed.*/
1349 order
= page_order(page
);
1350 if ((order
< MAX_ORDER
) && (order
>= pageblock_order
))
1351 return page
+ (1 << order
);
1354 return page
+ pageblock_nr_pages
;
1357 /* Checks if this range of memory is likely to be hot-removable. */
1358 int is_mem_section_removable(unsigned long start_pfn
, unsigned long nr_pages
)
1360 struct page
*page
= pfn_to_page(start_pfn
);
1361 struct page
*end_page
= page
+ nr_pages
;
1363 /* Check the starting page of each pageblock within the range */
1364 for (; page
< end_page
; page
= next_active_pageblock(page
)) {
1365 if (!is_pageblock_removable_nolock(page
))
1370 /* All pageblocks in the memory block are likely to be hot-removable */
1375 * Confirm all pages in a range [start, end) is belongs to the same zone.
1377 int test_pages_in_a_zone(unsigned long start_pfn
, unsigned long end_pfn
)
1379 unsigned long pfn
, sec_end_pfn
;
1380 struct zone
*zone
= NULL
;
1383 for (pfn
= start_pfn
, sec_end_pfn
= SECTION_ALIGN_UP(start_pfn
);
1385 pfn
= sec_end_pfn
+ 1, sec_end_pfn
+= PAGES_PER_SECTION
) {
1386 /* Make sure the memory section is present first */
1387 if (!present_section_nr(pfn_to_section_nr(pfn
)))
1389 for (; pfn
< sec_end_pfn
&& pfn
< end_pfn
;
1390 pfn
+= MAX_ORDER_NR_PAGES
) {
1392 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1393 while ((i
< MAX_ORDER_NR_PAGES
) &&
1394 !pfn_valid_within(pfn
+ i
))
1396 if (i
== MAX_ORDER_NR_PAGES
)
1398 page
= pfn_to_page(pfn
+ i
);
1399 if (zone
&& page_zone(page
) != zone
)
1401 zone
= page_zone(page
);
1408 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages
1409 * and hugepages). We scan pfn because it's much easier than scanning over
1410 * linked list. This function returns the pfn of the first found movable
1411 * page if it's found, otherwise 0.
1413 static unsigned long scan_movable_pages(unsigned long start
, unsigned long end
)
1417 for (pfn
= start
; pfn
< end
; pfn
++) {
1418 if (pfn_valid(pfn
)) {
1419 page
= pfn_to_page(pfn
);
1422 if (PageHuge(page
)) {
1423 if (page_huge_active(page
))
1426 pfn
= round_up(pfn
+ 1,
1427 1 << compound_order(page
)) - 1;
1434 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1436 do_migrate_range(unsigned long start_pfn
, unsigned long end_pfn
)
1440 int move_pages
= NR_OFFLINE_AT_ONCE_PAGES
;
1441 int not_managed
= 0;
1445 for (pfn
= start_pfn
; pfn
< end_pfn
&& move_pages
> 0; pfn
++) {
1446 if (!pfn_valid(pfn
))
1448 page
= pfn_to_page(pfn
);
1450 if (PageHuge(page
)) {
1451 struct page
*head
= compound_head(page
);
1452 pfn
= page_to_pfn(head
) + (1<<compound_order(head
)) - 1;
1453 if (compound_order(head
) > PFN_SECTION_SHIFT
) {
1457 if (isolate_huge_page(page
, &source
))
1458 move_pages
-= 1 << compound_order(head
);
1462 if (!get_page_unless_zero(page
))
1465 * We can skip free pages. And we can only deal with pages on
1468 ret
= isolate_lru_page(page
);
1469 if (!ret
) { /* Success */
1471 list_add_tail(&page
->lru
, &source
);
1473 inc_zone_page_state(page
, NR_ISOLATED_ANON
+
1474 page_is_file_cache(page
));
1477 #ifdef CONFIG_DEBUG_VM
1478 printk(KERN_ALERT
"removing pfn %lx from LRU failed\n",
1480 dump_page(page
, "failed to remove from LRU");
1483 /* Because we don't have big zone->lock. we should
1484 check this again here. */
1485 if (page_count(page
)) {
1492 if (!list_empty(&source
)) {
1494 putback_movable_pages(&source
);
1499 * alloc_migrate_target should be improooooved!!
1500 * migrate_pages returns # of failed pages.
1502 ret
= migrate_pages(&source
, alloc_migrate_target
, NULL
, 0,
1503 MIGRATE_SYNC
, MR_MEMORY_HOTPLUG
);
1505 putback_movable_pages(&source
);
1512 * remove from free_area[] and mark all as Reserved.
1515 offline_isolated_pages_cb(unsigned long start
, unsigned long nr_pages
,
1518 __offline_isolated_pages(start
, start
+ nr_pages
);
1523 offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
1525 walk_system_ram_range(start_pfn
, end_pfn
- start_pfn
, NULL
,
1526 offline_isolated_pages_cb
);
1530 * Check all pages in range, recoreded as memory resource, are isolated.
1533 check_pages_isolated_cb(unsigned long start_pfn
, unsigned long nr_pages
,
1537 long offlined
= *(long *)data
;
1538 ret
= test_pages_isolated(start_pfn
, start_pfn
+ nr_pages
, true);
1539 offlined
= nr_pages
;
1541 *(long *)data
+= offlined
;
1546 check_pages_isolated(unsigned long start_pfn
, unsigned long end_pfn
)
1551 ret
= walk_system_ram_range(start_pfn
, end_pfn
- start_pfn
, &offlined
,
1552 check_pages_isolated_cb
);
1554 offlined
= (long)ret
;
1558 #ifdef CONFIG_MOVABLE_NODE
1560 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1563 static bool can_offline_normal(struct zone
*zone
, unsigned long nr_pages
)
1567 #else /* CONFIG_MOVABLE_NODE */
1568 /* ensure the node has NORMAL memory if it is still online */
1569 static bool can_offline_normal(struct zone
*zone
, unsigned long nr_pages
)
1571 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1572 unsigned long present_pages
= 0;
1575 for (zt
= 0; zt
<= ZONE_NORMAL
; zt
++)
1576 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1578 if (present_pages
> nr_pages
)
1582 for (; zt
<= ZONE_MOVABLE
; zt
++)
1583 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1586 * we can't offline the last normal memory until all
1587 * higher memory is offlined.
1589 return present_pages
== 0;
1591 #endif /* CONFIG_MOVABLE_NODE */
1593 static int __init
cmdline_parse_movable_node(char *p
)
1595 #ifdef CONFIG_MOVABLE_NODE
1597 * Memory used by the kernel cannot be hot-removed because Linux
1598 * cannot migrate the kernel pages. When memory hotplug is
1599 * enabled, we should prevent memblock from allocating memory
1602 * ACPI SRAT records all hotpluggable memory ranges. But before
1603 * SRAT is parsed, we don't know about it.
1605 * The kernel image is loaded into memory at very early time. We
1606 * cannot prevent this anyway. So on NUMA system, we set any
1607 * node the kernel resides in as un-hotpluggable.
1609 * Since on modern servers, one node could have double-digit
1610 * gigabytes memory, we can assume the memory around the kernel
1611 * image is also un-hotpluggable. So before SRAT is parsed, just
1612 * allocate memory near the kernel image to try the best to keep
1613 * the kernel away from hotpluggable memory.
1615 memblock_set_bottom_up(true);
1616 movable_node_enabled
= true;
1618 pr_warn("movable_node option not supported\n");
1622 early_param("movable_node", cmdline_parse_movable_node
);
1624 /* check which state of node_states will be changed when offline memory */
1625 static void node_states_check_changes_offline(unsigned long nr_pages
,
1626 struct zone
*zone
, struct memory_notify
*arg
)
1628 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1629 unsigned long present_pages
= 0;
1630 enum zone_type zt
, zone_last
= ZONE_NORMAL
;
1633 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1634 * contains nodes which have zones of 0...ZONE_NORMAL,
1635 * set zone_last to ZONE_NORMAL.
1637 * If we don't have HIGHMEM nor movable node,
1638 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1639 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1641 if (N_MEMORY
== N_NORMAL_MEMORY
)
1642 zone_last
= ZONE_MOVABLE
;
1645 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1646 * If the memory to be offline is in a zone of 0...zone_last,
1647 * and it is the last present memory, 0...zone_last will
1648 * become empty after offline , thus we can determind we will
1649 * need to clear the node from node_states[N_NORMAL_MEMORY].
1651 for (zt
= 0; zt
<= zone_last
; zt
++)
1652 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1653 if (zone_idx(zone
) <= zone_last
&& nr_pages
>= present_pages
)
1654 arg
->status_change_nid_normal
= zone_to_nid(zone
);
1656 arg
->status_change_nid_normal
= -1;
1658 #ifdef CONFIG_HIGHMEM
1660 * If we have movable node, node_states[N_HIGH_MEMORY]
1661 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1662 * set zone_last to ZONE_HIGHMEM.
1664 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1665 * contains nodes which have zones of 0...ZONE_MOVABLE,
1666 * set zone_last to ZONE_MOVABLE.
1668 zone_last
= ZONE_HIGHMEM
;
1669 if (N_MEMORY
== N_HIGH_MEMORY
)
1670 zone_last
= ZONE_MOVABLE
;
1672 for (; zt
<= zone_last
; zt
++)
1673 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1674 if (zone_idx(zone
) <= zone_last
&& nr_pages
>= present_pages
)
1675 arg
->status_change_nid_high
= zone_to_nid(zone
);
1677 arg
->status_change_nid_high
= -1;
1679 arg
->status_change_nid_high
= arg
->status_change_nid_normal
;
1683 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1685 zone_last
= ZONE_MOVABLE
;
1688 * check whether node_states[N_HIGH_MEMORY] will be changed
1689 * If we try to offline the last present @nr_pages from the node,
1690 * we can determind we will need to clear the node from
1691 * node_states[N_HIGH_MEMORY].
1693 for (; zt
<= zone_last
; zt
++)
1694 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1695 if (nr_pages
>= present_pages
)
1696 arg
->status_change_nid
= zone_to_nid(zone
);
1698 arg
->status_change_nid
= -1;
1701 static void node_states_clear_node(int node
, struct memory_notify
*arg
)
1703 if (arg
->status_change_nid_normal
>= 0)
1704 node_clear_state(node
, N_NORMAL_MEMORY
);
1706 if ((N_MEMORY
!= N_NORMAL_MEMORY
) &&
1707 (arg
->status_change_nid_high
>= 0))
1708 node_clear_state(node
, N_HIGH_MEMORY
);
1710 if ((N_MEMORY
!= N_HIGH_MEMORY
) &&
1711 (arg
->status_change_nid
>= 0))
1712 node_clear_state(node
, N_MEMORY
);
1715 static int __ref
__offline_pages(unsigned long start_pfn
,
1716 unsigned long end_pfn
, unsigned long timeout
)
1718 unsigned long pfn
, nr_pages
, expire
;
1719 long offlined_pages
;
1720 int ret
, drain
, retry_max
, node
;
1721 unsigned long flags
;
1723 struct memory_notify arg
;
1725 /* at least, alignment against pageblock is necessary */
1726 if (!IS_ALIGNED(start_pfn
, pageblock_nr_pages
))
1728 if (!IS_ALIGNED(end_pfn
, pageblock_nr_pages
))
1730 /* This makes hotplug much easier...and readable.
1731 we assume this for now. .*/
1732 if (!test_pages_in_a_zone(start_pfn
, end_pfn
))
1735 zone
= page_zone(pfn_to_page(start_pfn
));
1736 node
= zone_to_nid(zone
);
1737 nr_pages
= end_pfn
- start_pfn
;
1739 if (zone_idx(zone
) <= ZONE_NORMAL
&& !can_offline_normal(zone
, nr_pages
))
1742 /* set above range as isolated */
1743 ret
= start_isolate_page_range(start_pfn
, end_pfn
,
1744 MIGRATE_MOVABLE
, true);
1748 arg
.start_pfn
= start_pfn
;
1749 arg
.nr_pages
= nr_pages
;
1750 node_states_check_changes_offline(nr_pages
, zone
, &arg
);
1752 ret
= memory_notify(MEM_GOING_OFFLINE
, &arg
);
1753 ret
= notifier_to_errno(ret
);
1755 goto failed_removal
;
1758 expire
= jiffies
+ timeout
;
1762 /* start memory hot removal */
1764 if (time_after(jiffies
, expire
))
1765 goto failed_removal
;
1767 if (signal_pending(current
))
1768 goto failed_removal
;
1771 lru_add_drain_all();
1773 drain_all_pages(zone
);
1776 pfn
= scan_movable_pages(start_pfn
, end_pfn
);
1777 if (pfn
) { /* We have movable pages */
1778 ret
= do_migrate_range(pfn
, end_pfn
);
1784 if (--retry_max
== 0)
1785 goto failed_removal
;
1791 /* drain all zone's lru pagevec, this is asynchronous... */
1792 lru_add_drain_all();
1794 /* drain pcp pages, this is synchronous. */
1795 drain_all_pages(zone
);
1797 * dissolve free hugepages in the memory block before doing offlining
1798 * actually in order to make hugetlbfs's object counting consistent.
1800 dissolve_free_huge_pages(start_pfn
, end_pfn
);
1802 offlined_pages
= check_pages_isolated(start_pfn
, end_pfn
);
1803 if (offlined_pages
< 0) {
1805 goto failed_removal
;
1807 printk(KERN_INFO
"Offlined Pages %ld\n", offlined_pages
);
1808 /* Ok, all of our target is isolated.
1809 We cannot do rollback at this point. */
1810 offline_isolated_pages(start_pfn
, end_pfn
);
1811 /* reset pagetype flags and makes migrate type to be MOVABLE */
1812 undo_isolate_page_range(start_pfn
, end_pfn
, MIGRATE_MOVABLE
);
1813 /* removal success */
1814 adjust_managed_page_count(pfn_to_page(start_pfn
), -offlined_pages
);
1815 zone
->present_pages
-= offlined_pages
;
1817 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
1818 zone
->zone_pgdat
->node_present_pages
-= offlined_pages
;
1819 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
1821 init_per_zone_wmark_min();
1823 if (!populated_zone(zone
)) {
1824 zone_pcp_reset(zone
);
1825 mutex_lock(&zonelists_mutex
);
1826 build_all_zonelists(NULL
, NULL
);
1827 mutex_unlock(&zonelists_mutex
);
1829 zone_pcp_update(zone
);
1831 node_states_clear_node(node
, &arg
);
1832 if (arg
.status_change_nid
>= 0)
1835 vm_total_pages
= nr_free_pagecache_pages();
1836 writeback_set_ratelimit();
1838 memory_notify(MEM_OFFLINE
, &arg
);
1842 printk(KERN_INFO
"memory offlining [mem %#010llx-%#010llx] failed\n",
1843 (unsigned long long) start_pfn
<< PAGE_SHIFT
,
1844 ((unsigned long long) end_pfn
<< PAGE_SHIFT
) - 1);
1845 memory_notify(MEM_CANCEL_OFFLINE
, &arg
);
1846 /* pushback to free area */
1847 undo_isolate_page_range(start_pfn
, end_pfn
, MIGRATE_MOVABLE
);
1851 /* Must be protected by mem_hotplug_begin() */
1852 int offline_pages(unsigned long start_pfn
, unsigned long nr_pages
)
1854 return __offline_pages(start_pfn
, start_pfn
+ nr_pages
, 120 * HZ
);
1856 #endif /* CONFIG_MEMORY_HOTREMOVE */
1859 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1860 * @start_pfn: start pfn of the memory range
1861 * @end_pfn: end pfn of the memory range
1862 * @arg: argument passed to func
1863 * @func: callback for each memory section walked
1865 * This function walks through all present mem sections in range
1866 * [start_pfn, end_pfn) and call func on each mem section.
1868 * Returns the return value of func.
1870 int walk_memory_range(unsigned long start_pfn
, unsigned long end_pfn
,
1871 void *arg
, int (*func
)(struct memory_block
*, void *))
1873 struct memory_block
*mem
= NULL
;
1874 struct mem_section
*section
;
1875 unsigned long pfn
, section_nr
;
1878 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
1879 section_nr
= pfn_to_section_nr(pfn
);
1880 if (!present_section_nr(section_nr
))
1883 section
= __nr_to_section(section_nr
);
1884 /* same memblock? */
1886 if ((section_nr
>= mem
->start_section_nr
) &&
1887 (section_nr
<= mem
->end_section_nr
))
1890 mem
= find_memory_block_hinted(section
, mem
);
1894 ret
= func(mem
, arg
);
1896 kobject_put(&mem
->dev
.kobj
);
1902 kobject_put(&mem
->dev
.kobj
);
1907 #ifdef CONFIG_MEMORY_HOTREMOVE
1908 static int check_memblock_offlined_cb(struct memory_block
*mem
, void *arg
)
1910 int ret
= !is_memblock_offlined(mem
);
1912 if (unlikely(ret
)) {
1913 phys_addr_t beginpa
, endpa
;
1915 beginpa
= PFN_PHYS(section_nr_to_pfn(mem
->start_section_nr
));
1916 endpa
= PFN_PHYS(section_nr_to_pfn(mem
->end_section_nr
+ 1))-1;
1917 pr_warn("removing memory fails, because memory "
1918 "[%pa-%pa] is onlined\n",
1925 static int check_cpu_on_node(pg_data_t
*pgdat
)
1929 for_each_present_cpu(cpu
) {
1930 if (cpu_to_node(cpu
) == pgdat
->node_id
)
1932 * the cpu on this node isn't removed, and we can't
1933 * offline this node.
1941 static void unmap_cpu_on_node(pg_data_t
*pgdat
)
1943 #ifdef CONFIG_ACPI_NUMA
1946 for_each_possible_cpu(cpu
)
1947 if (cpu_to_node(cpu
) == pgdat
->node_id
)
1948 numa_clear_node(cpu
);
1952 static int check_and_unmap_cpu_on_node(pg_data_t
*pgdat
)
1956 ret
= check_cpu_on_node(pgdat
);
1961 * the node will be offlined when we come here, so we can clear
1962 * the cpu_to_node() now.
1965 unmap_cpu_on_node(pgdat
);
1972 * Offline a node if all memory sections and cpus of the node are removed.
1974 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1975 * and online/offline operations before this call.
1977 void try_offline_node(int nid
)
1979 pg_data_t
*pgdat
= NODE_DATA(nid
);
1980 unsigned long start_pfn
= pgdat
->node_start_pfn
;
1981 unsigned long end_pfn
= start_pfn
+ pgdat
->node_spanned_pages
;
1985 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
1986 unsigned long section_nr
= pfn_to_section_nr(pfn
);
1988 if (!present_section_nr(section_nr
))
1991 if (pfn_to_nid(pfn
) != nid
)
1995 * some memory sections of this node are not removed, and we
1996 * can't offline node now.
2001 if (check_and_unmap_cpu_on_node(pgdat
))
2005 * all memory/cpu of this node are removed, we can offline this
2008 node_set_offline(nid
);
2009 unregister_one_node(nid
);
2011 /* free waittable in each zone */
2012 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2013 struct zone
*zone
= pgdat
->node_zones
+ i
;
2016 * wait_table may be allocated from boot memory,
2017 * here only free if it's allocated by vmalloc.
2019 if (is_vmalloc_addr(zone
->wait_table
)) {
2020 vfree(zone
->wait_table
);
2021 zone
->wait_table
= NULL
;
2025 EXPORT_SYMBOL(try_offline_node
);
2030 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2031 * and online/offline operations before this call, as required by
2032 * try_offline_node().
2034 void __ref
remove_memory(int nid
, u64 start
, u64 size
)
2038 BUG_ON(check_hotplug_memory_range(start
, size
));
2040 mem_hotplug_begin();
2043 * All memory blocks must be offlined before removing memory. Check
2044 * whether all memory blocks in question are offline and trigger a BUG()
2045 * if this is not the case.
2047 ret
= walk_memory_range(PFN_DOWN(start
), PFN_UP(start
+ size
- 1), NULL
,
2048 check_memblock_offlined_cb
);
2052 /* remove memmap entry */
2053 firmware_map_remove(start
, start
+ size
, "System RAM");
2054 memblock_free(start
, size
);
2055 memblock_remove(start
, size
);
2057 arch_remove_memory(start
, size
);
2059 try_offline_node(nid
);
2063 EXPORT_SYMBOL_GPL(remove_memory
);
2064 #endif /* CONFIG_MEMORY_HOTREMOVE */