1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/mm/memory_hotplug.c
8 #include <linux/stddef.h>
10 #include <linux/sched/signal.h>
11 #include <linux/swap.h>
12 #include <linux/interrupt.h>
13 #include <linux/pagemap.h>
14 #include <linux/compiler.h>
15 #include <linux/export.h>
16 #include <linux/pagevec.h>
17 #include <linux/writeback.h>
18 #include <linux/slab.h>
19 #include <linux/sysctl.h>
20 #include <linux/cpu.h>
21 #include <linux/memory.h>
22 #include <linux/memremap.h>
23 #include <linux/memory_hotplug.h>
24 #include <linux/highmem.h>
25 #include <linux/vmalloc.h>
26 #include <linux/ioport.h>
27 #include <linux/delay.h>
28 #include <linux/migrate.h>
29 #include <linux/page-isolation.h>
30 #include <linux/pfn.h>
31 #include <linux/suspend.h>
32 #include <linux/mm_inline.h>
33 #include <linux/firmware-map.h>
34 #include <linux/stop_machine.h>
35 #include <linux/hugetlb.h>
36 #include <linux/memblock.h>
37 #include <linux/compaction.h>
38 #include <linux/rmap.h>
40 #include <asm/tlbflush.h>
46 * online_page_callback contains pointer to current page onlining function.
47 * Initially it is generic_online_page(). If it is required it could be
48 * changed by calling set_online_page_callback() for callback registration
49 * and restore_online_page_callback() for generic callback restore.
52 static void generic_online_page(struct page
*page
, unsigned int order
);
54 static online_page_callback_t online_page_callback
= generic_online_page
;
55 static DEFINE_MUTEX(online_page_callback_lock
);
57 DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock
);
59 void get_online_mems(void)
61 percpu_down_read(&mem_hotplug_lock
);
64 void put_online_mems(void)
66 percpu_up_read(&mem_hotplug_lock
);
69 bool movable_node_enabled
= false;
71 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
72 bool memhp_auto_online
;
74 bool memhp_auto_online
= true;
76 EXPORT_SYMBOL_GPL(memhp_auto_online
);
78 static int __init
setup_memhp_default_state(char *str
)
80 if (!strcmp(str
, "online"))
81 memhp_auto_online
= true;
82 else if (!strcmp(str
, "offline"))
83 memhp_auto_online
= false;
87 __setup("memhp_default_state=", setup_memhp_default_state
);
89 void mem_hotplug_begin(void)
92 percpu_down_write(&mem_hotplug_lock
);
95 void mem_hotplug_done(void)
97 percpu_up_write(&mem_hotplug_lock
);
101 u64 max_mem_size
= U64_MAX
;
103 /* add this memory to iomem resource */
104 static struct resource
*register_memory_resource(u64 start
, u64 size
)
106 struct resource
*res
;
107 unsigned long flags
= IORESOURCE_SYSTEM_RAM
| IORESOURCE_BUSY
;
108 char *resource_name
= "System RAM";
110 if (start
+ size
> max_mem_size
)
111 return ERR_PTR(-E2BIG
);
114 * Request ownership of the new memory range. This might be
115 * a child of an existing resource that was present but
116 * not marked as busy.
118 res
= __request_region(&iomem_resource
, start
, size
,
119 resource_name
, flags
);
122 pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n",
123 start
, start
+ size
);
124 return ERR_PTR(-EEXIST
);
129 static void release_memory_resource(struct resource
*res
)
133 release_resource(res
);
138 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
139 void get_page_bootmem(unsigned long info
, struct page
*page
,
142 page
->freelist
= (void *)type
;
143 SetPagePrivate(page
);
144 set_page_private(page
, info
);
148 void put_page_bootmem(struct page
*page
)
152 type
= (unsigned long) page
->freelist
;
153 BUG_ON(type
< MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE
||
154 type
> MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE
);
156 if (page_ref_dec_return(page
) == 1) {
157 page
->freelist
= NULL
;
158 ClearPagePrivate(page
);
159 set_page_private(page
, 0);
160 INIT_LIST_HEAD(&page
->lru
);
161 free_reserved_page(page
);
165 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
166 #ifndef CONFIG_SPARSEMEM_VMEMMAP
167 static void register_page_bootmem_info_section(unsigned long start_pfn
)
169 unsigned long *usemap
, mapsize
, section_nr
, i
;
170 struct mem_section
*ms
;
171 struct page
*page
, *memmap
;
173 section_nr
= pfn_to_section_nr(start_pfn
);
174 ms
= __nr_to_section(section_nr
);
176 /* Get section's memmap address */
177 memmap
= sparse_decode_mem_map(ms
->section_mem_map
, section_nr
);
180 * Get page for the memmap's phys address
181 * XXX: need more consideration for sparse_vmemmap...
183 page
= virt_to_page(memmap
);
184 mapsize
= sizeof(struct page
) * PAGES_PER_SECTION
;
185 mapsize
= PAGE_ALIGN(mapsize
) >> PAGE_SHIFT
;
187 /* remember memmap's page */
188 for (i
= 0; i
< mapsize
; i
++, page
++)
189 get_page_bootmem(section_nr
, page
, SECTION_INFO
);
191 usemap
= ms
->pageblock_flags
;
192 page
= virt_to_page(usemap
);
194 mapsize
= PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT
;
196 for (i
= 0; i
< mapsize
; i
++, page
++)
197 get_page_bootmem(section_nr
, page
, MIX_SECTION_INFO
);
200 #else /* CONFIG_SPARSEMEM_VMEMMAP */
201 static void register_page_bootmem_info_section(unsigned long start_pfn
)
203 unsigned long *usemap
, mapsize
, section_nr
, i
;
204 struct mem_section
*ms
;
205 struct page
*page
, *memmap
;
207 section_nr
= pfn_to_section_nr(start_pfn
);
208 ms
= __nr_to_section(section_nr
);
210 memmap
= sparse_decode_mem_map(ms
->section_mem_map
, section_nr
);
212 register_page_bootmem_memmap(section_nr
, memmap
, PAGES_PER_SECTION
);
214 usemap
= ms
->pageblock_flags
;
215 page
= virt_to_page(usemap
);
217 mapsize
= PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT
;
219 for (i
= 0; i
< mapsize
; i
++, page
++)
220 get_page_bootmem(section_nr
, page
, MIX_SECTION_INFO
);
222 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
224 void __init
register_page_bootmem_info_node(struct pglist_data
*pgdat
)
226 unsigned long i
, pfn
, end_pfn
, nr_pages
;
227 int node
= pgdat
->node_id
;
230 nr_pages
= PAGE_ALIGN(sizeof(struct pglist_data
)) >> PAGE_SHIFT
;
231 page
= virt_to_page(pgdat
);
233 for (i
= 0; i
< nr_pages
; i
++, page
++)
234 get_page_bootmem(node
, page
, NODE_INFO
);
236 pfn
= pgdat
->node_start_pfn
;
237 end_pfn
= pgdat_end_pfn(pgdat
);
239 /* register section info */
240 for (; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
242 * Some platforms can assign the same pfn to multiple nodes - on
243 * node0 as well as nodeN. To avoid registering a pfn against
244 * multiple nodes we check that this pfn does not already
245 * reside in some other nodes.
247 if (pfn_valid(pfn
) && (early_pfn_to_nid(pfn
) == node
))
248 register_page_bootmem_info_section(pfn
);
251 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
253 static int __meminit
__add_section(int nid
, unsigned long phys_start_pfn
,
254 struct vmem_altmap
*altmap
, bool want_memblock
)
258 if (pfn_valid(phys_start_pfn
))
261 ret
= sparse_add_one_section(nid
, phys_start_pfn
, altmap
);
268 return hotplug_memory_register(nid
, __pfn_to_section(phys_start_pfn
));
272 * Reasonably generic function for adding memory. It is
273 * expected that archs that support memory hotplug will
274 * call this function after deciding the zone to which to
277 int __ref
__add_pages(int nid
, unsigned long phys_start_pfn
,
278 unsigned long nr_pages
, struct mhp_restrictions
*restrictions
)
282 int start_sec
, end_sec
;
283 struct vmem_altmap
*altmap
= restrictions
->altmap
;
285 /* during initialize mem_map, align hot-added range to section */
286 start_sec
= pfn_to_section_nr(phys_start_pfn
);
287 end_sec
= pfn_to_section_nr(phys_start_pfn
+ nr_pages
- 1);
291 * Validate altmap is within bounds of the total request
293 if (altmap
->base_pfn
!= phys_start_pfn
294 || vmem_altmap_offset(altmap
) > nr_pages
) {
295 pr_warn_once("memory add fail, invalid altmap\n");
302 for (i
= start_sec
; i
<= end_sec
; i
++) {
303 err
= __add_section(nid
, section_nr_to_pfn(i
), altmap
,
304 restrictions
->flags
& MHP_MEMBLOCK_API
);
307 * EEXIST is finally dealt with by ioresource collision
308 * check. see add_memory() => register_memory_resource()
309 * Warning will be printed if there is collision.
311 if (err
&& (err
!= -EEXIST
))
316 vmemmap_populate_print_last();
321 #ifdef CONFIG_MEMORY_HOTREMOVE
322 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
323 static unsigned long find_smallest_section_pfn(int nid
, struct zone
*zone
,
324 unsigned long start_pfn
,
325 unsigned long end_pfn
)
327 struct mem_section
*ms
;
329 for (; start_pfn
< end_pfn
; start_pfn
+= PAGES_PER_SECTION
) {
330 ms
= __pfn_to_section(start_pfn
);
332 if (unlikely(!valid_section(ms
)))
335 if (unlikely(pfn_to_nid(start_pfn
) != nid
))
338 if (zone
&& zone
!= page_zone(pfn_to_page(start_pfn
)))
347 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
348 static unsigned long find_biggest_section_pfn(int nid
, struct zone
*zone
,
349 unsigned long start_pfn
,
350 unsigned long end_pfn
)
352 struct mem_section
*ms
;
355 /* pfn is the end pfn of a memory section. */
357 for (; pfn
>= start_pfn
; pfn
-= PAGES_PER_SECTION
) {
358 ms
= __pfn_to_section(pfn
);
360 if (unlikely(!valid_section(ms
)))
363 if (unlikely(pfn_to_nid(pfn
) != nid
))
366 if (zone
&& zone
!= page_zone(pfn_to_page(pfn
)))
375 static void shrink_zone_span(struct zone
*zone
, unsigned long start_pfn
,
376 unsigned long end_pfn
)
378 unsigned long zone_start_pfn
= zone
->zone_start_pfn
;
379 unsigned long z
= zone_end_pfn(zone
); /* zone_end_pfn namespace clash */
380 unsigned long zone_end_pfn
= z
;
382 struct mem_section
*ms
;
383 int nid
= zone_to_nid(zone
);
385 zone_span_writelock(zone
);
386 if (zone_start_pfn
== start_pfn
) {
388 * If the section is smallest section in the zone, it need
389 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
390 * In this case, we find second smallest valid mem_section
391 * for shrinking zone.
393 pfn
= find_smallest_section_pfn(nid
, zone
, end_pfn
,
396 zone
->zone_start_pfn
= pfn
;
397 zone
->spanned_pages
= zone_end_pfn
- pfn
;
399 } else if (zone_end_pfn
== end_pfn
) {
401 * If the section is biggest section in the zone, it need
402 * shrink zone->spanned_pages.
403 * In this case, we find second biggest valid mem_section for
406 pfn
= find_biggest_section_pfn(nid
, zone
, zone_start_pfn
,
409 zone
->spanned_pages
= pfn
- zone_start_pfn
+ 1;
413 * The section is not biggest or smallest mem_section in the zone, it
414 * only creates a hole in the zone. So in this case, we need not
415 * change the zone. But perhaps, the zone has only hole data. Thus
416 * it check the zone has only hole or not.
418 pfn
= zone_start_pfn
;
419 for (; pfn
< zone_end_pfn
; pfn
+= PAGES_PER_SECTION
) {
420 ms
= __pfn_to_section(pfn
);
422 if (unlikely(!valid_section(ms
)))
425 if (page_zone(pfn_to_page(pfn
)) != zone
)
428 /* If the section is current section, it continues the loop */
429 if (start_pfn
== pfn
)
432 /* If we find valid section, we have nothing to do */
433 zone_span_writeunlock(zone
);
437 /* The zone has no valid section */
438 zone
->zone_start_pfn
= 0;
439 zone
->spanned_pages
= 0;
440 zone_span_writeunlock(zone
);
443 static void shrink_pgdat_span(struct pglist_data
*pgdat
,
444 unsigned long start_pfn
, unsigned long end_pfn
)
446 unsigned long pgdat_start_pfn
= pgdat
->node_start_pfn
;
447 unsigned long p
= pgdat_end_pfn(pgdat
); /* pgdat_end_pfn namespace clash */
448 unsigned long pgdat_end_pfn
= p
;
450 struct mem_section
*ms
;
451 int nid
= pgdat
->node_id
;
453 if (pgdat_start_pfn
== start_pfn
) {
455 * If the section is smallest section in the pgdat, it need
456 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
457 * In this case, we find second smallest valid mem_section
458 * for shrinking zone.
460 pfn
= find_smallest_section_pfn(nid
, NULL
, end_pfn
,
463 pgdat
->node_start_pfn
= pfn
;
464 pgdat
->node_spanned_pages
= pgdat_end_pfn
- pfn
;
466 } else if (pgdat_end_pfn
== end_pfn
) {
468 * If the section is biggest section in the pgdat, it need
469 * shrink pgdat->node_spanned_pages.
470 * In this case, we find second biggest valid mem_section for
473 pfn
= find_biggest_section_pfn(nid
, NULL
, pgdat_start_pfn
,
476 pgdat
->node_spanned_pages
= pfn
- pgdat_start_pfn
+ 1;
480 * If the section is not biggest or smallest mem_section in the pgdat,
481 * it only creates a hole in the pgdat. So in this case, we need not
483 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
484 * has only hole or not.
486 pfn
= pgdat_start_pfn
;
487 for (; pfn
< pgdat_end_pfn
; pfn
+= PAGES_PER_SECTION
) {
488 ms
= __pfn_to_section(pfn
);
490 if (unlikely(!valid_section(ms
)))
493 if (pfn_to_nid(pfn
) != nid
)
496 /* If the section is current section, it continues the loop */
497 if (start_pfn
== pfn
)
500 /* If we find valid section, we have nothing to do */
504 /* The pgdat has no valid section */
505 pgdat
->node_start_pfn
= 0;
506 pgdat
->node_spanned_pages
= 0;
509 static void __remove_zone(struct zone
*zone
, unsigned long start_pfn
)
511 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
512 int nr_pages
= PAGES_PER_SECTION
;
515 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
516 shrink_zone_span(zone
, start_pfn
, start_pfn
+ nr_pages
);
517 shrink_pgdat_span(pgdat
, start_pfn
, start_pfn
+ nr_pages
);
518 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
521 static void __remove_section(struct zone
*zone
, struct mem_section
*ms
,
522 unsigned long map_offset
,
523 struct vmem_altmap
*altmap
)
525 unsigned long start_pfn
;
528 if (WARN_ON_ONCE(!valid_section(ms
)))
531 unregister_memory_section(ms
);
533 scn_nr
= __section_nr(ms
);
534 start_pfn
= section_nr_to_pfn((unsigned long)scn_nr
);
535 __remove_zone(zone
, start_pfn
);
537 sparse_remove_one_section(zone
, ms
, map_offset
, altmap
);
541 * __remove_pages() - remove sections of pages from a zone
542 * @zone: zone from which pages need to be removed
543 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
544 * @nr_pages: number of pages to remove (must be multiple of section size)
545 * @altmap: alternative device page map or %NULL if default memmap is used
547 * Generic helper function to remove section mappings and sysfs entries
548 * for the section of the memory we are removing. Caller needs to make
549 * sure that pages are marked reserved and zones are adjust properly by
550 * calling offline_pages().
552 void __remove_pages(struct zone
*zone
, unsigned long phys_start_pfn
,
553 unsigned long nr_pages
, struct vmem_altmap
*altmap
)
556 unsigned long map_offset
= 0;
557 int sections_to_remove
;
559 /* In the ZONE_DEVICE case device driver owns the memory region */
560 if (is_dev_zone(zone
))
561 map_offset
= vmem_altmap_offset(altmap
);
563 clear_zone_contiguous(zone
);
566 * We can only remove entire sections
568 BUG_ON(phys_start_pfn
& ~PAGE_SECTION_MASK
);
569 BUG_ON(nr_pages
% PAGES_PER_SECTION
);
571 sections_to_remove
= nr_pages
/ PAGES_PER_SECTION
;
572 for (i
= 0; i
< sections_to_remove
; i
++) {
573 unsigned long pfn
= phys_start_pfn
+ i
*PAGES_PER_SECTION
;
576 __remove_section(zone
, __pfn_to_section(pfn
), map_offset
,
581 set_zone_contiguous(zone
);
583 #endif /* CONFIG_MEMORY_HOTREMOVE */
585 int set_online_page_callback(online_page_callback_t callback
)
590 mutex_lock(&online_page_callback_lock
);
592 if (online_page_callback
== generic_online_page
) {
593 online_page_callback
= callback
;
597 mutex_unlock(&online_page_callback_lock
);
602 EXPORT_SYMBOL_GPL(set_online_page_callback
);
604 int restore_online_page_callback(online_page_callback_t callback
)
609 mutex_lock(&online_page_callback_lock
);
611 if (online_page_callback
== callback
) {
612 online_page_callback
= generic_online_page
;
616 mutex_unlock(&online_page_callback_lock
);
621 EXPORT_SYMBOL_GPL(restore_online_page_callback
);
623 void __online_page_set_limits(struct page
*page
)
626 EXPORT_SYMBOL_GPL(__online_page_set_limits
);
628 void __online_page_increment_counters(struct page
*page
)
630 adjust_managed_page_count(page
, 1);
632 EXPORT_SYMBOL_GPL(__online_page_increment_counters
);
634 void __online_page_free(struct page
*page
)
636 __free_reserved_page(page
);
638 EXPORT_SYMBOL_GPL(__online_page_free
);
640 static void generic_online_page(struct page
*page
, unsigned int order
)
642 kernel_map_pages(page
, 1 << order
, 1);
643 __free_pages_core(page
, order
);
644 totalram_pages_add(1UL << order
);
645 #ifdef CONFIG_HIGHMEM
646 if (PageHighMem(page
))
647 totalhigh_pages_add(1UL << order
);
651 static int online_pages_blocks(unsigned long start
, unsigned long nr_pages
)
653 unsigned long end
= start
+ nr_pages
;
654 int order
, onlined_pages
= 0;
656 while (start
< end
) {
657 order
= min(MAX_ORDER
- 1,
658 get_order(PFN_PHYS(end
) - PFN_PHYS(start
)));
659 (*online_page_callback
)(pfn_to_page(start
), order
);
661 onlined_pages
+= (1UL << order
);
662 start
+= (1UL << order
);
664 return onlined_pages
;
667 static int online_pages_range(unsigned long start_pfn
, unsigned long nr_pages
,
670 unsigned long onlined_pages
= *(unsigned long *)arg
;
672 if (PageReserved(pfn_to_page(start_pfn
)))
673 onlined_pages
+= online_pages_blocks(start_pfn
, nr_pages
);
675 online_mem_sections(start_pfn
, start_pfn
+ nr_pages
);
677 *(unsigned long *)arg
= onlined_pages
;
681 /* check which state of node_states will be changed when online memory */
682 static void node_states_check_changes_online(unsigned long nr_pages
,
683 struct zone
*zone
, struct memory_notify
*arg
)
685 int nid
= zone_to_nid(zone
);
687 arg
->status_change_nid
= NUMA_NO_NODE
;
688 arg
->status_change_nid_normal
= NUMA_NO_NODE
;
689 arg
->status_change_nid_high
= NUMA_NO_NODE
;
691 if (!node_state(nid
, N_MEMORY
))
692 arg
->status_change_nid
= nid
;
693 if (zone_idx(zone
) <= ZONE_NORMAL
&& !node_state(nid
, N_NORMAL_MEMORY
))
694 arg
->status_change_nid_normal
= nid
;
695 #ifdef CONFIG_HIGHMEM
696 if (zone_idx(zone
) <= ZONE_HIGHMEM
&& !node_state(nid
, N_HIGH_MEMORY
))
697 arg
->status_change_nid_high
= nid
;
701 static void node_states_set_node(int node
, struct memory_notify
*arg
)
703 if (arg
->status_change_nid_normal
>= 0)
704 node_set_state(node
, N_NORMAL_MEMORY
);
706 if (arg
->status_change_nid_high
>= 0)
707 node_set_state(node
, N_HIGH_MEMORY
);
709 if (arg
->status_change_nid
>= 0)
710 node_set_state(node
, N_MEMORY
);
713 static void __meminit
resize_zone_range(struct zone
*zone
, unsigned long start_pfn
,
714 unsigned long nr_pages
)
716 unsigned long old_end_pfn
= zone_end_pfn(zone
);
718 if (zone_is_empty(zone
) || start_pfn
< zone
->zone_start_pfn
)
719 zone
->zone_start_pfn
= start_pfn
;
721 zone
->spanned_pages
= max(start_pfn
+ nr_pages
, old_end_pfn
) - zone
->zone_start_pfn
;
724 static void __meminit
resize_pgdat_range(struct pglist_data
*pgdat
, unsigned long start_pfn
,
725 unsigned long nr_pages
)
727 unsigned long old_end_pfn
= pgdat_end_pfn(pgdat
);
729 if (!pgdat
->node_spanned_pages
|| start_pfn
< pgdat
->node_start_pfn
)
730 pgdat
->node_start_pfn
= start_pfn
;
732 pgdat
->node_spanned_pages
= max(start_pfn
+ nr_pages
, old_end_pfn
) - pgdat
->node_start_pfn
;
735 void __ref
move_pfn_range_to_zone(struct zone
*zone
, unsigned long start_pfn
,
736 unsigned long nr_pages
, struct vmem_altmap
*altmap
)
738 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
739 int nid
= pgdat
->node_id
;
742 clear_zone_contiguous(zone
);
744 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
745 pgdat_resize_lock(pgdat
, &flags
);
746 zone_span_writelock(zone
);
747 if (zone_is_empty(zone
))
748 init_currently_empty_zone(zone
, start_pfn
, nr_pages
);
749 resize_zone_range(zone
, start_pfn
, nr_pages
);
750 zone_span_writeunlock(zone
);
751 resize_pgdat_range(pgdat
, start_pfn
, nr_pages
);
752 pgdat_resize_unlock(pgdat
, &flags
);
755 * TODO now we have a visible range of pages which are not associated
756 * with their zone properly. Not nice but set_pfnblock_flags_mask
757 * expects the zone spans the pfn range. All the pages in the range
758 * are reserved so nobody should be touching them so we should be safe
760 memmap_init_zone(nr_pages
, nid
, zone_idx(zone
), start_pfn
,
761 MEMMAP_HOTPLUG
, altmap
);
763 set_zone_contiguous(zone
);
767 * Returns a default kernel memory zone for the given pfn range.
768 * If no kernel zone covers this pfn range it will automatically go
769 * to the ZONE_NORMAL.
771 static struct zone
*default_kernel_zone_for_pfn(int nid
, unsigned long start_pfn
,
772 unsigned long nr_pages
)
774 struct pglist_data
*pgdat
= NODE_DATA(nid
);
777 for (zid
= 0; zid
<= ZONE_NORMAL
; zid
++) {
778 struct zone
*zone
= &pgdat
->node_zones
[zid
];
780 if (zone_intersects(zone
, start_pfn
, nr_pages
))
784 return &pgdat
->node_zones
[ZONE_NORMAL
];
787 static inline struct zone
*default_zone_for_pfn(int nid
, unsigned long start_pfn
,
788 unsigned long nr_pages
)
790 struct zone
*kernel_zone
= default_kernel_zone_for_pfn(nid
, start_pfn
,
792 struct zone
*movable_zone
= &NODE_DATA(nid
)->node_zones
[ZONE_MOVABLE
];
793 bool in_kernel
= zone_intersects(kernel_zone
, start_pfn
, nr_pages
);
794 bool in_movable
= zone_intersects(movable_zone
, start_pfn
, nr_pages
);
797 * We inherit the existing zone in a simple case where zones do not
798 * overlap in the given range
800 if (in_kernel
^ in_movable
)
801 return (in_kernel
) ? kernel_zone
: movable_zone
;
804 * If the range doesn't belong to any zone or two zones overlap in the
805 * given range then we use movable zone only if movable_node is
806 * enabled because we always online to a kernel zone by default.
808 return movable_node_enabled
? movable_zone
: kernel_zone
;
811 struct zone
* zone_for_pfn_range(int online_type
, int nid
, unsigned start_pfn
,
812 unsigned long nr_pages
)
814 if (online_type
== MMOP_ONLINE_KERNEL
)
815 return default_kernel_zone_for_pfn(nid
, start_pfn
, nr_pages
);
817 if (online_type
== MMOP_ONLINE_MOVABLE
)
818 return &NODE_DATA(nid
)->node_zones
[ZONE_MOVABLE
];
820 return default_zone_for_pfn(nid
, start_pfn
, nr_pages
);
824 * Associates the given pfn range with the given node and the zone appropriate
825 * for the given online type.
827 static struct zone
* __meminit
move_pfn_range(int online_type
, int nid
,
828 unsigned long start_pfn
, unsigned long nr_pages
)
832 zone
= zone_for_pfn_range(online_type
, nid
, start_pfn
, nr_pages
);
833 move_pfn_range_to_zone(zone
, start_pfn
, nr_pages
, NULL
);
837 int __ref
online_pages(unsigned long pfn
, unsigned long nr_pages
, int online_type
)
840 unsigned long onlined_pages
= 0;
842 int need_zonelists_rebuild
= 0;
845 struct memory_notify arg
;
846 struct memory_block
*mem
;
851 * We can't use pfn_to_nid() because nid might be stored in struct page
852 * which is not yet initialized. Instead, we find nid from memory block.
854 mem
= find_memory_block(__pfn_to_section(pfn
));
856 put_device(&mem
->dev
);
858 /* associate pfn range with the zone */
859 zone
= move_pfn_range(online_type
, nid
, pfn
, nr_pages
);
862 arg
.nr_pages
= nr_pages
;
863 node_states_check_changes_online(nr_pages
, zone
, &arg
);
865 ret
= memory_notify(MEM_GOING_ONLINE
, &arg
);
866 ret
= notifier_to_errno(ret
);
868 goto failed_addition
;
871 * If this zone is not populated, then it is not in zonelist.
872 * This means the page allocator ignores this zone.
873 * So, zonelist must be updated after online.
875 if (!populated_zone(zone
)) {
876 need_zonelists_rebuild
= 1;
877 setup_zone_pageset(zone
);
880 ret
= walk_system_ram_range(pfn
, nr_pages
, &onlined_pages
,
883 if (need_zonelists_rebuild
)
884 zone_pcp_reset(zone
);
885 goto failed_addition
;
888 zone
->present_pages
+= onlined_pages
;
890 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
891 zone
->zone_pgdat
->node_present_pages
+= onlined_pages
;
892 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
897 node_states_set_node(nid
, &arg
);
898 if (need_zonelists_rebuild
)
899 build_all_zonelists(NULL
);
901 zone_pcp_update(zone
);
904 init_per_zone_wmark_min();
911 vm_total_pages
= nr_free_pagecache_pages();
913 writeback_set_ratelimit();
916 memory_notify(MEM_ONLINE
, &arg
);
921 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
922 (unsigned long long) pfn
<< PAGE_SHIFT
,
923 (((unsigned long long) pfn
+ nr_pages
) << PAGE_SHIFT
) - 1);
924 memory_notify(MEM_CANCEL_ONLINE
, &arg
);
928 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
930 static void reset_node_present_pages(pg_data_t
*pgdat
)
934 for (z
= pgdat
->node_zones
; z
< pgdat
->node_zones
+ MAX_NR_ZONES
; z
++)
935 z
->present_pages
= 0;
937 pgdat
->node_present_pages
= 0;
940 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
941 static pg_data_t __ref
*hotadd_new_pgdat(int nid
, u64 start
)
943 struct pglist_data
*pgdat
;
944 unsigned long start_pfn
= PFN_DOWN(start
);
946 pgdat
= NODE_DATA(nid
);
948 pgdat
= arch_alloc_nodedata(nid
);
952 arch_refresh_nodedata(nid
, pgdat
);
955 * Reset the nr_zones, order and classzone_idx before reuse.
956 * Note that kswapd will init kswapd_classzone_idx properly
957 * when it starts in the near future.
960 pgdat
->kswapd_order
= 0;
961 pgdat
->kswapd_classzone_idx
= 0;
964 /* we can use NODE_DATA(nid) from here */
966 pgdat
->node_id
= nid
;
967 pgdat
->node_start_pfn
= start_pfn
;
969 /* init node's zones as empty zones, we don't have any present pages.*/
970 free_area_init_core_hotplug(nid
);
971 pgdat
->per_cpu_nodestats
= alloc_percpu(struct per_cpu_nodestat
);
974 * The node we allocated has no zone fallback lists. For avoiding
975 * to access not-initialized zonelist, build here.
977 build_all_zonelists(pgdat
);
980 * When memory is hot-added, all the memory is in offline state. So
981 * clear all zones' present_pages because they will be updated in
982 * online_pages() and offline_pages().
984 reset_node_managed_pages(pgdat
);
985 reset_node_present_pages(pgdat
);
990 static void rollback_node_hotadd(int nid
)
992 pg_data_t
*pgdat
= NODE_DATA(nid
);
994 arch_refresh_nodedata(nid
, NULL
);
995 free_percpu(pgdat
->per_cpu_nodestats
);
996 arch_free_nodedata(pgdat
);
1002 * try_online_node - online a node if offlined
1004 * @start: start addr of the node
1005 * @set_node_online: Whether we want to online the node
1006 * called by cpu_up() to online a node without onlined memory.
1009 * 1 -> a new node has been allocated
1010 * 0 -> the node is already online
1011 * -ENOMEM -> the node could not be allocated
1013 static int __try_online_node(int nid
, u64 start
, bool set_node_online
)
1018 if (node_online(nid
))
1021 pgdat
= hotadd_new_pgdat(nid
, start
);
1023 pr_err("Cannot online node %d due to NULL pgdat\n", nid
);
1028 if (set_node_online
) {
1029 node_set_online(nid
);
1030 ret
= register_one_node(nid
);
1038 * Users of this function always want to online/register the node
1040 int try_online_node(int nid
)
1044 mem_hotplug_begin();
1045 ret
= __try_online_node(nid
, 0, true);
1050 static int check_hotplug_memory_range(u64 start
, u64 size
)
1052 unsigned long block_sz
= memory_block_size_bytes();
1053 u64 block_nr_pages
= block_sz
>> PAGE_SHIFT
;
1054 u64 nr_pages
= size
>> PAGE_SHIFT
;
1055 u64 start_pfn
= PFN_DOWN(start
);
1057 /* memory range must be block size aligned */
1058 if (!nr_pages
|| !IS_ALIGNED(start_pfn
, block_nr_pages
) ||
1059 !IS_ALIGNED(nr_pages
, block_nr_pages
)) {
1060 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1061 block_sz
, start
, size
);
1068 static int online_memory_block(struct memory_block
*mem
, void *arg
)
1070 return device_online(&mem
->dev
);
1074 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1075 * and online/offline operations (triggered e.g. by sysfs).
1077 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1079 int __ref
add_memory_resource(int nid
, struct resource
*res
)
1081 struct mhp_restrictions restrictions
= {
1082 .flags
= MHP_MEMBLOCK_API
,
1085 bool new_node
= false;
1089 size
= resource_size(res
);
1091 ret
= check_hotplug_memory_range(start
, size
);
1095 mem_hotplug_begin();
1098 * Add new range to memblock so that when hotadd_new_pgdat() is called
1099 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1100 * this new range and calculate total pages correctly. The range will
1101 * be removed at hot-remove time.
1103 memblock_add_node(start
, size
, nid
);
1105 ret
= __try_online_node(nid
, start
, false);
1110 /* call arch's memory hotadd */
1111 ret
= arch_add_memory(nid
, start
, size
, &restrictions
);
1116 /* If sysfs file of new node can't be created, cpu on the node
1117 * can't be hot-added. There is no rollback way now.
1118 * So, check by BUG_ON() to catch it reluctantly..
1119 * We online node here. We can't roll back from here.
1121 node_set_online(nid
);
1122 ret
= __register_one_node(nid
);
1126 /* link memory sections under this node.*/
1127 ret
= link_mem_sections(nid
, PFN_DOWN(start
), PFN_UP(start
+ size
- 1));
1130 /* create new memmap entry */
1131 firmware_map_add_hotplug(start
, start
+ size
, "System RAM");
1133 /* device_online() will take the lock when calling online_pages() */
1136 /* online pages if requested */
1137 if (memhp_auto_online
)
1138 walk_memory_range(PFN_DOWN(start
), PFN_UP(start
+ size
- 1),
1139 NULL
, online_memory_block
);
1143 /* rollback pgdat allocation and others */
1145 rollback_node_hotadd(nid
);
1146 memblock_remove(start
, size
);
1151 /* requires device_hotplug_lock, see add_memory_resource() */
1152 int __ref
__add_memory(int nid
, u64 start
, u64 size
)
1154 struct resource
*res
;
1157 res
= register_memory_resource(start
, size
);
1159 return PTR_ERR(res
);
1161 ret
= add_memory_resource(nid
, res
);
1163 release_memory_resource(res
);
1167 int add_memory(int nid
, u64 start
, u64 size
)
1171 lock_device_hotplug();
1172 rc
= __add_memory(nid
, start
, size
);
1173 unlock_device_hotplug();
1177 EXPORT_SYMBOL_GPL(add_memory
);
1179 #ifdef CONFIG_MEMORY_HOTREMOVE
1181 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1182 * set and the size of the free page is given by page_order(). Using this,
1183 * the function determines if the pageblock contains only free pages.
1184 * Due to buddy contraints, a free page at least the size of a pageblock will
1185 * be located at the start of the pageblock
1187 static inline int pageblock_free(struct page
*page
)
1189 return PageBuddy(page
) && page_order(page
) >= pageblock_order
;
1192 /* Return the pfn of the start of the next active pageblock after a given pfn */
1193 static unsigned long next_active_pageblock(unsigned long pfn
)
1195 struct page
*page
= pfn_to_page(pfn
);
1197 /* Ensure the starting page is pageblock-aligned */
1198 BUG_ON(pfn
& (pageblock_nr_pages
- 1));
1200 /* If the entire pageblock is free, move to the end of free page */
1201 if (pageblock_free(page
)) {
1203 /* be careful. we don't have locks, page_order can be changed.*/
1204 order
= page_order(page
);
1205 if ((order
< MAX_ORDER
) && (order
>= pageblock_order
))
1206 return pfn
+ (1 << order
);
1209 return pfn
+ pageblock_nr_pages
;
1212 static bool is_pageblock_removable_nolock(unsigned long pfn
)
1214 struct page
*page
= pfn_to_page(pfn
);
1218 * We have to be careful here because we are iterating over memory
1219 * sections which are not zone aware so we might end up outside of
1220 * the zone but still within the section.
1221 * We have to take care about the node as well. If the node is offline
1222 * its NODE_DATA will be NULL - see page_zone.
1224 if (!node_online(page_to_nid(page
)))
1227 zone
= page_zone(page
);
1228 pfn
= page_to_pfn(page
);
1229 if (!zone_spans_pfn(zone
, pfn
))
1232 return !has_unmovable_pages(zone
, page
, 0, MIGRATE_MOVABLE
, SKIP_HWPOISON
);
1235 /* Checks if this range of memory is likely to be hot-removable. */
1236 bool is_mem_section_removable(unsigned long start_pfn
, unsigned long nr_pages
)
1238 unsigned long end_pfn
, pfn
;
1240 end_pfn
= min(start_pfn
+ nr_pages
,
1241 zone_end_pfn(page_zone(pfn_to_page(start_pfn
))));
1243 /* Check the starting page of each pageblock within the range */
1244 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
= next_active_pageblock(pfn
)) {
1245 if (!is_pageblock_removable_nolock(pfn
))
1250 /* All pageblocks in the memory block are likely to be hot-removable */
1255 * Confirm all pages in a range [start, end) belong to the same zone.
1256 * When true, return its valid [start, end).
1258 int test_pages_in_a_zone(unsigned long start_pfn
, unsigned long end_pfn
,
1259 unsigned long *valid_start
, unsigned long *valid_end
)
1261 unsigned long pfn
, sec_end_pfn
;
1262 unsigned long start
, end
;
1263 struct zone
*zone
= NULL
;
1266 for (pfn
= start_pfn
, sec_end_pfn
= SECTION_ALIGN_UP(start_pfn
+ 1);
1268 pfn
= sec_end_pfn
, sec_end_pfn
+= PAGES_PER_SECTION
) {
1269 /* Make sure the memory section is present first */
1270 if (!present_section_nr(pfn_to_section_nr(pfn
)))
1272 for (; pfn
< sec_end_pfn
&& pfn
< end_pfn
;
1273 pfn
+= MAX_ORDER_NR_PAGES
) {
1275 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1276 while ((i
< MAX_ORDER_NR_PAGES
) &&
1277 !pfn_valid_within(pfn
+ i
))
1279 if (i
== MAX_ORDER_NR_PAGES
|| pfn
+ i
>= end_pfn
)
1281 /* Check if we got outside of the zone */
1282 if (zone
&& !zone_spans_pfn(zone
, pfn
+ i
))
1284 page
= pfn_to_page(pfn
+ i
);
1285 if (zone
&& page_zone(page
) != zone
)
1289 zone
= page_zone(page
);
1290 end
= pfn
+ MAX_ORDER_NR_PAGES
;
1295 *valid_start
= start
;
1296 *valid_end
= min(end
, end_pfn
);
1304 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1305 * non-lru movable pages and hugepages). We scan pfn because it's much
1306 * easier than scanning over linked list. This function returns the pfn
1307 * of the first found movable page if it's found, otherwise 0.
1309 static unsigned long scan_movable_pages(unsigned long start
, unsigned long end
)
1313 for (pfn
= start
; pfn
< end
; pfn
++) {
1314 struct page
*page
, *head
;
1317 if (!pfn_valid(pfn
))
1319 page
= pfn_to_page(pfn
);
1322 if (__PageMovable(page
))
1325 if (!PageHuge(page
))
1327 head
= compound_head(page
);
1328 if (page_huge_active(head
))
1330 skip
= (1 << compound_order(head
)) - (page
- head
);
1336 static struct page
*new_node_page(struct page
*page
, unsigned long private)
1338 int nid
= page_to_nid(page
);
1339 nodemask_t nmask
= node_states
[N_MEMORY
];
1342 * try to allocate from a different node but reuse this node if there
1343 * are no other online nodes to be used (e.g. we are offlining a part
1344 * of the only existing node)
1346 node_clear(nid
, nmask
);
1347 if (nodes_empty(nmask
))
1348 node_set(nid
, nmask
);
1350 return new_page_nodemask(page
, nid
, &nmask
);
1354 do_migrate_range(unsigned long start_pfn
, unsigned long end_pfn
)
1361 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
1362 if (!pfn_valid(pfn
))
1364 page
= pfn_to_page(pfn
);
1366 if (PageHuge(page
)) {
1367 struct page
*head
= compound_head(page
);
1368 pfn
= page_to_pfn(head
) + (1<<compound_order(head
)) - 1;
1369 isolate_huge_page(head
, &source
);
1371 } else if (PageTransHuge(page
))
1372 pfn
= page_to_pfn(compound_head(page
))
1373 + hpage_nr_pages(page
) - 1;
1376 * HWPoison pages have elevated reference counts so the migration would
1377 * fail on them. It also doesn't make any sense to migrate them in the
1378 * first place. Still try to unmap such a page in case it is still mapped
1379 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1380 * the unmap as the catch all safety net).
1382 if (PageHWPoison(page
)) {
1383 if (WARN_ON(PageLRU(page
)))
1384 isolate_lru_page(page
);
1385 if (page_mapped(page
))
1386 try_to_unmap(page
, TTU_IGNORE_MLOCK
| TTU_IGNORE_ACCESS
);
1390 if (!get_page_unless_zero(page
))
1393 * We can skip free pages. And we can deal with pages on
1394 * LRU and non-lru movable pages.
1397 ret
= isolate_lru_page(page
);
1399 ret
= isolate_movable_page(page
, ISOLATE_UNEVICTABLE
);
1400 if (!ret
) { /* Success */
1401 list_add_tail(&page
->lru
, &source
);
1402 if (!__PageMovable(page
))
1403 inc_node_page_state(page
, NR_ISOLATED_ANON
+
1404 page_is_file_cache(page
));
1407 pr_warn("failed to isolate pfn %lx\n", pfn
);
1408 dump_page(page
, "isolation failed");
1412 if (!list_empty(&source
)) {
1413 /* Allocate a new page from the nearest neighbor node */
1414 ret
= migrate_pages(&source
, new_node_page
, NULL
, 0,
1415 MIGRATE_SYNC
, MR_MEMORY_HOTPLUG
);
1417 list_for_each_entry(page
, &source
, lru
) {
1418 pr_warn("migrating pfn %lx failed ret:%d ",
1419 page_to_pfn(page
), ret
);
1420 dump_page(page
, "migration failure");
1422 putback_movable_pages(&source
);
1430 * remove from free_area[] and mark all as Reserved.
1433 offline_isolated_pages_cb(unsigned long start
, unsigned long nr_pages
,
1436 unsigned long *offlined_pages
= (unsigned long *)data
;
1438 *offlined_pages
+= __offline_isolated_pages(start
, start
+ nr_pages
);
1443 * Check all pages in range, recoreded as memory resource, are isolated.
1446 check_pages_isolated_cb(unsigned long start_pfn
, unsigned long nr_pages
,
1449 return test_pages_isolated(start_pfn
, start_pfn
+ nr_pages
, true);
1452 static int __init
cmdline_parse_movable_node(char *p
)
1454 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1455 movable_node_enabled
= true;
1457 pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1461 early_param("movable_node", cmdline_parse_movable_node
);
1463 /* check which state of node_states will be changed when offline memory */
1464 static void node_states_check_changes_offline(unsigned long nr_pages
,
1465 struct zone
*zone
, struct memory_notify
*arg
)
1467 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1468 unsigned long present_pages
= 0;
1471 arg
->status_change_nid
= NUMA_NO_NODE
;
1472 arg
->status_change_nid_normal
= NUMA_NO_NODE
;
1473 arg
->status_change_nid_high
= NUMA_NO_NODE
;
1476 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1477 * If the memory to be offline is within the range
1478 * [0..ZONE_NORMAL], and it is the last present memory there,
1479 * the zones in that range will become empty after the offlining,
1480 * thus we can determine that we need to clear the node from
1481 * node_states[N_NORMAL_MEMORY].
1483 for (zt
= 0; zt
<= ZONE_NORMAL
; zt
++)
1484 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1485 if (zone_idx(zone
) <= ZONE_NORMAL
&& nr_pages
>= present_pages
)
1486 arg
->status_change_nid_normal
= zone_to_nid(zone
);
1488 #ifdef CONFIG_HIGHMEM
1490 * node_states[N_HIGH_MEMORY] contains nodes which
1491 * have normal memory or high memory.
1492 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1493 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1494 * we determine that the zones in that range become empty,
1495 * we need to clear the node for N_HIGH_MEMORY.
1497 present_pages
+= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1498 if (zone_idx(zone
) <= ZONE_HIGHMEM
&& nr_pages
>= present_pages
)
1499 arg
->status_change_nid_high
= zone_to_nid(zone
);
1503 * We have accounted the pages from [0..ZONE_NORMAL), and
1504 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1506 * Here we count the possible pages from ZONE_MOVABLE.
1507 * If after having accounted all the pages, we see that the nr_pages
1508 * to be offlined is over or equal to the accounted pages,
1509 * we know that the node will become empty, and so, we can clear
1510 * it for N_MEMORY as well.
1512 present_pages
+= pgdat
->node_zones
[ZONE_MOVABLE
].present_pages
;
1514 if (nr_pages
>= present_pages
)
1515 arg
->status_change_nid
= zone_to_nid(zone
);
1518 static void node_states_clear_node(int node
, struct memory_notify
*arg
)
1520 if (arg
->status_change_nid_normal
>= 0)
1521 node_clear_state(node
, N_NORMAL_MEMORY
);
1523 if (arg
->status_change_nid_high
>= 0)
1524 node_clear_state(node
, N_HIGH_MEMORY
);
1526 if (arg
->status_change_nid
>= 0)
1527 node_clear_state(node
, N_MEMORY
);
1530 static int __ref
__offline_pages(unsigned long start_pfn
,
1531 unsigned long end_pfn
)
1533 unsigned long pfn
, nr_pages
;
1534 unsigned long offlined_pages
= 0;
1535 int ret
, node
, nr_isolate_pageblock
;
1536 unsigned long flags
;
1537 unsigned long valid_start
, valid_end
;
1539 struct memory_notify arg
;
1542 mem_hotplug_begin();
1544 /* This makes hotplug much easier...and readable.
1545 we assume this for now. .*/
1546 if (!test_pages_in_a_zone(start_pfn
, end_pfn
, &valid_start
,
1549 reason
= "multizone range";
1550 goto failed_removal
;
1553 zone
= page_zone(pfn_to_page(valid_start
));
1554 node
= zone_to_nid(zone
);
1555 nr_pages
= end_pfn
- start_pfn
;
1557 /* set above range as isolated */
1558 ret
= start_isolate_page_range(start_pfn
, end_pfn
,
1560 SKIP_HWPOISON
| REPORT_FAILURE
);
1562 reason
= "failure to isolate range";
1563 goto failed_removal
;
1565 nr_isolate_pageblock
= ret
;
1567 arg
.start_pfn
= start_pfn
;
1568 arg
.nr_pages
= nr_pages
;
1569 node_states_check_changes_offline(nr_pages
, zone
, &arg
);
1571 ret
= memory_notify(MEM_GOING_OFFLINE
, &arg
);
1572 ret
= notifier_to_errno(ret
);
1574 reason
= "notifier failure";
1575 goto failed_removal_isolated
;
1579 for (pfn
= start_pfn
; pfn
;) {
1580 if (signal_pending(current
)) {
1582 reason
= "signal backoff";
1583 goto failed_removal_isolated
;
1587 lru_add_drain_all();
1589 pfn
= scan_movable_pages(pfn
, end_pfn
);
1592 * TODO: fatal migration failures should bail
1595 do_migrate_range(pfn
, end_pfn
);
1600 * Dissolve free hugepages in the memory block before doing
1601 * offlining actually in order to make hugetlbfs's object
1602 * counting consistent.
1604 ret
= dissolve_free_huge_pages(start_pfn
, end_pfn
);
1606 reason
= "failure to dissolve huge pages";
1607 goto failed_removal_isolated
;
1610 ret
= walk_system_ram_range(start_pfn
, end_pfn
- start_pfn
,
1611 NULL
, check_pages_isolated_cb
);
1614 /* Ok, all of our target is isolated.
1615 We cannot do rollback at this point. */
1616 walk_system_ram_range(start_pfn
, end_pfn
- start_pfn
,
1617 &offlined_pages
, offline_isolated_pages_cb
);
1618 pr_info("Offlined Pages %ld\n", offlined_pages
);
1620 * Onlining will reset pagetype flags and makes migrate type
1621 * MOVABLE, so just need to decrease the number of isolated
1622 * pageblocks zone counter here.
1624 spin_lock_irqsave(&zone
->lock
, flags
);
1625 zone
->nr_isolate_pageblock
-= nr_isolate_pageblock
;
1626 spin_unlock_irqrestore(&zone
->lock
, flags
);
1628 /* removal success */
1629 adjust_managed_page_count(pfn_to_page(start_pfn
), -offlined_pages
);
1630 zone
->present_pages
-= offlined_pages
;
1632 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
1633 zone
->zone_pgdat
->node_present_pages
-= offlined_pages
;
1634 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
1636 init_per_zone_wmark_min();
1638 if (!populated_zone(zone
)) {
1639 zone_pcp_reset(zone
);
1640 build_all_zonelists(NULL
);
1642 zone_pcp_update(zone
);
1644 node_states_clear_node(node
, &arg
);
1645 if (arg
.status_change_nid
>= 0) {
1647 kcompactd_stop(node
);
1650 vm_total_pages
= nr_free_pagecache_pages();
1651 writeback_set_ratelimit();
1653 memory_notify(MEM_OFFLINE
, &arg
);
1657 failed_removal_isolated
:
1658 undo_isolate_page_range(start_pfn
, end_pfn
, MIGRATE_MOVABLE
);
1659 memory_notify(MEM_CANCEL_OFFLINE
, &arg
);
1661 pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1662 (unsigned long long) start_pfn
<< PAGE_SHIFT
,
1663 ((unsigned long long) end_pfn
<< PAGE_SHIFT
) - 1,
1665 /* pushback to free area */
1670 int offline_pages(unsigned long start_pfn
, unsigned long nr_pages
)
1672 return __offline_pages(start_pfn
, start_pfn
+ nr_pages
);
1674 #endif /* CONFIG_MEMORY_HOTREMOVE */
1677 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1678 * @start_pfn: start pfn of the memory range
1679 * @end_pfn: end pfn of the memory range
1680 * @arg: argument passed to func
1681 * @func: callback for each memory section walked
1683 * This function walks through all present mem sections in range
1684 * [start_pfn, end_pfn) and call func on each mem section.
1686 * Returns the return value of func.
1688 int walk_memory_range(unsigned long start_pfn
, unsigned long end_pfn
,
1689 void *arg
, int (*func
)(struct memory_block
*, void *))
1691 struct memory_block
*mem
= NULL
;
1692 struct mem_section
*section
;
1693 unsigned long pfn
, section_nr
;
1696 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
1697 section_nr
= pfn_to_section_nr(pfn
);
1698 if (!present_section_nr(section_nr
))
1701 section
= __nr_to_section(section_nr
);
1702 /* same memblock? */
1704 if ((section_nr
>= mem
->start_section_nr
) &&
1705 (section_nr
<= mem
->end_section_nr
))
1708 mem
= find_memory_block_hinted(section
, mem
);
1712 ret
= func(mem
, arg
);
1714 kobject_put(&mem
->dev
.kobj
);
1720 kobject_put(&mem
->dev
.kobj
);
1725 #ifdef CONFIG_MEMORY_HOTREMOVE
1726 static int check_memblock_offlined_cb(struct memory_block
*mem
, void *arg
)
1728 int ret
= !is_memblock_offlined(mem
);
1730 if (unlikely(ret
)) {
1731 phys_addr_t beginpa
, endpa
;
1733 beginpa
= PFN_PHYS(section_nr_to_pfn(mem
->start_section_nr
));
1734 endpa
= PFN_PHYS(section_nr_to_pfn(mem
->end_section_nr
+ 1))-1;
1735 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1743 static int check_cpu_on_node(pg_data_t
*pgdat
)
1747 for_each_present_cpu(cpu
) {
1748 if (cpu_to_node(cpu
) == pgdat
->node_id
)
1750 * the cpu on this node isn't removed, and we can't
1751 * offline this node.
1763 * Offline a node if all memory sections and cpus of the node are removed.
1765 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1766 * and online/offline operations before this call.
1768 void try_offline_node(int nid
)
1770 pg_data_t
*pgdat
= NODE_DATA(nid
);
1771 unsigned long start_pfn
= pgdat
->node_start_pfn
;
1772 unsigned long end_pfn
= start_pfn
+ pgdat
->node_spanned_pages
;
1775 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
1776 unsigned long section_nr
= pfn_to_section_nr(pfn
);
1778 if (!present_section_nr(section_nr
))
1781 if (pfn_to_nid(pfn
) != nid
)
1785 * some memory sections of this node are not removed, and we
1786 * can't offline node now.
1791 if (check_cpu_on_node(pgdat
))
1795 * all memory/cpu of this node are removed, we can offline this
1798 node_set_offline(nid
);
1799 unregister_one_node(nid
);
1801 EXPORT_SYMBOL(try_offline_node
);
1803 static void __release_memory_resource(resource_size_t start
,
1804 resource_size_t size
)
1809 * When removing memory in the same granularity as it was added,
1810 * this function never fails. It might only fail if resources
1811 * have to be adjusted or split. We'll ignore the error, as
1812 * removing of memory cannot fail.
1814 ret
= release_mem_region_adjustable(&iomem_resource
, start
, size
);
1816 resource_size_t endres
= start
+ size
- 1;
1818 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
1819 &start
, &endres
, ret
);
1823 static int __ref
try_remove_memory(int nid
, u64 start
, u64 size
)
1827 BUG_ON(check_hotplug_memory_range(start
, size
));
1829 mem_hotplug_begin();
1832 * All memory blocks must be offlined before removing memory. Check
1833 * whether all memory blocks in question are offline and return error
1834 * if this is not the case.
1836 rc
= walk_memory_range(PFN_DOWN(start
), PFN_UP(start
+ size
- 1), NULL
,
1837 check_memblock_offlined_cb
);
1841 /* remove memmap entry */
1842 firmware_map_remove(start
, start
+ size
, "System RAM");
1843 memblock_free(start
, size
);
1844 memblock_remove(start
, size
);
1846 arch_remove_memory(nid
, start
, size
, NULL
);
1847 __release_memory_resource(start
, size
);
1849 try_offline_node(nid
);
1859 * @start: physical address of the region to remove
1860 * @size: size of the region to remove
1862 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1863 * and online/offline operations before this call, as required by
1864 * try_offline_node().
1866 void __remove_memory(int nid
, u64 start
, u64 size
)
1870 * trigger BUG() is some memory is not offlined prior to calling this
1873 if (try_remove_memory(nid
, start
, size
))
1878 * Remove memory if every memory block is offline, otherwise return -EBUSY is
1879 * some memory is not offline
1881 int remove_memory(int nid
, u64 start
, u64 size
)
1885 lock_device_hotplug();
1886 rc
= try_remove_memory(nid
, start
, size
);
1887 unlock_device_hotplug();
1891 EXPORT_SYMBOL_GPL(remove_memory
);
1892 #endif /* CONFIG_MEMORY_HOTREMOVE */