2 * linux/mm/memory_hotplug.c
7 #include <linux/stddef.h>
9 #include <linux/sched/signal.h>
10 #include <linux/swap.h>
11 #include <linux/interrupt.h>
12 #include <linux/pagemap.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/memremap.h>
22 #include <linux/memory_hotplug.h>
23 #include <linux/highmem.h>
24 #include <linux/vmalloc.h>
25 #include <linux/ioport.h>
26 #include <linux/delay.h>
27 #include <linux/migrate.h>
28 #include <linux/page-isolation.h>
29 #include <linux/pfn.h>
30 #include <linux/suspend.h>
31 #include <linux/mm_inline.h>
32 #include <linux/firmware-map.h>
33 #include <linux/stop_machine.h>
34 #include <linux/hugetlb.h>
35 #include <linux/memblock.h>
36 #include <linux/bootmem.h>
37 #include <linux/compaction.h>
39 #include <asm/tlbflush.h>
44 * online_page_callback contains pointer to current page onlining function.
45 * Initially it is generic_online_page(). If it is required it could be
46 * changed by calling set_online_page_callback() for callback registration
47 * and restore_online_page_callback() for generic callback restore.
50 static void generic_online_page(struct page
*page
);
52 static online_page_callback_t online_page_callback
= generic_online_page
;
53 static DEFINE_MUTEX(online_page_callback_lock
);
55 /* The same as the cpu_hotplug lock, but for memory hotplug. */
57 struct task_struct
*active_writer
;
58 struct mutex lock
; /* Synchronizes accesses to refcount, */
60 * Also blocks the new readers during
61 * an ongoing mem hotplug operation.
65 #ifdef CONFIG_DEBUG_LOCK_ALLOC
66 struct lockdep_map dep_map
;
69 .active_writer
= NULL
,
70 .lock
= __MUTEX_INITIALIZER(mem_hotplug
.lock
),
72 #ifdef CONFIG_DEBUG_LOCK_ALLOC
73 .dep_map
= {.name
= "mem_hotplug.lock" },
77 /* Lockdep annotations for get/put_online_mems() and mem_hotplug_begin/end() */
78 #define memhp_lock_acquire_read() lock_map_acquire_read(&mem_hotplug.dep_map)
79 #define memhp_lock_acquire() lock_map_acquire(&mem_hotplug.dep_map)
80 #define memhp_lock_release() lock_map_release(&mem_hotplug.dep_map)
82 bool movable_node_enabled
= false;
84 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
85 bool memhp_auto_online
;
87 bool memhp_auto_online
= true;
89 EXPORT_SYMBOL_GPL(memhp_auto_online
);
91 static int __init
setup_memhp_default_state(char *str
)
93 if (!strcmp(str
, "online"))
94 memhp_auto_online
= true;
95 else if (!strcmp(str
, "offline"))
96 memhp_auto_online
= false;
100 __setup("memhp_default_state=", setup_memhp_default_state
);
102 void get_online_mems(void)
105 if (mem_hotplug
.active_writer
== current
)
107 memhp_lock_acquire_read();
108 mutex_lock(&mem_hotplug
.lock
);
109 mem_hotplug
.refcount
++;
110 mutex_unlock(&mem_hotplug
.lock
);
114 void put_online_mems(void)
116 if (mem_hotplug
.active_writer
== current
)
118 mutex_lock(&mem_hotplug
.lock
);
120 if (WARN_ON(!mem_hotplug
.refcount
))
121 mem_hotplug
.refcount
++; /* try to fix things up */
123 if (!--mem_hotplug
.refcount
&& unlikely(mem_hotplug
.active_writer
))
124 wake_up_process(mem_hotplug
.active_writer
);
125 mutex_unlock(&mem_hotplug
.lock
);
126 memhp_lock_release();
130 /* Serializes write accesses to mem_hotplug.active_writer. */
131 static DEFINE_MUTEX(memory_add_remove_lock
);
133 void mem_hotplug_begin(void)
135 mutex_lock(&memory_add_remove_lock
);
137 mem_hotplug
.active_writer
= current
;
139 memhp_lock_acquire();
141 mutex_lock(&mem_hotplug
.lock
);
142 if (likely(!mem_hotplug
.refcount
))
144 __set_current_state(TASK_UNINTERRUPTIBLE
);
145 mutex_unlock(&mem_hotplug
.lock
);
150 void mem_hotplug_done(void)
152 mem_hotplug
.active_writer
= NULL
;
153 mutex_unlock(&mem_hotplug
.lock
);
154 memhp_lock_release();
155 mutex_unlock(&memory_add_remove_lock
);
158 /* add this memory to iomem resource */
159 static struct resource
*register_memory_resource(u64 start
, u64 size
)
161 struct resource
*res
;
162 res
= kzalloc(sizeof(struct resource
), GFP_KERNEL
);
164 return ERR_PTR(-ENOMEM
);
166 res
->name
= "System RAM";
168 res
->end
= start
+ size
- 1;
169 res
->flags
= IORESOURCE_SYSTEM_RAM
| IORESOURCE_BUSY
;
170 if (request_resource(&iomem_resource
, res
) < 0) {
171 pr_debug("System RAM resource %pR cannot be added\n", res
);
173 return ERR_PTR(-EEXIST
);
178 static void release_memory_resource(struct resource
*res
)
182 release_resource(res
);
187 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
188 void get_page_bootmem(unsigned long info
, struct page
*page
,
191 page
->freelist
= (void *)type
;
192 SetPagePrivate(page
);
193 set_page_private(page
, info
);
197 void put_page_bootmem(struct page
*page
)
201 type
= (unsigned long) page
->freelist
;
202 BUG_ON(type
< MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE
||
203 type
> MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE
);
205 if (page_ref_dec_return(page
) == 1) {
206 page
->freelist
= NULL
;
207 ClearPagePrivate(page
);
208 set_page_private(page
, 0);
209 INIT_LIST_HEAD(&page
->lru
);
210 free_reserved_page(page
);
214 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
215 #ifndef CONFIG_SPARSEMEM_VMEMMAP
216 static void register_page_bootmem_info_section(unsigned long start_pfn
)
218 unsigned long *usemap
, mapsize
, section_nr
, i
;
219 struct mem_section
*ms
;
220 struct page
*page
, *memmap
;
222 section_nr
= pfn_to_section_nr(start_pfn
);
223 ms
= __nr_to_section(section_nr
);
225 /* Get section's memmap address */
226 memmap
= sparse_decode_mem_map(ms
->section_mem_map
, section_nr
);
229 * Get page for the memmap's phys address
230 * XXX: need more consideration for sparse_vmemmap...
232 page
= virt_to_page(memmap
);
233 mapsize
= sizeof(struct page
) * PAGES_PER_SECTION
;
234 mapsize
= PAGE_ALIGN(mapsize
) >> PAGE_SHIFT
;
236 /* remember memmap's page */
237 for (i
= 0; i
< mapsize
; i
++, page
++)
238 get_page_bootmem(section_nr
, page
, SECTION_INFO
);
240 usemap
= __nr_to_section(section_nr
)->pageblock_flags
;
241 page
= virt_to_page(usemap
);
243 mapsize
= PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT
;
245 for (i
= 0; i
< mapsize
; i
++, page
++)
246 get_page_bootmem(section_nr
, page
, MIX_SECTION_INFO
);
249 #else /* CONFIG_SPARSEMEM_VMEMMAP */
250 static void register_page_bootmem_info_section(unsigned long start_pfn
)
252 unsigned long *usemap
, mapsize
, section_nr
, i
;
253 struct mem_section
*ms
;
254 struct page
*page
, *memmap
;
256 if (!pfn_valid(start_pfn
))
259 section_nr
= pfn_to_section_nr(start_pfn
);
260 ms
= __nr_to_section(section_nr
);
262 memmap
= sparse_decode_mem_map(ms
->section_mem_map
, section_nr
);
264 register_page_bootmem_memmap(section_nr
, memmap
, PAGES_PER_SECTION
);
266 usemap
= __nr_to_section(section_nr
)->pageblock_flags
;
267 page
= virt_to_page(usemap
);
269 mapsize
= PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT
;
271 for (i
= 0; i
< mapsize
; i
++, page
++)
272 get_page_bootmem(section_nr
, page
, MIX_SECTION_INFO
);
274 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
276 void __init
register_page_bootmem_info_node(struct pglist_data
*pgdat
)
278 unsigned long i
, pfn
, end_pfn
, nr_pages
;
279 int node
= pgdat
->node_id
;
282 nr_pages
= PAGE_ALIGN(sizeof(struct pglist_data
)) >> PAGE_SHIFT
;
283 page
= virt_to_page(pgdat
);
285 for (i
= 0; i
< nr_pages
; i
++, page
++)
286 get_page_bootmem(node
, page
, NODE_INFO
);
288 pfn
= pgdat
->node_start_pfn
;
289 end_pfn
= pgdat_end_pfn(pgdat
);
291 /* register section info */
292 for (; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
294 * Some platforms can assign the same pfn to multiple nodes - on
295 * node0 as well as nodeN. To avoid registering a pfn against
296 * multiple nodes we check that this pfn does not already
297 * reside in some other nodes.
299 if (pfn_valid(pfn
) && (early_pfn_to_nid(pfn
) == node
))
300 register_page_bootmem_info_section(pfn
);
303 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
305 static int __meminit
__add_section(int nid
, unsigned long phys_start_pfn
,
311 if (pfn_valid(phys_start_pfn
))
314 ret
= sparse_add_one_section(NODE_DATA(nid
), phys_start_pfn
);
319 * Make all the pages reserved so that nobody will stumble over half
321 * FIXME: We also have to associate it with a node because pfn_to_node
322 * relies on having page with the proper node.
324 for (i
= 0; i
< PAGES_PER_SECTION
; i
++) {
325 unsigned long pfn
= phys_start_pfn
+ i
;
330 page
= pfn_to_page(pfn
);
331 set_page_node(page
, nid
);
332 SetPageReserved(page
);
338 return register_new_memory(nid
, __pfn_to_section(phys_start_pfn
));
342 * Reasonably generic function for adding memory. It is
343 * expected that archs that support memory hotplug will
344 * call this function after deciding the zone to which to
347 int __ref
__add_pages(int nid
, unsigned long phys_start_pfn
,
348 unsigned long nr_pages
, bool want_memblock
)
352 int start_sec
, end_sec
;
353 struct vmem_altmap
*altmap
;
355 /* during initialize mem_map, align hot-added range to section */
356 start_sec
= pfn_to_section_nr(phys_start_pfn
);
357 end_sec
= pfn_to_section_nr(phys_start_pfn
+ nr_pages
- 1);
359 altmap
= to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn
));
362 * Validate altmap is within bounds of the total request
364 if (altmap
->base_pfn
!= phys_start_pfn
365 || vmem_altmap_offset(altmap
) > nr_pages
) {
366 pr_warn_once("memory add fail, invalid altmap\n");
373 for (i
= start_sec
; i
<= end_sec
; i
++) {
374 err
= __add_section(nid
, section_nr_to_pfn(i
), want_memblock
);
377 * EEXIST is finally dealt with by ioresource collision
378 * check. see add_memory() => register_memory_resource()
379 * Warning will be printed if there is collision.
381 if (err
&& (err
!= -EEXIST
))
385 vmemmap_populate_print_last();
389 EXPORT_SYMBOL_GPL(__add_pages
);
391 #ifdef CONFIG_MEMORY_HOTREMOVE
392 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
393 static int find_smallest_section_pfn(int nid
, struct zone
*zone
,
394 unsigned long start_pfn
,
395 unsigned long end_pfn
)
397 struct mem_section
*ms
;
399 for (; start_pfn
< end_pfn
; start_pfn
+= PAGES_PER_SECTION
) {
400 ms
= __pfn_to_section(start_pfn
);
402 if (unlikely(!valid_section(ms
)))
405 if (unlikely(pfn_to_nid(start_pfn
) != nid
))
408 if (zone
&& zone
!= page_zone(pfn_to_page(start_pfn
)))
417 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
418 static int find_biggest_section_pfn(int nid
, struct zone
*zone
,
419 unsigned long start_pfn
,
420 unsigned long end_pfn
)
422 struct mem_section
*ms
;
425 /* pfn is the end pfn of a memory section. */
427 for (; pfn
>= start_pfn
; pfn
-= PAGES_PER_SECTION
) {
428 ms
= __pfn_to_section(pfn
);
430 if (unlikely(!valid_section(ms
)))
433 if (unlikely(pfn_to_nid(pfn
) != nid
))
436 if (zone
&& zone
!= page_zone(pfn_to_page(pfn
)))
445 static void shrink_zone_span(struct zone
*zone
, unsigned long start_pfn
,
446 unsigned long end_pfn
)
448 unsigned long zone_start_pfn
= zone
->zone_start_pfn
;
449 unsigned long z
= zone_end_pfn(zone
); /* zone_end_pfn namespace clash */
450 unsigned long zone_end_pfn
= z
;
452 struct mem_section
*ms
;
453 int nid
= zone_to_nid(zone
);
455 zone_span_writelock(zone
);
456 if (zone_start_pfn
== start_pfn
) {
458 * If the section is smallest section in the zone, it need
459 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
460 * In this case, we find second smallest valid mem_section
461 * for shrinking zone.
463 pfn
= find_smallest_section_pfn(nid
, zone
, end_pfn
,
466 zone
->zone_start_pfn
= pfn
;
467 zone
->spanned_pages
= zone_end_pfn
- pfn
;
469 } else if (zone_end_pfn
== end_pfn
) {
471 * If the section is biggest section in the zone, it need
472 * shrink zone->spanned_pages.
473 * In this case, we find second biggest valid mem_section for
476 pfn
= find_biggest_section_pfn(nid
, zone
, zone_start_pfn
,
479 zone
->spanned_pages
= pfn
- zone_start_pfn
+ 1;
483 * The section is not biggest or smallest mem_section in the zone, it
484 * only creates a hole in the zone. So in this case, we need not
485 * change the zone. But perhaps, the zone has only hole data. Thus
486 * it check the zone has only hole or not.
488 pfn
= zone_start_pfn
;
489 for (; pfn
< zone_end_pfn
; pfn
+= PAGES_PER_SECTION
) {
490 ms
= __pfn_to_section(pfn
);
492 if (unlikely(!valid_section(ms
)))
495 if (page_zone(pfn_to_page(pfn
)) != zone
)
498 /* If the section is current section, it continues the loop */
499 if (start_pfn
== pfn
)
502 /* If we find valid section, we have nothing to do */
503 zone_span_writeunlock(zone
);
507 /* The zone has no valid section */
508 zone
->zone_start_pfn
= 0;
509 zone
->spanned_pages
= 0;
510 zone_span_writeunlock(zone
);
513 static void shrink_pgdat_span(struct pglist_data
*pgdat
,
514 unsigned long start_pfn
, unsigned long end_pfn
)
516 unsigned long pgdat_start_pfn
= pgdat
->node_start_pfn
;
517 unsigned long p
= pgdat_end_pfn(pgdat
); /* pgdat_end_pfn namespace clash */
518 unsigned long pgdat_end_pfn
= p
;
520 struct mem_section
*ms
;
521 int nid
= pgdat
->node_id
;
523 if (pgdat_start_pfn
== start_pfn
) {
525 * If the section is smallest section in the pgdat, it need
526 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
527 * In this case, we find second smallest valid mem_section
528 * for shrinking zone.
530 pfn
= find_smallest_section_pfn(nid
, NULL
, end_pfn
,
533 pgdat
->node_start_pfn
= pfn
;
534 pgdat
->node_spanned_pages
= pgdat_end_pfn
- pfn
;
536 } else if (pgdat_end_pfn
== end_pfn
) {
538 * If the section is biggest section in the pgdat, it need
539 * shrink pgdat->node_spanned_pages.
540 * In this case, we find second biggest valid mem_section for
543 pfn
= find_biggest_section_pfn(nid
, NULL
, pgdat_start_pfn
,
546 pgdat
->node_spanned_pages
= pfn
- pgdat_start_pfn
+ 1;
550 * If the section is not biggest or smallest mem_section in the pgdat,
551 * it only creates a hole in the pgdat. So in this case, we need not
553 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
554 * has only hole or not.
556 pfn
= pgdat_start_pfn
;
557 for (; pfn
< pgdat_end_pfn
; pfn
+= PAGES_PER_SECTION
) {
558 ms
= __pfn_to_section(pfn
);
560 if (unlikely(!valid_section(ms
)))
563 if (pfn_to_nid(pfn
) != nid
)
566 /* If the section is current section, it continues the loop */
567 if (start_pfn
== pfn
)
570 /* If we find valid section, we have nothing to do */
574 /* The pgdat has no valid section */
575 pgdat
->node_start_pfn
= 0;
576 pgdat
->node_spanned_pages
= 0;
579 static void __remove_zone(struct zone
*zone
, unsigned long start_pfn
)
581 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
582 int nr_pages
= PAGES_PER_SECTION
;
586 zone_type
= zone
- pgdat
->node_zones
;
588 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
589 shrink_zone_span(zone
, start_pfn
, start_pfn
+ nr_pages
);
590 shrink_pgdat_span(pgdat
, start_pfn
, start_pfn
+ nr_pages
);
591 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
594 static int __remove_section(struct zone
*zone
, struct mem_section
*ms
,
595 unsigned long map_offset
)
597 unsigned long start_pfn
;
601 if (!valid_section(ms
))
604 ret
= unregister_memory_section(ms
);
608 scn_nr
= __section_nr(ms
);
609 start_pfn
= section_nr_to_pfn(scn_nr
);
610 __remove_zone(zone
, start_pfn
);
612 sparse_remove_one_section(zone
, ms
, map_offset
);
617 * __remove_pages() - remove sections of pages from a zone
618 * @zone: zone from which pages need to be removed
619 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
620 * @nr_pages: number of pages to remove (must be multiple of section size)
622 * Generic helper function to remove section mappings and sysfs entries
623 * for the section of the memory we are removing. Caller needs to make
624 * sure that pages are marked reserved and zones are adjust properly by
625 * calling offline_pages().
627 int __remove_pages(struct zone
*zone
, unsigned long phys_start_pfn
,
628 unsigned long nr_pages
)
631 unsigned long map_offset
= 0;
632 int sections_to_remove
, ret
= 0;
634 /* In the ZONE_DEVICE case device driver owns the memory region */
635 if (is_dev_zone(zone
)) {
636 struct page
*page
= pfn_to_page(phys_start_pfn
);
637 struct vmem_altmap
*altmap
;
639 altmap
= to_vmem_altmap((unsigned long) page
);
641 map_offset
= vmem_altmap_offset(altmap
);
643 resource_size_t start
, size
;
645 start
= phys_start_pfn
<< PAGE_SHIFT
;
646 size
= nr_pages
* PAGE_SIZE
;
648 ret
= release_mem_region_adjustable(&iomem_resource
, start
,
651 resource_size_t endres
= start
+ size
- 1;
653 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
654 &start
, &endres
, ret
);
658 clear_zone_contiguous(zone
);
661 * We can only remove entire sections
663 BUG_ON(phys_start_pfn
& ~PAGE_SECTION_MASK
);
664 BUG_ON(nr_pages
% PAGES_PER_SECTION
);
666 sections_to_remove
= nr_pages
/ PAGES_PER_SECTION
;
667 for (i
= 0; i
< sections_to_remove
; i
++) {
668 unsigned long pfn
= phys_start_pfn
+ i
*PAGES_PER_SECTION
;
670 ret
= __remove_section(zone
, __pfn_to_section(pfn
), map_offset
);
676 set_zone_contiguous(zone
);
680 #endif /* CONFIG_MEMORY_HOTREMOVE */
682 int set_online_page_callback(online_page_callback_t callback
)
687 mutex_lock(&online_page_callback_lock
);
689 if (online_page_callback
== generic_online_page
) {
690 online_page_callback
= callback
;
694 mutex_unlock(&online_page_callback_lock
);
699 EXPORT_SYMBOL_GPL(set_online_page_callback
);
701 int restore_online_page_callback(online_page_callback_t callback
)
706 mutex_lock(&online_page_callback_lock
);
708 if (online_page_callback
== callback
) {
709 online_page_callback
= generic_online_page
;
713 mutex_unlock(&online_page_callback_lock
);
718 EXPORT_SYMBOL_GPL(restore_online_page_callback
);
720 void __online_page_set_limits(struct page
*page
)
723 EXPORT_SYMBOL_GPL(__online_page_set_limits
);
725 void __online_page_increment_counters(struct page
*page
)
727 adjust_managed_page_count(page
, 1);
729 EXPORT_SYMBOL_GPL(__online_page_increment_counters
);
731 void __online_page_free(struct page
*page
)
733 __free_reserved_page(page
);
735 EXPORT_SYMBOL_GPL(__online_page_free
);
737 static void generic_online_page(struct page
*page
)
739 __online_page_set_limits(page
);
740 __online_page_increment_counters(page
);
741 __online_page_free(page
);
744 static int online_pages_range(unsigned long start_pfn
, unsigned long nr_pages
,
748 unsigned long onlined_pages
= *(unsigned long *)arg
;
751 if (PageReserved(pfn_to_page(start_pfn
)))
752 for (i
= 0; i
< nr_pages
; i
++) {
753 page
= pfn_to_page(start_pfn
+ i
);
754 (*online_page_callback
)(page
);
758 online_mem_sections(start_pfn
, start_pfn
+ nr_pages
);
760 *(unsigned long *)arg
= onlined_pages
;
764 /* check which state of node_states will be changed when online memory */
765 static void node_states_check_changes_online(unsigned long nr_pages
,
766 struct zone
*zone
, struct memory_notify
*arg
)
768 int nid
= zone_to_nid(zone
);
769 enum zone_type zone_last
= ZONE_NORMAL
;
772 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
773 * contains nodes which have zones of 0...ZONE_NORMAL,
774 * set zone_last to ZONE_NORMAL.
776 * If we don't have HIGHMEM nor movable node,
777 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
778 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
780 if (N_MEMORY
== N_NORMAL_MEMORY
)
781 zone_last
= ZONE_MOVABLE
;
784 * if the memory to be online is in a zone of 0...zone_last, and
785 * the zones of 0...zone_last don't have memory before online, we will
786 * need to set the node to node_states[N_NORMAL_MEMORY] after
787 * the memory is online.
789 if (zone_idx(zone
) <= zone_last
&& !node_state(nid
, N_NORMAL_MEMORY
))
790 arg
->status_change_nid_normal
= nid
;
792 arg
->status_change_nid_normal
= -1;
794 #ifdef CONFIG_HIGHMEM
796 * If we have movable node, node_states[N_HIGH_MEMORY]
797 * contains nodes which have zones of 0...ZONE_HIGHMEM,
798 * set zone_last to ZONE_HIGHMEM.
800 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
801 * contains nodes which have zones of 0...ZONE_MOVABLE,
802 * set zone_last to ZONE_MOVABLE.
804 zone_last
= ZONE_HIGHMEM
;
805 if (N_MEMORY
== N_HIGH_MEMORY
)
806 zone_last
= ZONE_MOVABLE
;
808 if (zone_idx(zone
) <= zone_last
&& !node_state(nid
, N_HIGH_MEMORY
))
809 arg
->status_change_nid_high
= nid
;
811 arg
->status_change_nid_high
= -1;
813 arg
->status_change_nid_high
= arg
->status_change_nid_normal
;
817 * if the node don't have memory befor online, we will need to
818 * set the node to node_states[N_MEMORY] after the memory
821 if (!node_state(nid
, N_MEMORY
))
822 arg
->status_change_nid
= nid
;
824 arg
->status_change_nid
= -1;
827 static void node_states_set_node(int node
, struct memory_notify
*arg
)
829 if (arg
->status_change_nid_normal
>= 0)
830 node_set_state(node
, N_NORMAL_MEMORY
);
832 if (arg
->status_change_nid_high
>= 0)
833 node_set_state(node
, N_HIGH_MEMORY
);
835 node_set_state(node
, N_MEMORY
);
838 bool allow_online_pfn_range(int nid
, unsigned long pfn
, unsigned long nr_pages
, int online_type
)
840 struct pglist_data
*pgdat
= NODE_DATA(nid
);
841 struct zone
*movable_zone
= &pgdat
->node_zones
[ZONE_MOVABLE
];
842 struct zone
*default_zone
= default_zone_for_pfn(nid
, pfn
, nr_pages
);
845 * TODO there shouldn't be any inherent reason to have ZONE_NORMAL
846 * physically before ZONE_MOVABLE. All we need is they do not
847 * overlap. Historically we didn't allow ZONE_NORMAL after ZONE_MOVABLE
848 * though so let's stick with it for simplicity for now.
849 * TODO make sure we do not overlap with ZONE_DEVICE
851 if (online_type
== MMOP_ONLINE_KERNEL
) {
852 if (zone_is_empty(movable_zone
))
854 return movable_zone
->zone_start_pfn
>= pfn
+ nr_pages
;
855 } else if (online_type
== MMOP_ONLINE_MOVABLE
) {
856 return zone_end_pfn(default_zone
) <= pfn
;
859 /* MMOP_ONLINE_KEEP will always succeed and inherits the current zone */
860 return online_type
== MMOP_ONLINE_KEEP
;
863 static void __meminit
resize_zone_range(struct zone
*zone
, unsigned long start_pfn
,
864 unsigned long nr_pages
)
866 unsigned long old_end_pfn
= zone_end_pfn(zone
);
868 if (zone_is_empty(zone
) || start_pfn
< zone
->zone_start_pfn
)
869 zone
->zone_start_pfn
= start_pfn
;
871 zone
->spanned_pages
= max(start_pfn
+ nr_pages
, old_end_pfn
) - zone
->zone_start_pfn
;
874 static void __meminit
resize_pgdat_range(struct pglist_data
*pgdat
, unsigned long start_pfn
,
875 unsigned long nr_pages
)
877 unsigned long old_end_pfn
= pgdat_end_pfn(pgdat
);
879 if (!pgdat
->node_spanned_pages
|| start_pfn
< pgdat
->node_start_pfn
)
880 pgdat
->node_start_pfn
= start_pfn
;
882 pgdat
->node_spanned_pages
= max(start_pfn
+ nr_pages
, old_end_pfn
) - pgdat
->node_start_pfn
;
885 void __ref
move_pfn_range_to_zone(struct zone
*zone
,
886 unsigned long start_pfn
, unsigned long nr_pages
)
888 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
889 int nid
= pgdat
->node_id
;
892 if (zone_is_empty(zone
))
893 init_currently_empty_zone(zone
, start_pfn
, nr_pages
);
895 clear_zone_contiguous(zone
);
897 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
898 pgdat_resize_lock(pgdat
, &flags
);
899 zone_span_writelock(zone
);
900 resize_zone_range(zone
, start_pfn
, nr_pages
);
901 zone_span_writeunlock(zone
);
902 resize_pgdat_range(pgdat
, start_pfn
, nr_pages
);
903 pgdat_resize_unlock(pgdat
, &flags
);
906 * TODO now we have a visible range of pages which are not associated
907 * with their zone properly. Not nice but set_pfnblock_flags_mask
908 * expects the zone spans the pfn range. All the pages in the range
909 * are reserved so nobody should be touching them so we should be safe
911 memmap_init_zone(nr_pages
, nid
, zone_idx(zone
), start_pfn
, MEMMAP_HOTPLUG
);
913 set_zone_contiguous(zone
);
917 * Returns a default kernel memory zone for the given pfn range.
918 * If no kernel zone covers this pfn range it will automatically go
919 * to the ZONE_NORMAL.
921 struct zone
*default_zone_for_pfn(int nid
, unsigned long start_pfn
,
922 unsigned long nr_pages
)
924 struct pglist_data
*pgdat
= NODE_DATA(nid
);
927 for (zid
= 0; zid
<= ZONE_NORMAL
; zid
++) {
928 struct zone
*zone
= &pgdat
->node_zones
[zid
];
930 if (zone_intersects(zone
, start_pfn
, nr_pages
))
934 return &pgdat
->node_zones
[ZONE_NORMAL
];
937 static inline bool movable_pfn_range(int nid
, struct zone
*default_zone
,
938 unsigned long start_pfn
, unsigned long nr_pages
)
940 if (!allow_online_pfn_range(nid
, start_pfn
, nr_pages
,
944 if (!movable_node_is_enabled())
947 return !zone_intersects(default_zone
, start_pfn
, nr_pages
);
951 * Associates the given pfn range with the given node and the zone appropriate
952 * for the given online type.
954 static struct zone
* __meminit
move_pfn_range(int online_type
, int nid
,
955 unsigned long start_pfn
, unsigned long nr_pages
)
957 struct pglist_data
*pgdat
= NODE_DATA(nid
);
958 struct zone
*zone
= default_zone_for_pfn(nid
, start_pfn
, nr_pages
);
960 if (online_type
== MMOP_ONLINE_KEEP
) {
961 struct zone
*movable_zone
= &pgdat
->node_zones
[ZONE_MOVABLE
];
963 * MMOP_ONLINE_KEEP defaults to MMOP_ONLINE_KERNEL but use
964 * movable zone if that is not possible (e.g. we are within
965 * or past the existing movable zone). movable_node overrides
966 * this default and defaults to movable zone
968 if (movable_pfn_range(nid
, zone
, start_pfn
, nr_pages
))
970 } else if (online_type
== MMOP_ONLINE_MOVABLE
) {
971 zone
= &pgdat
->node_zones
[ZONE_MOVABLE
];
974 move_pfn_range_to_zone(zone
, start_pfn
, nr_pages
);
978 /* Must be protected by mem_hotplug_begin() */
979 int __ref
online_pages(unsigned long pfn
, unsigned long nr_pages
, int online_type
)
982 unsigned long onlined_pages
= 0;
984 int need_zonelists_rebuild
= 0;
987 struct memory_notify arg
;
989 nid
= pfn_to_nid(pfn
);
990 if (!allow_online_pfn_range(nid
, pfn
, nr_pages
, online_type
))
993 /* associate pfn range with the zone */
994 zone
= move_pfn_range(online_type
, nid
, pfn
, nr_pages
);
997 arg
.nr_pages
= nr_pages
;
998 node_states_check_changes_online(nr_pages
, zone
, &arg
);
1000 ret
= memory_notify(MEM_GOING_ONLINE
, &arg
);
1001 ret
= notifier_to_errno(ret
);
1003 goto failed_addition
;
1006 * If this zone is not populated, then it is not in zonelist.
1007 * This means the page allocator ignores this zone.
1008 * So, zonelist must be updated after online.
1010 mutex_lock(&zonelists_mutex
);
1011 if (!populated_zone(zone
)) {
1012 need_zonelists_rebuild
= 1;
1013 build_all_zonelists(NULL
, zone
);
1016 ret
= walk_system_ram_range(pfn
, nr_pages
, &onlined_pages
,
1017 online_pages_range
);
1019 if (need_zonelists_rebuild
)
1020 zone_pcp_reset(zone
);
1021 mutex_unlock(&zonelists_mutex
);
1022 goto failed_addition
;
1025 zone
->present_pages
+= onlined_pages
;
1027 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
1028 zone
->zone_pgdat
->node_present_pages
+= onlined_pages
;
1029 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
1031 if (onlined_pages
) {
1032 node_states_set_node(nid
, &arg
);
1033 if (need_zonelists_rebuild
)
1034 build_all_zonelists(NULL
, NULL
);
1036 zone_pcp_update(zone
);
1039 mutex_unlock(&zonelists_mutex
);
1041 init_per_zone_wmark_min();
1043 if (onlined_pages
) {
1048 vm_total_pages
= nr_free_pagecache_pages();
1050 writeback_set_ratelimit();
1053 memory_notify(MEM_ONLINE
, &arg
);
1057 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
1058 (unsigned long long) pfn
<< PAGE_SHIFT
,
1059 (((unsigned long long) pfn
+ nr_pages
) << PAGE_SHIFT
) - 1);
1060 memory_notify(MEM_CANCEL_ONLINE
, &arg
);
1063 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1065 static void reset_node_present_pages(pg_data_t
*pgdat
)
1069 for (z
= pgdat
->node_zones
; z
< pgdat
->node_zones
+ MAX_NR_ZONES
; z
++)
1070 z
->present_pages
= 0;
1072 pgdat
->node_present_pages
= 0;
1075 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1076 static pg_data_t __ref
*hotadd_new_pgdat(int nid
, u64 start
)
1078 struct pglist_data
*pgdat
;
1079 unsigned long zones_size
[MAX_NR_ZONES
] = {0};
1080 unsigned long zholes_size
[MAX_NR_ZONES
] = {0};
1081 unsigned long start_pfn
= PFN_DOWN(start
);
1083 pgdat
= NODE_DATA(nid
);
1085 pgdat
= arch_alloc_nodedata(nid
);
1089 arch_refresh_nodedata(nid
, pgdat
);
1092 * Reset the nr_zones, order and classzone_idx before reuse.
1093 * Note that kswapd will init kswapd_classzone_idx properly
1094 * when it starts in the near future.
1096 pgdat
->nr_zones
= 0;
1097 pgdat
->kswapd_order
= 0;
1098 pgdat
->kswapd_classzone_idx
= 0;
1101 /* we can use NODE_DATA(nid) from here */
1103 /* init node's zones as empty zones, we don't have any present pages.*/
1104 free_area_init_node(nid
, zones_size
, start_pfn
, zholes_size
);
1105 pgdat
->per_cpu_nodestats
= alloc_percpu(struct per_cpu_nodestat
);
1108 * The node we allocated has no zone fallback lists. For avoiding
1109 * to access not-initialized zonelist, build here.
1111 mutex_lock(&zonelists_mutex
);
1112 build_all_zonelists(pgdat
, NULL
);
1113 mutex_unlock(&zonelists_mutex
);
1116 * zone->managed_pages is set to an approximate value in
1117 * free_area_init_core(), which will cause
1118 * /sys/device/system/node/nodeX/meminfo has wrong data.
1119 * So reset it to 0 before any memory is onlined.
1121 reset_node_managed_pages(pgdat
);
1124 * When memory is hot-added, all the memory is in offline state. So
1125 * clear all zones' present_pages because they will be updated in
1126 * online_pages() and offline_pages().
1128 reset_node_present_pages(pgdat
);
1133 static void rollback_node_hotadd(int nid
, pg_data_t
*pgdat
)
1135 arch_refresh_nodedata(nid
, NULL
);
1136 free_percpu(pgdat
->per_cpu_nodestats
);
1137 arch_free_nodedata(pgdat
);
1143 * try_online_node - online a node if offlined
1145 * called by cpu_up() to online a node without onlined memory.
1147 int try_online_node(int nid
)
1152 if (node_online(nid
))
1155 mem_hotplug_begin();
1156 pgdat
= hotadd_new_pgdat(nid
, 0);
1158 pr_err("Cannot online node %d due to NULL pgdat\n", nid
);
1162 node_set_online(nid
);
1163 ret
= register_one_node(nid
);
1166 if (pgdat
->node_zonelists
->_zonerefs
->zone
== NULL
) {
1167 mutex_lock(&zonelists_mutex
);
1168 build_all_zonelists(NULL
, NULL
);
1169 mutex_unlock(&zonelists_mutex
);
1177 static int check_hotplug_memory_range(u64 start
, u64 size
)
1179 u64 start_pfn
= PFN_DOWN(start
);
1180 u64 nr_pages
= size
>> PAGE_SHIFT
;
1182 /* Memory range must be aligned with section */
1183 if ((start_pfn
& ~PAGE_SECTION_MASK
) ||
1184 (nr_pages
% PAGES_PER_SECTION
) || (!nr_pages
)) {
1185 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1186 (unsigned long long)start
,
1187 (unsigned long long)size
);
1194 static int online_memory_block(struct memory_block
*mem
, void *arg
)
1196 return device_online(&mem
->dev
);
1199 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1200 int __ref
add_memory_resource(int nid
, struct resource
*res
, bool online
)
1203 pg_data_t
*pgdat
= NULL
;
1209 size
= resource_size(res
);
1211 ret
= check_hotplug_memory_range(start
, size
);
1215 { /* Stupid hack to suppress address-never-null warning */
1216 void *p
= NODE_DATA(nid
);
1220 mem_hotplug_begin();
1223 * Add new range to memblock so that when hotadd_new_pgdat() is called
1224 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1225 * this new range and calculate total pages correctly. The range will
1226 * be removed at hot-remove time.
1228 memblock_add_node(start
, size
, nid
);
1230 new_node
= !node_online(nid
);
1232 pgdat
= hotadd_new_pgdat(nid
, start
);
1238 /* call arch's memory hotadd */
1239 ret
= arch_add_memory(nid
, start
, size
, true);
1244 /* we online node here. we can't roll back from here. */
1245 node_set_online(nid
);
1248 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
1249 unsigned long nr_pages
= size
>> PAGE_SHIFT
;
1251 ret
= __register_one_node(nid
);
1256 * link memory sections under this node. This is already
1257 * done when creatig memory section in register_new_memory
1258 * but that depends to have the node registered so offline
1259 * nodes have to go through register_node.
1260 * TODO clean up this mess.
1262 ret
= link_mem_sections(nid
, start_pfn
, nr_pages
);
1265 * If sysfs file of new node can't create, cpu on the node
1266 * can't be hot-added. There is no rollback way now.
1267 * So, check by BUG_ON() to catch it reluctantly..
1272 /* create new memmap entry */
1273 firmware_map_add_hotplug(start
, start
+ size
, "System RAM");
1275 /* online pages if requested */
1277 walk_memory_range(PFN_DOWN(start
), PFN_UP(start
+ size
- 1),
1278 NULL
, online_memory_block
);
1283 /* rollback pgdat allocation and others */
1284 if (new_pgdat
&& pgdat
)
1285 rollback_node_hotadd(nid
, pgdat
);
1286 memblock_remove(start
, size
);
1292 EXPORT_SYMBOL_GPL(add_memory_resource
);
1294 int __ref
add_memory(int nid
, u64 start
, u64 size
)
1296 struct resource
*res
;
1299 res
= register_memory_resource(start
, size
);
1301 return PTR_ERR(res
);
1303 ret
= add_memory_resource(nid
, res
, memhp_auto_online
);
1305 release_memory_resource(res
);
1308 EXPORT_SYMBOL_GPL(add_memory
);
1310 #ifdef CONFIG_MEMORY_HOTREMOVE
1312 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1313 * set and the size of the free page is given by page_order(). Using this,
1314 * the function determines if the pageblock contains only free pages.
1315 * Due to buddy contraints, a free page at least the size of a pageblock will
1316 * be located at the start of the pageblock
1318 static inline int pageblock_free(struct page
*page
)
1320 return PageBuddy(page
) && page_order(page
) >= pageblock_order
;
1323 /* Return the start of the next active pageblock after a given page */
1324 static struct page
*next_active_pageblock(struct page
*page
)
1326 /* Ensure the starting page is pageblock-aligned */
1327 BUG_ON(page_to_pfn(page
) & (pageblock_nr_pages
- 1));
1329 /* If the entire pageblock is free, move to the end of free page */
1330 if (pageblock_free(page
)) {
1332 /* be careful. we don't have locks, page_order can be changed.*/
1333 order
= page_order(page
);
1334 if ((order
< MAX_ORDER
) && (order
>= pageblock_order
))
1335 return page
+ (1 << order
);
1338 return page
+ pageblock_nr_pages
;
1341 /* Checks if this range of memory is likely to be hot-removable. */
1342 bool is_mem_section_removable(unsigned long start_pfn
, unsigned long nr_pages
)
1344 struct page
*page
= pfn_to_page(start_pfn
);
1345 struct page
*end_page
= page
+ nr_pages
;
1347 /* Check the starting page of each pageblock within the range */
1348 for (; page
< end_page
; page
= next_active_pageblock(page
)) {
1349 if (!is_pageblock_removable_nolock(page
))
1354 /* All pageblocks in the memory block are likely to be hot-removable */
1359 * Confirm all pages in a range [start, end) belong to the same zone.
1360 * When true, return its valid [start, end).
1362 int test_pages_in_a_zone(unsigned long start_pfn
, unsigned long end_pfn
,
1363 unsigned long *valid_start
, unsigned long *valid_end
)
1365 unsigned long pfn
, sec_end_pfn
;
1366 unsigned long start
, end
;
1367 struct zone
*zone
= NULL
;
1370 for (pfn
= start_pfn
, sec_end_pfn
= SECTION_ALIGN_UP(start_pfn
+ 1);
1372 pfn
= sec_end_pfn
, sec_end_pfn
+= PAGES_PER_SECTION
) {
1373 /* Make sure the memory section is present first */
1374 if (!present_section_nr(pfn_to_section_nr(pfn
)))
1376 for (; pfn
< sec_end_pfn
&& pfn
< end_pfn
;
1377 pfn
+= MAX_ORDER_NR_PAGES
) {
1379 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1380 while ((i
< MAX_ORDER_NR_PAGES
) &&
1381 !pfn_valid_within(pfn
+ i
))
1383 if (i
== MAX_ORDER_NR_PAGES
|| pfn
+ i
>= end_pfn
)
1385 page
= pfn_to_page(pfn
+ i
);
1386 if (zone
&& page_zone(page
) != zone
)
1390 zone
= page_zone(page
);
1391 end
= pfn
+ MAX_ORDER_NR_PAGES
;
1396 *valid_start
= start
;
1397 *valid_end
= min(end
, end_pfn
);
1405 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1406 * non-lru movable pages and hugepages). We scan pfn because it's much
1407 * easier than scanning over linked list. This function returns the pfn
1408 * of the first found movable page if it's found, otherwise 0.
1410 static unsigned long scan_movable_pages(unsigned long start
, unsigned long end
)
1414 for (pfn
= start
; pfn
< end
; pfn
++) {
1415 if (pfn_valid(pfn
)) {
1416 page
= pfn_to_page(pfn
);
1419 if (__PageMovable(page
))
1421 if (PageHuge(page
)) {
1422 if (page_huge_active(page
))
1425 pfn
= round_up(pfn
+ 1,
1426 1 << compound_order(page
)) - 1;
1433 static struct page
*new_node_page(struct page
*page
, unsigned long private,
1436 gfp_t gfp_mask
= GFP_USER
| __GFP_MOVABLE
;
1437 int nid
= page_to_nid(page
);
1438 nodemask_t nmask
= node_states
[N_MEMORY
];
1439 struct page
*new_page
= NULL
;
1442 * TODO: allocate a destination hugepage from a nearest neighbor node,
1443 * accordance with memory policy of the user process if possible. For
1444 * now as a simple work-around, we use the next node for destination.
1447 return alloc_huge_page_node(page_hstate(compound_head(page
)),
1448 next_node_in(nid
, nmask
));
1450 node_clear(nid
, nmask
);
1452 if (PageHighMem(page
)
1453 || (zone_idx(page_zone(page
)) == ZONE_MOVABLE
))
1454 gfp_mask
|= __GFP_HIGHMEM
;
1456 if (!nodes_empty(nmask
))
1457 new_page
= __alloc_pages_nodemask(gfp_mask
, 0, nid
, &nmask
);
1459 new_page
= __alloc_pages(gfp_mask
, 0, nid
);
1464 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1466 do_migrate_range(unsigned long start_pfn
, unsigned long end_pfn
)
1470 int move_pages
= NR_OFFLINE_AT_ONCE_PAGES
;
1471 int not_managed
= 0;
1475 for (pfn
= start_pfn
; pfn
< end_pfn
&& move_pages
> 0; pfn
++) {
1476 if (!pfn_valid(pfn
))
1478 page
= pfn_to_page(pfn
);
1480 if (PageHuge(page
)) {
1481 struct page
*head
= compound_head(page
);
1482 pfn
= page_to_pfn(head
) + (1<<compound_order(head
)) - 1;
1483 if (compound_order(head
) > PFN_SECTION_SHIFT
) {
1487 if (isolate_huge_page(page
, &source
))
1488 move_pages
-= 1 << compound_order(head
);
1492 if (!get_page_unless_zero(page
))
1495 * We can skip free pages. And we can deal with pages on
1496 * LRU and non-lru movable pages.
1499 ret
= isolate_lru_page(page
);
1501 ret
= isolate_movable_page(page
, ISOLATE_UNEVICTABLE
);
1502 if (!ret
) { /* Success */
1504 list_add_tail(&page
->lru
, &source
);
1506 if (!__PageMovable(page
))
1507 inc_node_page_state(page
, NR_ISOLATED_ANON
+
1508 page_is_file_cache(page
));
1511 #ifdef CONFIG_DEBUG_VM
1512 pr_alert("failed to isolate pfn %lx\n", pfn
);
1513 dump_page(page
, "isolation failed");
1516 /* Because we don't have big zone->lock. we should
1517 check this again here. */
1518 if (page_count(page
)) {
1525 if (!list_empty(&source
)) {
1527 putback_movable_pages(&source
);
1531 /* Allocate a new page from the nearest neighbor node */
1532 ret
= migrate_pages(&source
, new_node_page
, NULL
, 0,
1533 MIGRATE_SYNC
, MR_MEMORY_HOTPLUG
);
1535 putback_movable_pages(&source
);
1542 * remove from free_area[] and mark all as Reserved.
1545 offline_isolated_pages_cb(unsigned long start
, unsigned long nr_pages
,
1548 __offline_isolated_pages(start
, start
+ nr_pages
);
1553 offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
1555 walk_system_ram_range(start_pfn
, end_pfn
- start_pfn
, NULL
,
1556 offline_isolated_pages_cb
);
1560 * Check all pages in range, recoreded as memory resource, are isolated.
1563 check_pages_isolated_cb(unsigned long start_pfn
, unsigned long nr_pages
,
1567 long offlined
= *(long *)data
;
1568 ret
= test_pages_isolated(start_pfn
, start_pfn
+ nr_pages
, true);
1569 offlined
= nr_pages
;
1571 *(long *)data
+= offlined
;
1576 check_pages_isolated(unsigned long start_pfn
, unsigned long end_pfn
)
1581 ret
= walk_system_ram_range(start_pfn
, end_pfn
- start_pfn
, &offlined
,
1582 check_pages_isolated_cb
);
1584 offlined
= (long)ret
;
1588 static int __init
cmdline_parse_movable_node(char *p
)
1590 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1591 movable_node_enabled
= true;
1593 pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1597 early_param("movable_node", cmdline_parse_movable_node
);
1599 /* check which state of node_states will be changed when offline memory */
1600 static void node_states_check_changes_offline(unsigned long nr_pages
,
1601 struct zone
*zone
, struct memory_notify
*arg
)
1603 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1604 unsigned long present_pages
= 0;
1605 enum zone_type zt
, zone_last
= ZONE_NORMAL
;
1608 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1609 * contains nodes which have zones of 0...ZONE_NORMAL,
1610 * set zone_last to ZONE_NORMAL.
1612 * If we don't have HIGHMEM nor movable node,
1613 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1614 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1616 if (N_MEMORY
== N_NORMAL_MEMORY
)
1617 zone_last
= ZONE_MOVABLE
;
1620 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1621 * If the memory to be offline is in a zone of 0...zone_last,
1622 * and it is the last present memory, 0...zone_last will
1623 * become empty after offline , thus we can determind we will
1624 * need to clear the node from node_states[N_NORMAL_MEMORY].
1626 for (zt
= 0; zt
<= zone_last
; zt
++)
1627 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1628 if (zone_idx(zone
) <= zone_last
&& nr_pages
>= present_pages
)
1629 arg
->status_change_nid_normal
= zone_to_nid(zone
);
1631 arg
->status_change_nid_normal
= -1;
1633 #ifdef CONFIG_HIGHMEM
1635 * If we have movable node, node_states[N_HIGH_MEMORY]
1636 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1637 * set zone_last to ZONE_HIGHMEM.
1639 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1640 * contains nodes which have zones of 0...ZONE_MOVABLE,
1641 * set zone_last to ZONE_MOVABLE.
1643 zone_last
= ZONE_HIGHMEM
;
1644 if (N_MEMORY
== N_HIGH_MEMORY
)
1645 zone_last
= ZONE_MOVABLE
;
1647 for (; zt
<= zone_last
; zt
++)
1648 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1649 if (zone_idx(zone
) <= zone_last
&& nr_pages
>= present_pages
)
1650 arg
->status_change_nid_high
= zone_to_nid(zone
);
1652 arg
->status_change_nid_high
= -1;
1654 arg
->status_change_nid_high
= arg
->status_change_nid_normal
;
1658 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1660 zone_last
= ZONE_MOVABLE
;
1663 * check whether node_states[N_HIGH_MEMORY] will be changed
1664 * If we try to offline the last present @nr_pages from the node,
1665 * we can determind we will need to clear the node from
1666 * node_states[N_HIGH_MEMORY].
1668 for (; zt
<= zone_last
; zt
++)
1669 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1670 if (nr_pages
>= present_pages
)
1671 arg
->status_change_nid
= zone_to_nid(zone
);
1673 arg
->status_change_nid
= -1;
1676 static void node_states_clear_node(int node
, struct memory_notify
*arg
)
1678 if (arg
->status_change_nid_normal
>= 0)
1679 node_clear_state(node
, N_NORMAL_MEMORY
);
1681 if ((N_MEMORY
!= N_NORMAL_MEMORY
) &&
1682 (arg
->status_change_nid_high
>= 0))
1683 node_clear_state(node
, N_HIGH_MEMORY
);
1685 if ((N_MEMORY
!= N_HIGH_MEMORY
) &&
1686 (arg
->status_change_nid
>= 0))
1687 node_clear_state(node
, N_MEMORY
);
1690 static int __ref
__offline_pages(unsigned long start_pfn
,
1691 unsigned long end_pfn
, unsigned long timeout
)
1693 unsigned long pfn
, nr_pages
, expire
;
1694 long offlined_pages
;
1695 int ret
, drain
, retry_max
, node
;
1696 unsigned long flags
;
1697 unsigned long valid_start
, valid_end
;
1699 struct memory_notify arg
;
1701 /* at least, alignment against pageblock is necessary */
1702 if (!IS_ALIGNED(start_pfn
, pageblock_nr_pages
))
1704 if (!IS_ALIGNED(end_pfn
, pageblock_nr_pages
))
1706 /* This makes hotplug much easier...and readable.
1707 we assume this for now. .*/
1708 if (!test_pages_in_a_zone(start_pfn
, end_pfn
, &valid_start
, &valid_end
))
1711 zone
= page_zone(pfn_to_page(valid_start
));
1712 node
= zone_to_nid(zone
);
1713 nr_pages
= end_pfn
- start_pfn
;
1715 /* set above range as isolated */
1716 ret
= start_isolate_page_range(start_pfn
, end_pfn
,
1717 MIGRATE_MOVABLE
, true);
1721 arg
.start_pfn
= start_pfn
;
1722 arg
.nr_pages
= nr_pages
;
1723 node_states_check_changes_offline(nr_pages
, zone
, &arg
);
1725 ret
= memory_notify(MEM_GOING_OFFLINE
, &arg
);
1726 ret
= notifier_to_errno(ret
);
1728 goto failed_removal
;
1731 expire
= jiffies
+ timeout
;
1735 /* start memory hot removal */
1737 if (time_after(jiffies
, expire
))
1738 goto failed_removal
;
1740 if (signal_pending(current
))
1741 goto failed_removal
;
1744 lru_add_drain_all();
1746 drain_all_pages(zone
);
1749 pfn
= scan_movable_pages(start_pfn
, end_pfn
);
1750 if (pfn
) { /* We have movable pages */
1751 ret
= do_migrate_range(pfn
, end_pfn
);
1757 if (--retry_max
== 0)
1758 goto failed_removal
;
1764 /* drain all zone's lru pagevec, this is asynchronous... */
1765 lru_add_drain_all();
1767 /* drain pcp pages, this is synchronous. */
1768 drain_all_pages(zone
);
1770 * dissolve free hugepages in the memory block before doing offlining
1771 * actually in order to make hugetlbfs's object counting consistent.
1773 ret
= dissolve_free_huge_pages(start_pfn
, end_pfn
);
1775 goto failed_removal
;
1777 offlined_pages
= check_pages_isolated(start_pfn
, end_pfn
);
1778 if (offlined_pages
< 0) {
1780 goto failed_removal
;
1782 pr_info("Offlined Pages %ld\n", offlined_pages
);
1783 /* Ok, all of our target is isolated.
1784 We cannot do rollback at this point. */
1785 offline_isolated_pages(start_pfn
, end_pfn
);
1786 /* reset pagetype flags and makes migrate type to be MOVABLE */
1787 undo_isolate_page_range(start_pfn
, end_pfn
, MIGRATE_MOVABLE
);
1788 /* removal success */
1789 adjust_managed_page_count(pfn_to_page(start_pfn
), -offlined_pages
);
1790 zone
->present_pages
-= offlined_pages
;
1792 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
1793 zone
->zone_pgdat
->node_present_pages
-= offlined_pages
;
1794 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
1796 init_per_zone_wmark_min();
1798 if (!populated_zone(zone
)) {
1799 zone_pcp_reset(zone
);
1800 mutex_lock(&zonelists_mutex
);
1801 build_all_zonelists(NULL
, NULL
);
1802 mutex_unlock(&zonelists_mutex
);
1804 zone_pcp_update(zone
);
1806 node_states_clear_node(node
, &arg
);
1807 if (arg
.status_change_nid
>= 0) {
1809 kcompactd_stop(node
);
1812 vm_total_pages
= nr_free_pagecache_pages();
1813 writeback_set_ratelimit();
1815 memory_notify(MEM_OFFLINE
, &arg
);
1819 pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n",
1820 (unsigned long long) start_pfn
<< PAGE_SHIFT
,
1821 ((unsigned long long) end_pfn
<< PAGE_SHIFT
) - 1);
1822 memory_notify(MEM_CANCEL_OFFLINE
, &arg
);
1823 /* pushback to free area */
1824 undo_isolate_page_range(start_pfn
, end_pfn
, MIGRATE_MOVABLE
);
1828 /* Must be protected by mem_hotplug_begin() */
1829 int offline_pages(unsigned long start_pfn
, unsigned long nr_pages
)
1831 return __offline_pages(start_pfn
, start_pfn
+ nr_pages
, 120 * HZ
);
1833 #endif /* CONFIG_MEMORY_HOTREMOVE */
1836 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1837 * @start_pfn: start pfn of the memory range
1838 * @end_pfn: end pfn of the memory range
1839 * @arg: argument passed to func
1840 * @func: callback for each memory section walked
1842 * This function walks through all present mem sections in range
1843 * [start_pfn, end_pfn) and call func on each mem section.
1845 * Returns the return value of func.
1847 int walk_memory_range(unsigned long start_pfn
, unsigned long end_pfn
,
1848 void *arg
, int (*func
)(struct memory_block
*, void *))
1850 struct memory_block
*mem
= NULL
;
1851 struct mem_section
*section
;
1852 unsigned long pfn
, section_nr
;
1855 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
1856 section_nr
= pfn_to_section_nr(pfn
);
1857 if (!present_section_nr(section_nr
))
1860 section
= __nr_to_section(section_nr
);
1861 /* same memblock? */
1863 if ((section_nr
>= mem
->start_section_nr
) &&
1864 (section_nr
<= mem
->end_section_nr
))
1867 mem
= find_memory_block_hinted(section
, mem
);
1871 ret
= func(mem
, arg
);
1873 kobject_put(&mem
->dev
.kobj
);
1879 kobject_put(&mem
->dev
.kobj
);
1884 #ifdef CONFIG_MEMORY_HOTREMOVE
1885 static int check_memblock_offlined_cb(struct memory_block
*mem
, void *arg
)
1887 int ret
= !is_memblock_offlined(mem
);
1889 if (unlikely(ret
)) {
1890 phys_addr_t beginpa
, endpa
;
1892 beginpa
= PFN_PHYS(section_nr_to_pfn(mem
->start_section_nr
));
1893 endpa
= PFN_PHYS(section_nr_to_pfn(mem
->end_section_nr
+ 1))-1;
1894 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1901 static int check_cpu_on_node(pg_data_t
*pgdat
)
1905 for_each_present_cpu(cpu
) {
1906 if (cpu_to_node(cpu
) == pgdat
->node_id
)
1908 * the cpu on this node isn't removed, and we can't
1909 * offline this node.
1917 static void unmap_cpu_on_node(pg_data_t
*pgdat
)
1919 #ifdef CONFIG_ACPI_NUMA
1922 for_each_possible_cpu(cpu
)
1923 if (cpu_to_node(cpu
) == pgdat
->node_id
)
1924 numa_clear_node(cpu
);
1928 static int check_and_unmap_cpu_on_node(pg_data_t
*pgdat
)
1932 ret
= check_cpu_on_node(pgdat
);
1937 * the node will be offlined when we come here, so we can clear
1938 * the cpu_to_node() now.
1941 unmap_cpu_on_node(pgdat
);
1948 * Offline a node if all memory sections and cpus of the node are removed.
1950 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1951 * and online/offline operations before this call.
1953 void try_offline_node(int nid
)
1955 pg_data_t
*pgdat
= NODE_DATA(nid
);
1956 unsigned long start_pfn
= pgdat
->node_start_pfn
;
1957 unsigned long end_pfn
= start_pfn
+ pgdat
->node_spanned_pages
;
1960 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
1961 unsigned long section_nr
= pfn_to_section_nr(pfn
);
1963 if (!present_section_nr(section_nr
))
1966 if (pfn_to_nid(pfn
) != nid
)
1970 * some memory sections of this node are not removed, and we
1971 * can't offline node now.
1976 if (check_and_unmap_cpu_on_node(pgdat
))
1980 * all memory/cpu of this node are removed, we can offline this
1983 node_set_offline(nid
);
1984 unregister_one_node(nid
);
1986 EXPORT_SYMBOL(try_offline_node
);
1991 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1992 * and online/offline operations before this call, as required by
1993 * try_offline_node().
1995 void __ref
remove_memory(int nid
, u64 start
, u64 size
)
1999 BUG_ON(check_hotplug_memory_range(start
, size
));
2001 mem_hotplug_begin();
2004 * All memory blocks must be offlined before removing memory. Check
2005 * whether all memory blocks in question are offline and trigger a BUG()
2006 * if this is not the case.
2008 ret
= walk_memory_range(PFN_DOWN(start
), PFN_UP(start
+ size
- 1), NULL
,
2009 check_memblock_offlined_cb
);
2013 /* remove memmap entry */
2014 firmware_map_remove(start
, start
+ size
, "System RAM");
2015 memblock_free(start
, size
);
2016 memblock_remove(start
, size
);
2018 arch_remove_memory(start
, size
);
2020 try_offline_node(nid
);
2024 EXPORT_SYMBOL_GPL(remove_memory
);
2025 #endif /* CONFIG_MEMORY_HOTREMOVE */