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>
32 /* add this memory to iomem resource */
33 static struct resource
*register_memory_resource(u64 start
, u64 size
)
36 res
= kzalloc(sizeof(struct resource
), GFP_KERNEL
);
39 res
->name
= "System RAM";
41 res
->end
= start
+ size
- 1;
42 res
->flags
= IORESOURCE_MEM
| IORESOURCE_BUSY
;
43 if (request_resource(&iomem_resource
, res
) < 0) {
44 printk("System RAM resource %llx - %llx cannot be added\n",
45 (unsigned long long)res
->start
, (unsigned long long)res
->end
);
52 static void release_memory_resource(struct resource
*res
)
56 release_resource(res
);
61 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
62 #ifndef CONFIG_SPARSEMEM_VMEMMAP
63 static void get_page_bootmem(unsigned long info
, struct page
*page
, int magic
)
65 atomic_set(&page
->_mapcount
, magic
);
67 set_page_private(page
, info
);
68 atomic_inc(&page
->_count
);
71 void put_page_bootmem(struct page
*page
)
75 magic
= atomic_read(&page
->_mapcount
);
78 if (atomic_dec_return(&page
->_count
) == 1) {
79 ClearPagePrivate(page
);
80 set_page_private(page
, 0);
81 reset_page_mapcount(page
);
82 __free_pages_bootmem(page
, 0);
87 void register_page_bootmem_info_section(unsigned long start_pfn
)
89 unsigned long *usemap
, mapsize
, section_nr
, i
;
90 struct mem_section
*ms
;
91 struct page
*page
, *memmap
;
93 if (!pfn_valid(start_pfn
))
96 section_nr
= pfn_to_section_nr(start_pfn
);
97 ms
= __nr_to_section(section_nr
);
99 /* Get section's memmap address */
100 memmap
= sparse_decode_mem_map(ms
->section_mem_map
, section_nr
);
103 * Get page for the memmap's phys address
104 * XXX: need more consideration for sparse_vmemmap...
106 page
= virt_to_page(memmap
);
107 mapsize
= sizeof(struct page
) * PAGES_PER_SECTION
;
108 mapsize
= PAGE_ALIGN(mapsize
) >> PAGE_SHIFT
;
110 /* remember memmap's page */
111 for (i
= 0; i
< mapsize
; i
++, page
++)
112 get_page_bootmem(section_nr
, page
, SECTION_INFO
);
114 usemap
= __nr_to_section(section_nr
)->pageblock_flags
;
115 page
= virt_to_page(usemap
);
117 mapsize
= PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT
;
119 for (i
= 0; i
< mapsize
; i
++, page
++)
120 get_page_bootmem(section_nr
, page
, MIX_INFO
);
124 void register_page_bootmem_info_node(struct pglist_data
*pgdat
)
126 unsigned long i
, pfn
, end_pfn
, nr_pages
;
127 int node
= pgdat
->node_id
;
131 nr_pages
= PAGE_ALIGN(sizeof(struct pglist_data
)) >> PAGE_SHIFT
;
132 page
= virt_to_page(pgdat
);
134 for (i
= 0; i
< nr_pages
; i
++, page
++)
135 get_page_bootmem(node
, page
, NODE_INFO
);
137 zone
= &pgdat
->node_zones
[0];
138 for (; zone
< pgdat
->node_zones
+ MAX_NR_ZONES
- 1; zone
++) {
139 if (zone
->wait_table
) {
140 nr_pages
= zone
->wait_table_hash_nr_entries
141 * sizeof(wait_queue_head_t
);
142 nr_pages
= PAGE_ALIGN(nr_pages
) >> PAGE_SHIFT
;
143 page
= virt_to_page(zone
->wait_table
);
145 for (i
= 0; i
< nr_pages
; i
++, page
++)
146 get_page_bootmem(node
, page
, NODE_INFO
);
150 pfn
= pgdat
->node_start_pfn
;
151 end_pfn
= pfn
+ pgdat
->node_spanned_pages
;
153 /* register_section info */
154 for (; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
)
155 register_page_bootmem_info_section(pfn
);
158 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
160 static int __add_zone(struct zone
*zone
, unsigned long phys_start_pfn
)
162 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
163 int nr_pages
= PAGES_PER_SECTION
;
164 int nid
= pgdat
->node_id
;
167 zone_type
= zone
- pgdat
->node_zones
;
168 if (!zone
->wait_table
) {
170 ret
= init_currently_empty_zone(zone
, phys_start_pfn
,
171 nr_pages
, MEMMAP_HOTPLUG
);
175 memmap_init_zone(nr_pages
, nid
, zone_type
,
176 phys_start_pfn
, MEMMAP_HOTPLUG
);
180 static int __add_section(struct zone
*zone
, unsigned long phys_start_pfn
)
182 int nr_pages
= PAGES_PER_SECTION
;
185 if (pfn_valid(phys_start_pfn
))
188 ret
= sparse_add_one_section(zone
, phys_start_pfn
, nr_pages
);
193 ret
= __add_zone(zone
, phys_start_pfn
);
198 return register_new_memory(__pfn_to_section(phys_start_pfn
));
201 #ifdef CONFIG_SPARSEMEM_VMEMMAP
202 static int __remove_section(struct zone
*zone
, struct mem_section
*ms
)
205 * XXX: Freeing memmap with vmemmap is not implement yet.
206 * This should be removed later.
211 static int __remove_section(struct zone
*zone
, struct mem_section
*ms
)
214 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
217 if (!valid_section(ms
))
220 ret
= unregister_memory_section(ms
);
224 pgdat_resize_lock(pgdat
, &flags
);
225 sparse_remove_one_section(zone
, ms
);
226 pgdat_resize_unlock(pgdat
, &flags
);
232 * Reasonably generic function for adding memory. It is
233 * expected that archs that support memory hotplug will
234 * call this function after deciding the zone to which to
237 int __add_pages(struct zone
*zone
, unsigned long phys_start_pfn
,
238 unsigned long nr_pages
)
242 int start_sec
, end_sec
;
243 /* during initialize mem_map, align hot-added range to section */
244 start_sec
= pfn_to_section_nr(phys_start_pfn
);
245 end_sec
= pfn_to_section_nr(phys_start_pfn
+ nr_pages
- 1);
247 for (i
= start_sec
; i
<= end_sec
; i
++) {
248 err
= __add_section(zone
, i
<< PFN_SECTION_SHIFT
);
251 * EEXIST is finally dealt with by ioresource collision
252 * check. see add_memory() => register_memory_resource()
253 * Warning will be printed if there is collision.
255 if (err
&& (err
!= -EEXIST
))
262 EXPORT_SYMBOL_GPL(__add_pages
);
265 * __remove_pages() - remove sections of pages from a zone
266 * @zone: zone from which pages need to be removed
267 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
268 * @nr_pages: number of pages to remove (must be multiple of section size)
270 * Generic helper function to remove section mappings and sysfs entries
271 * for the section of the memory we are removing. Caller needs to make
272 * sure that pages are marked reserved and zones are adjust properly by
273 * calling offline_pages().
275 int __remove_pages(struct zone
*zone
, unsigned long phys_start_pfn
,
276 unsigned long nr_pages
)
278 unsigned long i
, ret
= 0;
279 int sections_to_remove
;
282 * We can only remove entire sections
284 BUG_ON(phys_start_pfn
& ~PAGE_SECTION_MASK
);
285 BUG_ON(nr_pages
% PAGES_PER_SECTION
);
287 release_mem_region(phys_start_pfn
<< PAGE_SHIFT
, nr_pages
* PAGE_SIZE
);
289 sections_to_remove
= nr_pages
/ PAGES_PER_SECTION
;
290 for (i
= 0; i
< sections_to_remove
; i
++) {
291 unsigned long pfn
= phys_start_pfn
+ i
*PAGES_PER_SECTION
;
292 ret
= __remove_section(zone
, __pfn_to_section(pfn
));
298 EXPORT_SYMBOL_GPL(__remove_pages
);
300 static void grow_zone_span(struct zone
*zone
,
301 unsigned long start_pfn
, unsigned long end_pfn
)
303 unsigned long old_zone_end_pfn
;
305 zone_span_writelock(zone
);
307 old_zone_end_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
308 if (start_pfn
< zone
->zone_start_pfn
)
309 zone
->zone_start_pfn
= start_pfn
;
311 zone
->spanned_pages
= max(old_zone_end_pfn
, end_pfn
) -
312 zone
->zone_start_pfn
;
314 zone_span_writeunlock(zone
);
317 static void grow_pgdat_span(struct pglist_data
*pgdat
,
318 unsigned long start_pfn
, unsigned long end_pfn
)
320 unsigned long old_pgdat_end_pfn
=
321 pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
323 if (start_pfn
< pgdat
->node_start_pfn
)
324 pgdat
->node_start_pfn
= start_pfn
;
326 pgdat
->node_spanned_pages
= max(old_pgdat_end_pfn
, end_pfn
) -
327 pgdat
->node_start_pfn
;
330 void online_page(struct page
*page
)
335 #ifdef CONFIG_HIGHMEM
336 if (PageHighMem(page
))
340 #ifdef CONFIG_FLATMEM
341 max_mapnr
= max(page_to_pfn(page
), max_mapnr
);
344 ClearPageReserved(page
);
345 init_page_count(page
);
349 static int online_pages_range(unsigned long start_pfn
, unsigned long nr_pages
,
353 unsigned long onlined_pages
= *(unsigned long *)arg
;
355 if (PageReserved(pfn_to_page(start_pfn
)))
356 for (i
= 0; i
< nr_pages
; i
++) {
357 page
= pfn_to_page(start_pfn
+ i
);
361 *(unsigned long *)arg
= onlined_pages
;
366 int online_pages(unsigned long pfn
, unsigned long nr_pages
)
369 unsigned long onlined_pages
= 0;
371 int need_zonelists_rebuild
= 0;
374 struct memory_notify arg
;
377 arg
.nr_pages
= nr_pages
;
378 arg
.status_change_nid
= -1;
380 nid
= page_to_nid(pfn_to_page(pfn
));
381 if (node_present_pages(nid
) == 0)
382 arg
.status_change_nid
= nid
;
384 ret
= memory_notify(MEM_GOING_ONLINE
, &arg
);
385 ret
= notifier_to_errno(ret
);
387 memory_notify(MEM_CANCEL_ONLINE
, &arg
);
391 * This doesn't need a lock to do pfn_to_page().
392 * The section can't be removed here because of the
393 * memory_block->state_mutex.
395 zone
= page_zone(pfn_to_page(pfn
));
396 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
397 grow_zone_span(zone
, pfn
, pfn
+ nr_pages
);
398 grow_pgdat_span(zone
->zone_pgdat
, pfn
, pfn
+ nr_pages
);
399 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
402 * If this zone is not populated, then it is not in zonelist.
403 * This means the page allocator ignores this zone.
404 * So, zonelist must be updated after online.
406 if (!populated_zone(zone
))
407 need_zonelists_rebuild
= 1;
409 walk_memory_resource(pfn
, nr_pages
, &onlined_pages
,
411 zone
->present_pages
+= onlined_pages
;
412 zone
->zone_pgdat
->node_present_pages
+= onlined_pages
;
414 setup_per_zone_pages_min();
416 kswapd_run(zone_to_nid(zone
));
417 node_set_state(zone_to_nid(zone
), N_HIGH_MEMORY
);
420 if (need_zonelists_rebuild
)
421 build_all_zonelists();
422 vm_total_pages
= nr_free_pagecache_pages();
423 writeback_set_ratelimit();
426 memory_notify(MEM_ONLINE
, &arg
);
430 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
432 static pg_data_t
*hotadd_new_pgdat(int nid
, u64 start
)
434 struct pglist_data
*pgdat
;
435 unsigned long zones_size
[MAX_NR_ZONES
] = {0};
436 unsigned long zholes_size
[MAX_NR_ZONES
] = {0};
437 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
439 pgdat
= arch_alloc_nodedata(nid
);
443 arch_refresh_nodedata(nid
, pgdat
);
445 /* we can use NODE_DATA(nid) from here */
447 /* init node's zones as empty zones, we don't have any present pages.*/
448 free_area_init_node(nid
, pgdat
, zones_size
, start_pfn
, zholes_size
);
453 static void rollback_node_hotadd(int nid
, pg_data_t
*pgdat
)
455 arch_refresh_nodedata(nid
, NULL
);
456 arch_free_nodedata(pgdat
);
461 int add_memory(int nid
, u64 start
, u64 size
)
463 pg_data_t
*pgdat
= NULL
;
465 struct resource
*res
;
468 res
= register_memory_resource(start
, size
);
472 if (!node_online(nid
)) {
473 pgdat
= hotadd_new_pgdat(nid
, start
);
479 /* call arch's memory hotadd */
480 ret
= arch_add_memory(nid
, start
, size
);
485 /* we online node here. we can't roll back from here. */
486 node_set_online(nid
);
488 cpuset_track_online_nodes();
491 ret
= register_one_node(nid
);
493 * If sysfs file of new node can't create, cpu on the node
494 * can't be hot-added. There is no rollback way now.
495 * So, check by BUG_ON() to catch it reluctantly..
502 /* rollback pgdat allocation and others */
504 rollback_node_hotadd(nid
, pgdat
);
506 release_memory_resource(res
);
510 EXPORT_SYMBOL_GPL(add_memory
);
512 #ifdef CONFIG_MEMORY_HOTREMOVE
514 * Confirm all pages in a range [start, end) is belongs to the same zone.
516 static int test_pages_in_a_zone(unsigned long start_pfn
, unsigned long end_pfn
)
519 struct zone
*zone
= NULL
;
522 for (pfn
= start_pfn
;
524 pfn
+= MAX_ORDER_NR_PAGES
) {
526 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
527 while ((i
< MAX_ORDER_NR_PAGES
) && !pfn_valid_within(pfn
+ i
))
529 if (i
== MAX_ORDER_NR_PAGES
)
531 page
= pfn_to_page(pfn
+ i
);
532 if (zone
&& page_zone(page
) != zone
)
534 zone
= page_zone(page
);
540 * Scanning pfn is much easier than scanning lru list.
541 * Scan pfn from start to end and Find LRU page.
543 int scan_lru_pages(unsigned long start
, unsigned long end
)
547 for (pfn
= start
; pfn
< end
; pfn
++) {
548 if (pfn_valid(pfn
)) {
549 page
= pfn_to_page(pfn
);
558 hotremove_migrate_alloc(struct page
*page
,
559 unsigned long private,
562 /* This should be improoooooved!! */
563 return alloc_page(GFP_HIGHUSER_PAGECACHE
);
567 #define NR_OFFLINE_AT_ONCE_PAGES (256)
569 do_migrate_range(unsigned long start_pfn
, unsigned long end_pfn
)
573 int move_pages
= NR_OFFLINE_AT_ONCE_PAGES
;
578 for (pfn
= start_pfn
; pfn
< end_pfn
&& move_pages
> 0; pfn
++) {
581 page
= pfn_to_page(pfn
);
582 if (!page_count(page
))
585 * We can skip free pages. And we can only deal with pages on
588 ret
= isolate_lru_page(page
, &source
);
589 if (!ret
) { /* Success */
592 /* Becasue we don't have big zone->lock. we should
593 check this again here. */
594 if (page_count(page
))
596 #ifdef CONFIG_DEBUG_VM
597 printk(KERN_INFO
"removing from LRU failed"
599 pfn
, page_count(page
), page
->flags
);
605 if (!list_empty(&source
))
606 putback_lru_pages(&source
);
610 if (list_empty(&source
))
612 /* this function returns # of failed pages */
613 ret
= migrate_pages(&source
, hotremove_migrate_alloc
, 0);
620 * remove from free_area[] and mark all as Reserved.
623 offline_isolated_pages_cb(unsigned long start
, unsigned long nr_pages
,
626 __offline_isolated_pages(start
, start
+ nr_pages
);
631 offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
633 walk_memory_resource(start_pfn
, end_pfn
- start_pfn
, NULL
,
634 offline_isolated_pages_cb
);
638 * Check all pages in range, recoreded as memory resource, are isolated.
641 check_pages_isolated_cb(unsigned long start_pfn
, unsigned long nr_pages
,
645 long offlined
= *(long *)data
;
646 ret
= test_pages_isolated(start_pfn
, start_pfn
+ nr_pages
);
649 *(long *)data
+= offlined
;
654 check_pages_isolated(unsigned long start_pfn
, unsigned long end_pfn
)
659 ret
= walk_memory_resource(start_pfn
, end_pfn
- start_pfn
, &offlined
,
660 check_pages_isolated_cb
);
662 offlined
= (long)ret
;
666 int offline_pages(unsigned long start_pfn
,
667 unsigned long end_pfn
, unsigned long timeout
)
669 unsigned long pfn
, nr_pages
, expire
;
671 int ret
, drain
, retry_max
, node
;
673 struct memory_notify arg
;
675 BUG_ON(start_pfn
>= end_pfn
);
676 /* at least, alignment against pageblock is necessary */
677 if (!IS_ALIGNED(start_pfn
, pageblock_nr_pages
))
679 if (!IS_ALIGNED(end_pfn
, pageblock_nr_pages
))
681 /* This makes hotplug much easier...and readable.
682 we assume this for now. .*/
683 if (!test_pages_in_a_zone(start_pfn
, end_pfn
))
686 zone
= page_zone(pfn_to_page(start_pfn
));
687 node
= zone_to_nid(zone
);
688 nr_pages
= end_pfn
- start_pfn
;
690 /* set above range as isolated */
691 ret
= start_isolate_page_range(start_pfn
, end_pfn
);
695 arg
.start_pfn
= start_pfn
;
696 arg
.nr_pages
= nr_pages
;
697 arg
.status_change_nid
= -1;
698 if (nr_pages
>= node_present_pages(node
))
699 arg
.status_change_nid
= node
;
701 ret
= memory_notify(MEM_GOING_OFFLINE
, &arg
);
702 ret
= notifier_to_errno(ret
);
707 expire
= jiffies
+ timeout
;
711 /* start memory hot removal */
713 if (time_after(jiffies
, expire
))
716 if (signal_pending(current
))
721 flush_scheduled_work();
726 pfn
= scan_lru_pages(start_pfn
, end_pfn
);
727 if (pfn
) { /* We have page on LRU */
728 ret
= do_migrate_range(pfn
, end_pfn
);
734 if (--retry_max
== 0)
741 /* drain all zone's lru pagevec, this is asyncronous... */
743 flush_scheduled_work();
745 /* drain pcp pages , this is synchrouns. */
748 offlined_pages
= check_pages_isolated(start_pfn
, end_pfn
);
749 if (offlined_pages
< 0) {
753 printk(KERN_INFO
"Offlined Pages %ld\n", offlined_pages
);
754 /* Ok, all of our target is islaoted.
755 We cannot do rollback at this point. */
756 offline_isolated_pages(start_pfn
, end_pfn
);
757 /* reset pagetype flags and makes migrate type to be MOVABLE */
758 undo_isolate_page_range(start_pfn
, end_pfn
);
759 /* removal success */
760 zone
->present_pages
-= offlined_pages
;
761 zone
->zone_pgdat
->node_present_pages
-= offlined_pages
;
762 totalram_pages
-= offlined_pages
;
763 num_physpages
-= offlined_pages
;
765 vm_total_pages
= nr_free_pagecache_pages();
766 writeback_set_ratelimit();
768 memory_notify(MEM_OFFLINE
, &arg
);
772 printk(KERN_INFO
"memory offlining %lx to %lx failed\n",
774 memory_notify(MEM_CANCEL_OFFLINE
, &arg
);
775 /* pushback to free area */
776 undo_isolate_page_range(start_pfn
, end_pfn
);
781 int remove_memory(u64 start
, u64 size
)
785 EXPORT_SYMBOL_GPL(remove_memory
);
786 #endif /* CONFIG_MEMORY_HOTREMOVE */