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/module.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/cpuset.h>
26 #include <linux/delay.h>
27 #include <linux/migrate.h>
28 #include <linux/page-isolation.h>
30 #include <asm/tlbflush.h>
34 /* add this memory to iomem resource */
35 static struct resource
*register_memory_resource(u64 start
, u64 size
)
38 res
= kzalloc(sizeof(struct resource
), GFP_KERNEL
);
41 res
->name
= "System RAM";
43 res
->end
= start
+ size
- 1;
44 res
->flags
= IORESOURCE_MEM
| IORESOURCE_BUSY
;
45 if (request_resource(&iomem_resource
, res
) < 0) {
46 printk("System RAM resource %llx - %llx cannot be added\n",
47 (unsigned long long)res
->start
, (unsigned long long)res
->end
);
54 static void release_memory_resource(struct resource
*res
)
58 release_resource(res
);
63 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
64 #ifndef CONFIG_SPARSEMEM_VMEMMAP
65 static void get_page_bootmem(unsigned long info
, struct page
*page
, int magic
)
67 atomic_set(&page
->_mapcount
, magic
);
69 set_page_private(page
, info
);
70 atomic_inc(&page
->_count
);
73 void put_page_bootmem(struct page
*page
)
77 magic
= atomic_read(&page
->_mapcount
);
80 if (atomic_dec_return(&page
->_count
) == 1) {
81 ClearPagePrivate(page
);
82 set_page_private(page
, 0);
83 reset_page_mapcount(page
);
84 __free_pages_bootmem(page
, 0);
89 void register_page_bootmem_info_section(unsigned long start_pfn
)
91 unsigned long *usemap
, mapsize
, section_nr
, i
;
92 struct mem_section
*ms
;
93 struct page
*page
, *memmap
;
95 if (!pfn_valid(start_pfn
))
98 section_nr
= pfn_to_section_nr(start_pfn
);
99 ms
= __nr_to_section(section_nr
);
101 /* Get section's memmap address */
102 memmap
= sparse_decode_mem_map(ms
->section_mem_map
, section_nr
);
105 * Get page for the memmap's phys address
106 * XXX: need more consideration for sparse_vmemmap...
108 page
= virt_to_page(memmap
);
109 mapsize
= sizeof(struct page
) * PAGES_PER_SECTION
;
110 mapsize
= PAGE_ALIGN(mapsize
) >> PAGE_SHIFT
;
112 /* remember memmap's page */
113 for (i
= 0; i
< mapsize
; i
++, page
++)
114 get_page_bootmem(section_nr
, page
, SECTION_INFO
);
116 usemap
= __nr_to_section(section_nr
)->pageblock_flags
;
117 page
= virt_to_page(usemap
);
119 mapsize
= PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT
;
121 for (i
= 0; i
< mapsize
; i
++, page
++)
122 get_page_bootmem(section_nr
, page
, MIX_INFO
);
126 void register_page_bootmem_info_node(struct pglist_data
*pgdat
)
128 unsigned long i
, pfn
, end_pfn
, nr_pages
;
129 int node
= pgdat
->node_id
;
133 nr_pages
= PAGE_ALIGN(sizeof(struct pglist_data
)) >> PAGE_SHIFT
;
134 page
= virt_to_page(pgdat
);
136 for (i
= 0; i
< nr_pages
; i
++, page
++)
137 get_page_bootmem(node
, page
, NODE_INFO
);
139 zone
= &pgdat
->node_zones
[0];
140 for (; zone
< pgdat
->node_zones
+ MAX_NR_ZONES
- 1; zone
++) {
141 if (zone
->wait_table
) {
142 nr_pages
= zone
->wait_table_hash_nr_entries
143 * sizeof(wait_queue_head_t
);
144 nr_pages
= PAGE_ALIGN(nr_pages
) >> PAGE_SHIFT
;
145 page
= virt_to_page(zone
->wait_table
);
147 for (i
= 0; i
< nr_pages
; i
++, page
++)
148 get_page_bootmem(node
, page
, NODE_INFO
);
152 pfn
= pgdat
->node_start_pfn
;
153 end_pfn
= pfn
+ pgdat
->node_spanned_pages
;
155 /* register_section info */
156 for (; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
)
157 register_page_bootmem_info_section(pfn
);
160 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
162 static int __add_zone(struct zone
*zone
, unsigned long phys_start_pfn
)
164 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
165 int nr_pages
= PAGES_PER_SECTION
;
166 int nid
= pgdat
->node_id
;
169 zone_type
= zone
- pgdat
->node_zones
;
170 if (!zone
->wait_table
)
171 return init_currently_empty_zone(zone
, phys_start_pfn
,
172 nr_pages
, MEMMAP_HOTPLUG
);
173 memmap_init_zone(nr_pages
, nid
, zone_type
,
174 phys_start_pfn
, MEMMAP_HOTPLUG
);
178 static int __add_section(struct zone
*zone
, unsigned long phys_start_pfn
)
180 int nr_pages
= PAGES_PER_SECTION
;
183 if (pfn_valid(phys_start_pfn
))
186 ret
= sparse_add_one_section(zone
, phys_start_pfn
, nr_pages
);
191 ret
= __add_zone(zone
, phys_start_pfn
);
196 return register_new_memory(__pfn_to_section(phys_start_pfn
));
199 #ifdef CONFIG_SPARSEMEM_VMEMMAP
200 static int __remove_section(struct zone
*zone
, struct mem_section
*ms
)
203 * XXX: Freeing memmap with vmemmap is not implement yet.
204 * This should be removed later.
209 static int __remove_section(struct zone
*zone
, struct mem_section
*ms
)
212 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
215 if (!valid_section(ms
))
218 ret
= unregister_memory_section(ms
);
222 pgdat_resize_lock(pgdat
, &flags
);
223 sparse_remove_one_section(zone
, ms
);
224 pgdat_resize_unlock(pgdat
, &flags
);
230 * Reasonably generic function for adding memory. It is
231 * expected that archs that support memory hotplug will
232 * call this function after deciding the zone to which to
235 int __add_pages(struct zone
*zone
, unsigned long phys_start_pfn
,
236 unsigned long nr_pages
)
240 int start_sec
, end_sec
;
241 /* during initialize mem_map, align hot-added range to section */
242 start_sec
= pfn_to_section_nr(phys_start_pfn
);
243 end_sec
= pfn_to_section_nr(phys_start_pfn
+ nr_pages
- 1);
245 for (i
= start_sec
; i
<= end_sec
; i
++) {
246 err
= __add_section(zone
, i
<< PFN_SECTION_SHIFT
);
249 * EEXIST is finally dealt with by ioresource collision
250 * check. see add_memory() => register_memory_resource()
251 * Warning will be printed if there is collision.
253 if (err
&& (err
!= -EEXIST
))
260 EXPORT_SYMBOL_GPL(__add_pages
);
263 * __remove_pages() - remove sections of pages from a zone
264 * @zone: zone from which pages need to be removed
265 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
266 * @nr_pages: number of pages to remove (must be multiple of section size)
268 * Generic helper function to remove section mappings and sysfs entries
269 * for the section of the memory we are removing. Caller needs to make
270 * sure that pages are marked reserved and zones are adjust properly by
271 * calling offline_pages().
273 int __remove_pages(struct zone
*zone
, unsigned long phys_start_pfn
,
274 unsigned long nr_pages
)
276 unsigned long i
, ret
= 0;
277 int sections_to_remove
;
280 * We can only remove entire sections
282 BUG_ON(phys_start_pfn
& ~PAGE_SECTION_MASK
);
283 BUG_ON(nr_pages
% PAGES_PER_SECTION
);
285 release_mem_region(phys_start_pfn
<< PAGE_SHIFT
, nr_pages
* PAGE_SIZE
);
287 sections_to_remove
= nr_pages
/ PAGES_PER_SECTION
;
288 for (i
= 0; i
< sections_to_remove
; i
++) {
289 unsigned long pfn
= phys_start_pfn
+ i
*PAGES_PER_SECTION
;
290 ret
= __remove_section(zone
, __pfn_to_section(pfn
));
296 EXPORT_SYMBOL_GPL(__remove_pages
);
298 static void grow_zone_span(struct zone
*zone
,
299 unsigned long start_pfn
, unsigned long end_pfn
)
301 unsigned long old_zone_end_pfn
;
303 zone_span_writelock(zone
);
305 old_zone_end_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
306 if (start_pfn
< zone
->zone_start_pfn
)
307 zone
->zone_start_pfn
= start_pfn
;
309 zone
->spanned_pages
= max(old_zone_end_pfn
, end_pfn
) -
310 zone
->zone_start_pfn
;
312 zone_span_writeunlock(zone
);
315 static void grow_pgdat_span(struct pglist_data
*pgdat
,
316 unsigned long start_pfn
, unsigned long end_pfn
)
318 unsigned long old_pgdat_end_pfn
=
319 pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
321 if (start_pfn
< pgdat
->node_start_pfn
)
322 pgdat
->node_start_pfn
= start_pfn
;
324 pgdat
->node_spanned_pages
= max(old_pgdat_end_pfn
, end_pfn
) -
325 pgdat
->node_start_pfn
;
328 void online_page(struct page
*page
)
333 #ifdef CONFIG_HIGHMEM
334 if (PageHighMem(page
))
338 #ifdef CONFIG_FLATMEM
339 max_mapnr
= max(page_to_pfn(page
), max_mapnr
);
342 ClearPageReserved(page
);
343 init_page_count(page
);
347 static int online_pages_range(unsigned long start_pfn
, unsigned long nr_pages
,
351 unsigned long onlined_pages
= *(unsigned long *)arg
;
353 if (PageReserved(pfn_to_page(start_pfn
)))
354 for (i
= 0; i
< nr_pages
; i
++) {
355 page
= pfn_to_page(start_pfn
+ i
);
359 *(unsigned long *)arg
= onlined_pages
;
364 int online_pages(unsigned long pfn
, unsigned long nr_pages
)
367 unsigned long onlined_pages
= 0;
369 int need_zonelists_rebuild
= 0;
372 struct memory_notify arg
;
375 arg
.nr_pages
= nr_pages
;
376 arg
.status_change_nid
= -1;
378 nid
= page_to_nid(pfn_to_page(pfn
));
379 if (node_present_pages(nid
) == 0)
380 arg
.status_change_nid
= nid
;
382 ret
= memory_notify(MEM_GOING_ONLINE
, &arg
);
383 ret
= notifier_to_errno(ret
);
385 memory_notify(MEM_CANCEL_ONLINE
, &arg
);
389 * This doesn't need a lock to do pfn_to_page().
390 * The section can't be removed here because of the
391 * memory_block->state_mutex.
393 zone
= page_zone(pfn_to_page(pfn
));
394 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
395 grow_zone_span(zone
, pfn
, pfn
+ nr_pages
);
396 grow_pgdat_span(zone
->zone_pgdat
, pfn
, pfn
+ nr_pages
);
397 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
400 * If this zone is not populated, then it is not in zonelist.
401 * This means the page allocator ignores this zone.
402 * So, zonelist must be updated after online.
404 if (!populated_zone(zone
))
405 need_zonelists_rebuild
= 1;
407 ret
= walk_memory_resource(pfn
, nr_pages
, &onlined_pages
,
410 printk(KERN_DEBUG
"online_pages %lx at %lx failed\n",
412 memory_notify(MEM_CANCEL_ONLINE
, &arg
);
416 zone
->present_pages
+= onlined_pages
;
417 zone
->zone_pgdat
->node_present_pages
+= onlined_pages
;
419 setup_per_zone_pages_min();
421 kswapd_run(zone_to_nid(zone
));
422 node_set_state(zone_to_nid(zone
), N_HIGH_MEMORY
);
425 if (need_zonelists_rebuild
)
426 build_all_zonelists();
427 vm_total_pages
= nr_free_pagecache_pages();
428 writeback_set_ratelimit();
431 memory_notify(MEM_ONLINE
, &arg
);
435 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
437 static pg_data_t
*hotadd_new_pgdat(int nid
, u64 start
)
439 struct pglist_data
*pgdat
;
440 unsigned long zones_size
[MAX_NR_ZONES
] = {0};
441 unsigned long zholes_size
[MAX_NR_ZONES
] = {0};
442 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
444 pgdat
= arch_alloc_nodedata(nid
);
448 arch_refresh_nodedata(nid
, pgdat
);
450 /* we can use NODE_DATA(nid) from here */
452 /* init node's zones as empty zones, we don't have any present pages.*/
453 free_area_init_node(nid
, pgdat
, zones_size
, start_pfn
, zholes_size
);
458 static void rollback_node_hotadd(int nid
, pg_data_t
*pgdat
)
460 arch_refresh_nodedata(nid
, NULL
);
461 arch_free_nodedata(pgdat
);
466 int add_memory(int nid
, u64 start
, u64 size
)
468 pg_data_t
*pgdat
= NULL
;
470 struct resource
*res
;
473 res
= register_memory_resource(start
, size
);
477 if (!node_online(nid
)) {
478 pgdat
= hotadd_new_pgdat(nid
, start
);
484 /* call arch's memory hotadd */
485 ret
= arch_add_memory(nid
, start
, size
);
490 /* we online node here. we can't roll back from here. */
491 node_set_online(nid
);
493 cpuset_track_online_nodes();
496 ret
= register_one_node(nid
);
498 * If sysfs file of new node can't create, cpu on the node
499 * can't be hot-added. There is no rollback way now.
500 * So, check by BUG_ON() to catch it reluctantly..
507 /* rollback pgdat allocation and others */
509 rollback_node_hotadd(nid
, pgdat
);
511 release_memory_resource(res
);
515 EXPORT_SYMBOL_GPL(add_memory
);
517 #ifdef CONFIG_MEMORY_HOTREMOVE
519 * Confirm all pages in a range [start, end) is belongs to the same zone.
521 static int test_pages_in_a_zone(unsigned long start_pfn
, unsigned long end_pfn
)
524 struct zone
*zone
= NULL
;
527 for (pfn
= start_pfn
;
529 pfn
+= MAX_ORDER_NR_PAGES
) {
531 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
532 while ((i
< MAX_ORDER_NR_PAGES
) && !pfn_valid_within(pfn
+ i
))
534 if (i
== MAX_ORDER_NR_PAGES
)
536 page
= pfn_to_page(pfn
+ i
);
537 if (zone
&& page_zone(page
) != zone
)
539 zone
= page_zone(page
);
545 * Scanning pfn is much easier than scanning lru list.
546 * Scan pfn from start to end and Find LRU page.
548 int scan_lru_pages(unsigned long start
, unsigned long end
)
552 for (pfn
= start
; pfn
< end
; pfn
++) {
553 if (pfn_valid(pfn
)) {
554 page
= pfn_to_page(pfn
);
563 hotremove_migrate_alloc(struct page
*page
,
564 unsigned long private,
567 /* This should be improoooooved!! */
568 return alloc_page(GFP_HIGHUSER_PAGECACHE
);
572 #define NR_OFFLINE_AT_ONCE_PAGES (256)
574 do_migrate_range(unsigned long start_pfn
, unsigned long end_pfn
)
578 int move_pages
= NR_OFFLINE_AT_ONCE_PAGES
;
583 for (pfn
= start_pfn
; pfn
< end_pfn
&& move_pages
> 0; pfn
++) {
586 page
= pfn_to_page(pfn
);
587 if (!page_count(page
))
590 * We can skip free pages. And we can only deal with pages on
593 ret
= isolate_lru_page(page
, &source
);
594 if (!ret
) { /* Success */
597 /* Becasue we don't have big zone->lock. we should
598 check this again here. */
599 if (page_count(page
))
601 #ifdef CONFIG_DEBUG_VM
602 printk(KERN_INFO
"removing from LRU failed"
604 pfn
, page_count(page
), page
->flags
);
610 if (!list_empty(&source
))
611 putback_lru_pages(&source
);
615 if (list_empty(&source
))
617 /* this function returns # of failed pages */
618 ret
= migrate_pages(&source
, hotremove_migrate_alloc
, 0);
625 * remove from free_area[] and mark all as Reserved.
628 offline_isolated_pages_cb(unsigned long start
, unsigned long nr_pages
,
631 __offline_isolated_pages(start
, start
+ nr_pages
);
636 offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
638 walk_memory_resource(start_pfn
, end_pfn
- start_pfn
, NULL
,
639 offline_isolated_pages_cb
);
643 * Check all pages in range, recoreded as memory resource, are isolated.
646 check_pages_isolated_cb(unsigned long start_pfn
, unsigned long nr_pages
,
650 long offlined
= *(long *)data
;
651 ret
= test_pages_isolated(start_pfn
, start_pfn
+ nr_pages
);
654 *(long *)data
+= offlined
;
659 check_pages_isolated(unsigned long start_pfn
, unsigned long end_pfn
)
664 ret
= walk_memory_resource(start_pfn
, end_pfn
- start_pfn
, &offlined
,
665 check_pages_isolated_cb
);
667 offlined
= (long)ret
;
671 int offline_pages(unsigned long start_pfn
,
672 unsigned long end_pfn
, unsigned long timeout
)
674 unsigned long pfn
, nr_pages
, expire
;
676 int ret
, drain
, retry_max
, node
;
678 struct memory_notify arg
;
680 BUG_ON(start_pfn
>= end_pfn
);
681 /* at least, alignment against pageblock is necessary */
682 if (!IS_ALIGNED(start_pfn
, pageblock_nr_pages
))
684 if (!IS_ALIGNED(end_pfn
, pageblock_nr_pages
))
686 /* This makes hotplug much easier...and readable.
687 we assume this for now. .*/
688 if (!test_pages_in_a_zone(start_pfn
, end_pfn
))
691 zone
= page_zone(pfn_to_page(start_pfn
));
692 node
= zone_to_nid(zone
);
693 nr_pages
= end_pfn
- start_pfn
;
695 /* set above range as isolated */
696 ret
= start_isolate_page_range(start_pfn
, end_pfn
);
700 arg
.start_pfn
= start_pfn
;
701 arg
.nr_pages
= nr_pages
;
702 arg
.status_change_nid
= -1;
703 if (nr_pages
>= node_present_pages(node
))
704 arg
.status_change_nid
= node
;
706 ret
= memory_notify(MEM_GOING_OFFLINE
, &arg
);
707 ret
= notifier_to_errno(ret
);
712 expire
= jiffies
+ timeout
;
716 /* start memory hot removal */
718 if (time_after(jiffies
, expire
))
721 if (signal_pending(current
))
726 flush_scheduled_work();
731 pfn
= scan_lru_pages(start_pfn
, end_pfn
);
732 if (pfn
) { /* We have page on LRU */
733 ret
= do_migrate_range(pfn
, end_pfn
);
739 if (--retry_max
== 0)
746 /* drain all zone's lru pagevec, this is asyncronous... */
748 flush_scheduled_work();
750 /* drain pcp pages , this is synchrouns. */
753 offlined_pages
= check_pages_isolated(start_pfn
, end_pfn
);
754 if (offlined_pages
< 0) {
758 printk(KERN_INFO
"Offlined Pages %ld\n", offlined_pages
);
759 /* Ok, all of our target is islaoted.
760 We cannot do rollback at this point. */
761 offline_isolated_pages(start_pfn
, end_pfn
);
762 /* reset pagetype flags and makes migrate type to be MOVABLE */
763 undo_isolate_page_range(start_pfn
, end_pfn
);
764 /* removal success */
765 zone
->present_pages
-= offlined_pages
;
766 zone
->zone_pgdat
->node_present_pages
-= offlined_pages
;
767 totalram_pages
-= offlined_pages
;
768 num_physpages
-= offlined_pages
;
770 vm_total_pages
= nr_free_pagecache_pages();
771 writeback_set_ratelimit();
773 memory_notify(MEM_OFFLINE
, &arg
);
777 printk(KERN_INFO
"memory offlining %lx to %lx failed\n",
779 memory_notify(MEM_CANCEL_OFFLINE
, &arg
);
780 /* pushback to free area */
781 undo_isolate_page_range(start_pfn
, end_pfn
);
786 int remove_memory(u64 start
, u64 size
)
790 EXPORT_SYMBOL_GPL(remove_memory
);
791 #endif /* CONFIG_MEMORY_HOTREMOVE */