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 online_page_callback_t online_page_callback
= generic_online_page
;
53 static DEFINE_MUTEX(online_page_callback_lock
);
55 DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock
);
57 void get_online_mems(void)
59 percpu_down_read(&mem_hotplug_lock
);
62 void put_online_mems(void)
64 percpu_up_read(&mem_hotplug_lock
);
67 bool movable_node_enabled
= false;
69 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
70 int memhp_default_online_type
= MMOP_OFFLINE
;
72 int memhp_default_online_type
= MMOP_ONLINE
;
75 static int __init
setup_memhp_default_state(char *str
)
77 const int online_type
= memhp_online_type_from_str(str
);
80 memhp_default_online_type
= online_type
;
84 __setup("memhp_default_state=", setup_memhp_default_state
);
86 void mem_hotplug_begin(void)
89 percpu_down_write(&mem_hotplug_lock
);
92 void mem_hotplug_done(void)
94 percpu_up_write(&mem_hotplug_lock
);
98 u64 max_mem_size
= U64_MAX
;
100 /* add this memory to iomem resource */
101 static struct resource
*register_memory_resource(u64 start
, u64 size
)
103 struct resource
*res
;
104 unsigned long flags
= IORESOURCE_SYSTEM_RAM
| IORESOURCE_BUSY
;
105 char *resource_name
= "System RAM";
108 * Make sure value parsed from 'mem=' only restricts memory adding
109 * while booting, so that memory hotplug won't be impacted. Please
110 * refer to document of 'mem=' in kernel-parameters.txt for more
113 if (start
+ size
> max_mem_size
&& system_state
< SYSTEM_RUNNING
)
114 return ERR_PTR(-E2BIG
);
117 * Request ownership of the new memory range. This might be
118 * a child of an existing resource that was present but
119 * not marked as busy.
121 res
= __request_region(&iomem_resource
, start
, size
,
122 resource_name
, flags
);
125 pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n",
126 start
, start
+ size
);
127 return ERR_PTR(-EEXIST
);
132 static void release_memory_resource(struct resource
*res
)
136 release_resource(res
);
140 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
141 void get_page_bootmem(unsigned long info
, struct page
*page
,
144 page
->freelist
= (void *)type
;
145 SetPagePrivate(page
);
146 set_page_private(page
, info
);
150 void put_page_bootmem(struct page
*page
)
154 type
= (unsigned long) page
->freelist
;
155 BUG_ON(type
< MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE
||
156 type
> MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE
);
158 if (page_ref_dec_return(page
) == 1) {
159 page
->freelist
= NULL
;
160 ClearPagePrivate(page
);
161 set_page_private(page
, 0);
162 INIT_LIST_HEAD(&page
->lru
);
163 free_reserved_page(page
);
167 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
168 #ifndef CONFIG_SPARSEMEM_VMEMMAP
169 static void register_page_bootmem_info_section(unsigned long start_pfn
)
171 unsigned long mapsize
, section_nr
, i
;
172 struct mem_section
*ms
;
173 struct page
*page
, *memmap
;
174 struct mem_section_usage
*usage
;
176 section_nr
= pfn_to_section_nr(start_pfn
);
177 ms
= __nr_to_section(section_nr
);
179 /* Get section's memmap address */
180 memmap
= sparse_decode_mem_map(ms
->section_mem_map
, section_nr
);
183 * Get page for the memmap's phys address
184 * XXX: need more consideration for sparse_vmemmap...
186 page
= virt_to_page(memmap
);
187 mapsize
= sizeof(struct page
) * PAGES_PER_SECTION
;
188 mapsize
= PAGE_ALIGN(mapsize
) >> PAGE_SHIFT
;
190 /* remember memmap's page */
191 for (i
= 0; i
< mapsize
; i
++, page
++)
192 get_page_bootmem(section_nr
, page
, SECTION_INFO
);
195 page
= virt_to_page(usage
);
197 mapsize
= PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT
;
199 for (i
= 0; i
< mapsize
; i
++, page
++)
200 get_page_bootmem(section_nr
, page
, MIX_SECTION_INFO
);
203 #else /* CONFIG_SPARSEMEM_VMEMMAP */
204 static void register_page_bootmem_info_section(unsigned long start_pfn
)
206 unsigned long mapsize
, section_nr
, i
;
207 struct mem_section
*ms
;
208 struct page
*page
, *memmap
;
209 struct mem_section_usage
*usage
;
211 section_nr
= pfn_to_section_nr(start_pfn
);
212 ms
= __nr_to_section(section_nr
);
214 memmap
= sparse_decode_mem_map(ms
->section_mem_map
, section_nr
);
216 register_page_bootmem_memmap(section_nr
, memmap
, PAGES_PER_SECTION
);
219 page
= virt_to_page(usage
);
221 mapsize
= PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT
;
223 for (i
= 0; i
< mapsize
; i
++, page
++)
224 get_page_bootmem(section_nr
, page
, MIX_SECTION_INFO
);
226 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
228 void __init
register_page_bootmem_info_node(struct pglist_data
*pgdat
)
230 unsigned long i
, pfn
, end_pfn
, nr_pages
;
231 int node
= pgdat
->node_id
;
234 nr_pages
= PAGE_ALIGN(sizeof(struct pglist_data
)) >> PAGE_SHIFT
;
235 page
= virt_to_page(pgdat
);
237 for (i
= 0; i
< nr_pages
; i
++, page
++)
238 get_page_bootmem(node
, page
, NODE_INFO
);
240 pfn
= pgdat
->node_start_pfn
;
241 end_pfn
= pgdat_end_pfn(pgdat
);
243 /* register section info */
244 for (; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
246 * Some platforms can assign the same pfn to multiple nodes - on
247 * node0 as well as nodeN. To avoid registering a pfn against
248 * multiple nodes we check that this pfn does not already
249 * reside in some other nodes.
251 if (pfn_valid(pfn
) && (early_pfn_to_nid(pfn
) == node
))
252 register_page_bootmem_info_section(pfn
);
255 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
257 static int check_pfn_span(unsigned long pfn
, unsigned long nr_pages
,
261 * Disallow all operations smaller than a sub-section and only
262 * allow operations smaller than a section for
263 * SPARSEMEM_VMEMMAP. Note that check_hotplug_memory_range()
264 * enforces a larger memory_block_size_bytes() granularity for
265 * memory that will be marked online, so this check should only
266 * fire for direct arch_{add,remove}_memory() users outside of
267 * add_memory_resource().
269 unsigned long min_align
;
271 if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP
))
272 min_align
= PAGES_PER_SUBSECTION
;
274 min_align
= PAGES_PER_SECTION
;
275 if (!IS_ALIGNED(pfn
, min_align
)
276 || !IS_ALIGNED(nr_pages
, min_align
)) {
277 WARN(1, "Misaligned __%s_pages start: %#lx end: #%lx\n",
278 reason
, pfn
, pfn
+ nr_pages
- 1);
284 static int check_hotplug_memory_addressable(unsigned long pfn
,
285 unsigned long nr_pages
)
287 const u64 max_addr
= PFN_PHYS(pfn
+ nr_pages
) - 1;
289 if (max_addr
>> MAX_PHYSMEM_BITS
) {
290 const u64 max_allowed
= (1ull << (MAX_PHYSMEM_BITS
+ 1)) - 1;
292 "Hotplugged memory exceeds maximum addressable address, range=%#llx-%#llx, maximum=%#llx\n",
293 (u64
)PFN_PHYS(pfn
), max_addr
, max_allowed
);
301 * Reasonably generic function for adding memory. It is
302 * expected that archs that support memory hotplug will
303 * call this function after deciding the zone to which to
306 int __ref
__add_pages(int nid
, unsigned long pfn
, unsigned long nr_pages
,
307 struct mhp_restrictions
*restrictions
)
309 const unsigned long end_pfn
= pfn
+ nr_pages
;
310 unsigned long cur_nr_pages
;
312 struct vmem_altmap
*altmap
= restrictions
->altmap
;
314 err
= check_hotplug_memory_addressable(pfn
, nr_pages
);
320 * Validate altmap is within bounds of the total request
322 if (altmap
->base_pfn
!= pfn
323 || vmem_altmap_offset(altmap
) > nr_pages
) {
324 pr_warn_once("memory add fail, invalid altmap\n");
330 err
= check_pfn_span(pfn
, nr_pages
, "add");
334 for (; pfn
< end_pfn
; pfn
+= cur_nr_pages
) {
335 /* Select all remaining pages up to the next section boundary */
336 cur_nr_pages
= min(end_pfn
- pfn
,
337 SECTION_ALIGN_UP(pfn
+ 1) - pfn
);
338 err
= sparse_add_section(nid
, pfn
, cur_nr_pages
, altmap
);
343 vmemmap_populate_print_last();
347 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
348 static unsigned long find_smallest_section_pfn(int nid
, struct zone
*zone
,
349 unsigned long start_pfn
,
350 unsigned long end_pfn
)
352 for (; start_pfn
< end_pfn
; start_pfn
+= PAGES_PER_SUBSECTION
) {
353 if (unlikely(!pfn_to_online_page(start_pfn
)))
356 if (unlikely(pfn_to_nid(start_pfn
) != nid
))
359 if (zone
!= page_zone(pfn_to_page(start_pfn
)))
368 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
369 static unsigned long find_biggest_section_pfn(int nid
, struct zone
*zone
,
370 unsigned long start_pfn
,
371 unsigned long end_pfn
)
375 /* pfn is the end pfn of a memory section. */
377 for (; pfn
>= start_pfn
; pfn
-= PAGES_PER_SUBSECTION
) {
378 if (unlikely(!pfn_to_online_page(pfn
)))
381 if (unlikely(pfn_to_nid(pfn
) != nid
))
384 if (zone
!= page_zone(pfn_to_page(pfn
)))
393 static void shrink_zone_span(struct zone
*zone
, unsigned long start_pfn
,
394 unsigned long end_pfn
)
397 int nid
= zone_to_nid(zone
);
399 zone_span_writelock(zone
);
400 if (zone
->zone_start_pfn
== start_pfn
) {
402 * If the section is smallest section in the zone, it need
403 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
404 * In this case, we find second smallest valid mem_section
405 * for shrinking zone.
407 pfn
= find_smallest_section_pfn(nid
, zone
, end_pfn
,
410 zone
->spanned_pages
= zone_end_pfn(zone
) - pfn
;
411 zone
->zone_start_pfn
= pfn
;
413 zone
->zone_start_pfn
= 0;
414 zone
->spanned_pages
= 0;
416 } else if (zone_end_pfn(zone
) == end_pfn
) {
418 * If the section is biggest section in the zone, it need
419 * shrink zone->spanned_pages.
420 * In this case, we find second biggest valid mem_section for
423 pfn
= find_biggest_section_pfn(nid
, zone
, zone
->zone_start_pfn
,
426 zone
->spanned_pages
= pfn
- zone
->zone_start_pfn
+ 1;
428 zone
->zone_start_pfn
= 0;
429 zone
->spanned_pages
= 0;
432 zone_span_writeunlock(zone
);
435 static void update_pgdat_span(struct pglist_data
*pgdat
)
437 unsigned long node_start_pfn
= 0, node_end_pfn
= 0;
440 for (zone
= pgdat
->node_zones
;
441 zone
< pgdat
->node_zones
+ MAX_NR_ZONES
; zone
++) {
442 unsigned long zone_end_pfn
= zone
->zone_start_pfn
+
445 /* No need to lock the zones, they can't change. */
446 if (!zone
->spanned_pages
)
449 node_start_pfn
= zone
->zone_start_pfn
;
450 node_end_pfn
= zone_end_pfn
;
454 if (zone_end_pfn
> node_end_pfn
)
455 node_end_pfn
= zone_end_pfn
;
456 if (zone
->zone_start_pfn
< node_start_pfn
)
457 node_start_pfn
= zone
->zone_start_pfn
;
460 pgdat
->node_start_pfn
= node_start_pfn
;
461 pgdat
->node_spanned_pages
= node_end_pfn
- node_start_pfn
;
464 void __ref
remove_pfn_range_from_zone(struct zone
*zone
,
465 unsigned long start_pfn
,
466 unsigned long nr_pages
)
468 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
471 /* Poison struct pages because they are now uninitialized again. */
472 page_init_poison(pfn_to_page(start_pfn
), sizeof(struct page
) * nr_pages
);
474 #ifdef CONFIG_ZONE_DEVICE
476 * Zone shrinking code cannot properly deal with ZONE_DEVICE. So
477 * we will not try to shrink the zones - which is okay as
478 * set_zone_contiguous() cannot deal with ZONE_DEVICE either way.
480 if (zone_idx(zone
) == ZONE_DEVICE
)
484 clear_zone_contiguous(zone
);
486 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
487 shrink_zone_span(zone
, start_pfn
, start_pfn
+ nr_pages
);
488 update_pgdat_span(pgdat
);
489 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
491 set_zone_contiguous(zone
);
494 static void __remove_section(unsigned long pfn
, unsigned long nr_pages
,
495 unsigned long map_offset
,
496 struct vmem_altmap
*altmap
)
498 struct mem_section
*ms
= __pfn_to_section(pfn
);
500 if (WARN_ON_ONCE(!valid_section(ms
)))
503 sparse_remove_section(ms
, pfn
, nr_pages
, map_offset
, altmap
);
507 * __remove_pages() - remove sections of pages
508 * @pfn: starting pageframe (must be aligned to start of a section)
509 * @nr_pages: number of pages to remove (must be multiple of section size)
510 * @altmap: alternative device page map or %NULL if default memmap is used
512 * Generic helper function to remove section mappings and sysfs entries
513 * for the section of the memory we are removing. Caller needs to make
514 * sure that pages are marked reserved and zones are adjust properly by
515 * calling offline_pages().
517 void __remove_pages(unsigned long pfn
, unsigned long nr_pages
,
518 struct vmem_altmap
*altmap
)
520 const unsigned long end_pfn
= pfn
+ nr_pages
;
521 unsigned long cur_nr_pages
;
522 unsigned long map_offset
= 0;
524 map_offset
= vmem_altmap_offset(altmap
);
526 if (check_pfn_span(pfn
, nr_pages
, "remove"))
529 for (; pfn
< end_pfn
; pfn
+= cur_nr_pages
) {
531 /* Select all remaining pages up to the next section boundary */
532 cur_nr_pages
= min(end_pfn
- pfn
,
533 SECTION_ALIGN_UP(pfn
+ 1) - pfn
);
534 __remove_section(pfn
, cur_nr_pages
, map_offset
, altmap
);
539 int set_online_page_callback(online_page_callback_t callback
)
544 mutex_lock(&online_page_callback_lock
);
546 if (online_page_callback
== generic_online_page
) {
547 online_page_callback
= callback
;
551 mutex_unlock(&online_page_callback_lock
);
556 EXPORT_SYMBOL_GPL(set_online_page_callback
);
558 int restore_online_page_callback(online_page_callback_t callback
)
563 mutex_lock(&online_page_callback_lock
);
565 if (online_page_callback
== callback
) {
566 online_page_callback
= generic_online_page
;
570 mutex_unlock(&online_page_callback_lock
);
575 EXPORT_SYMBOL_GPL(restore_online_page_callback
);
577 void generic_online_page(struct page
*page
, unsigned int order
)
580 * Freeing the page with debug_pagealloc enabled will try to unmap it,
581 * so we should map it first. This is better than introducing a special
582 * case in page freeing fast path.
584 if (debug_pagealloc_enabled_static())
585 kernel_map_pages(page
, 1 << order
, 1);
586 __free_pages_core(page
, order
);
587 totalram_pages_add(1UL << order
);
588 #ifdef CONFIG_HIGHMEM
589 if (PageHighMem(page
))
590 totalhigh_pages_add(1UL << order
);
593 EXPORT_SYMBOL_GPL(generic_online_page
);
595 static int online_pages_range(unsigned long start_pfn
, unsigned long nr_pages
,
598 const unsigned long end_pfn
= start_pfn
+ nr_pages
;
603 * Online the pages. The callback might decide to keep some pages
604 * PG_reserved (to add them to the buddy later), but we still account
605 * them as being online/belonging to this zone ("present").
607 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= 1ul << order
) {
608 order
= min(MAX_ORDER
- 1, get_order(PFN_PHYS(end_pfn
- pfn
)));
609 /* __free_pages_core() wants pfns to be aligned to the order */
610 if (WARN_ON_ONCE(!IS_ALIGNED(pfn
, 1ul << order
)))
612 (*online_page_callback
)(pfn_to_page(pfn
), order
);
615 /* mark all involved sections as online */
616 online_mem_sections(start_pfn
, end_pfn
);
618 *(unsigned long *)arg
+= nr_pages
;
622 /* check which state of node_states will be changed when online memory */
623 static void node_states_check_changes_online(unsigned long nr_pages
,
624 struct zone
*zone
, struct memory_notify
*arg
)
626 int nid
= zone_to_nid(zone
);
628 arg
->status_change_nid
= NUMA_NO_NODE
;
629 arg
->status_change_nid_normal
= NUMA_NO_NODE
;
630 arg
->status_change_nid_high
= NUMA_NO_NODE
;
632 if (!node_state(nid
, N_MEMORY
))
633 arg
->status_change_nid
= nid
;
634 if (zone_idx(zone
) <= ZONE_NORMAL
&& !node_state(nid
, N_NORMAL_MEMORY
))
635 arg
->status_change_nid_normal
= nid
;
636 #ifdef CONFIG_HIGHMEM
637 if (zone_idx(zone
) <= ZONE_HIGHMEM
&& !node_state(nid
, N_HIGH_MEMORY
))
638 arg
->status_change_nid_high
= nid
;
642 static void node_states_set_node(int node
, struct memory_notify
*arg
)
644 if (arg
->status_change_nid_normal
>= 0)
645 node_set_state(node
, N_NORMAL_MEMORY
);
647 if (arg
->status_change_nid_high
>= 0)
648 node_set_state(node
, N_HIGH_MEMORY
);
650 if (arg
->status_change_nid
>= 0)
651 node_set_state(node
, N_MEMORY
);
654 static void __meminit
resize_zone_range(struct zone
*zone
, unsigned long start_pfn
,
655 unsigned long nr_pages
)
657 unsigned long old_end_pfn
= zone_end_pfn(zone
);
659 if (zone_is_empty(zone
) || start_pfn
< zone
->zone_start_pfn
)
660 zone
->zone_start_pfn
= start_pfn
;
662 zone
->spanned_pages
= max(start_pfn
+ nr_pages
, old_end_pfn
) - zone
->zone_start_pfn
;
665 static void __meminit
resize_pgdat_range(struct pglist_data
*pgdat
, unsigned long start_pfn
,
666 unsigned long nr_pages
)
668 unsigned long old_end_pfn
= pgdat_end_pfn(pgdat
);
670 if (!pgdat
->node_spanned_pages
|| start_pfn
< pgdat
->node_start_pfn
)
671 pgdat
->node_start_pfn
= start_pfn
;
673 pgdat
->node_spanned_pages
= max(start_pfn
+ nr_pages
, old_end_pfn
) - pgdat
->node_start_pfn
;
677 * Associate the pfn range with the given zone, initializing the memmaps
678 * and resizing the pgdat/zone data to span the added pages. After this
679 * call, all affected pages are PG_reserved.
681 void __ref
move_pfn_range_to_zone(struct zone
*zone
, unsigned long start_pfn
,
682 unsigned long nr_pages
, struct vmem_altmap
*altmap
)
684 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
685 int nid
= pgdat
->node_id
;
688 clear_zone_contiguous(zone
);
690 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
691 pgdat_resize_lock(pgdat
, &flags
);
692 zone_span_writelock(zone
);
693 if (zone_is_empty(zone
))
694 init_currently_empty_zone(zone
, start_pfn
, nr_pages
);
695 resize_zone_range(zone
, start_pfn
, nr_pages
);
696 zone_span_writeunlock(zone
);
697 resize_pgdat_range(pgdat
, start_pfn
, nr_pages
);
698 pgdat_resize_unlock(pgdat
, &flags
);
701 * TODO now we have a visible range of pages which are not associated
702 * with their zone properly. Not nice but set_pfnblock_flags_mask
703 * expects the zone spans the pfn range. All the pages in the range
704 * are reserved so nobody should be touching them so we should be safe
706 memmap_init_zone(nr_pages
, nid
, zone_idx(zone
), start_pfn
,
707 MEMMAP_HOTPLUG
, altmap
);
709 set_zone_contiguous(zone
);
713 * Returns a default kernel memory zone for the given pfn range.
714 * If no kernel zone covers this pfn range it will automatically go
715 * to the ZONE_NORMAL.
717 static struct zone
*default_kernel_zone_for_pfn(int nid
, unsigned long start_pfn
,
718 unsigned long nr_pages
)
720 struct pglist_data
*pgdat
= NODE_DATA(nid
);
723 for (zid
= 0; zid
<= ZONE_NORMAL
; zid
++) {
724 struct zone
*zone
= &pgdat
->node_zones
[zid
];
726 if (zone_intersects(zone
, start_pfn
, nr_pages
))
730 return &pgdat
->node_zones
[ZONE_NORMAL
];
733 static inline struct zone
*default_zone_for_pfn(int nid
, unsigned long start_pfn
,
734 unsigned long nr_pages
)
736 struct zone
*kernel_zone
= default_kernel_zone_for_pfn(nid
, start_pfn
,
738 struct zone
*movable_zone
= &NODE_DATA(nid
)->node_zones
[ZONE_MOVABLE
];
739 bool in_kernel
= zone_intersects(kernel_zone
, start_pfn
, nr_pages
);
740 bool in_movable
= zone_intersects(movable_zone
, start_pfn
, nr_pages
);
743 * We inherit the existing zone in a simple case where zones do not
744 * overlap in the given range
746 if (in_kernel
^ in_movable
)
747 return (in_kernel
) ? kernel_zone
: movable_zone
;
750 * If the range doesn't belong to any zone or two zones overlap in the
751 * given range then we use movable zone only if movable_node is
752 * enabled because we always online to a kernel zone by default.
754 return movable_node_enabled
? movable_zone
: kernel_zone
;
757 struct zone
* zone_for_pfn_range(int online_type
, int nid
, unsigned start_pfn
,
758 unsigned long nr_pages
)
760 if (online_type
== MMOP_ONLINE_KERNEL
)
761 return default_kernel_zone_for_pfn(nid
, start_pfn
, nr_pages
);
763 if (online_type
== MMOP_ONLINE_MOVABLE
)
764 return &NODE_DATA(nid
)->node_zones
[ZONE_MOVABLE
];
766 return default_zone_for_pfn(nid
, start_pfn
, nr_pages
);
769 int __ref
online_pages(unsigned long pfn
, unsigned long nr_pages
,
770 int online_type
, int nid
)
773 unsigned long onlined_pages
= 0;
775 int need_zonelists_rebuild
= 0;
777 struct memory_notify arg
;
781 /* associate pfn range with the zone */
782 zone
= zone_for_pfn_range(online_type
, nid
, pfn
, nr_pages
);
783 move_pfn_range_to_zone(zone
, pfn
, nr_pages
, NULL
);
786 arg
.nr_pages
= nr_pages
;
787 node_states_check_changes_online(nr_pages
, zone
, &arg
);
789 ret
= memory_notify(MEM_GOING_ONLINE
, &arg
);
790 ret
= notifier_to_errno(ret
);
792 goto failed_addition
;
795 * If this zone is not populated, then it is not in zonelist.
796 * This means the page allocator ignores this zone.
797 * So, zonelist must be updated after online.
799 if (!populated_zone(zone
)) {
800 need_zonelists_rebuild
= 1;
801 setup_zone_pageset(zone
);
804 ret
= walk_system_ram_range(pfn
, nr_pages
, &onlined_pages
,
807 /* not a single memory resource was applicable */
808 if (need_zonelists_rebuild
)
809 zone_pcp_reset(zone
);
810 goto failed_addition
;
813 zone
->present_pages
+= onlined_pages
;
815 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
816 zone
->zone_pgdat
->node_present_pages
+= onlined_pages
;
817 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
821 node_states_set_node(nid
, &arg
);
822 if (need_zonelists_rebuild
)
823 build_all_zonelists(NULL
);
825 zone_pcp_update(zone
);
827 init_per_zone_wmark_min();
832 vm_total_pages
= nr_free_pagecache_pages();
834 writeback_set_ratelimit();
836 memory_notify(MEM_ONLINE
, &arg
);
841 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
842 (unsigned long long) pfn
<< PAGE_SHIFT
,
843 (((unsigned long long) pfn
+ nr_pages
) << PAGE_SHIFT
) - 1);
844 memory_notify(MEM_CANCEL_ONLINE
, &arg
);
845 remove_pfn_range_from_zone(zone
, pfn
, nr_pages
);
849 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
851 static void reset_node_present_pages(pg_data_t
*pgdat
)
855 for (z
= pgdat
->node_zones
; z
< pgdat
->node_zones
+ MAX_NR_ZONES
; z
++)
856 z
->present_pages
= 0;
858 pgdat
->node_present_pages
= 0;
861 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
862 static pg_data_t __ref
*hotadd_new_pgdat(int nid
, u64 start
)
864 struct pglist_data
*pgdat
;
865 unsigned long start_pfn
= PFN_DOWN(start
);
867 pgdat
= NODE_DATA(nid
);
869 pgdat
= arch_alloc_nodedata(nid
);
873 pgdat
->per_cpu_nodestats
=
874 alloc_percpu(struct per_cpu_nodestat
);
875 arch_refresh_nodedata(nid
, pgdat
);
879 * Reset the nr_zones, order and classzone_idx before reuse.
880 * Note that kswapd will init kswapd_classzone_idx properly
881 * when it starts in the near future.
884 pgdat
->kswapd_order
= 0;
885 pgdat
->kswapd_classzone_idx
= 0;
886 for_each_online_cpu(cpu
) {
887 struct per_cpu_nodestat
*p
;
889 p
= per_cpu_ptr(pgdat
->per_cpu_nodestats
, cpu
);
890 memset(p
, 0, sizeof(*p
));
894 /* we can use NODE_DATA(nid) from here */
896 pgdat
->node_id
= nid
;
897 pgdat
->node_start_pfn
= start_pfn
;
899 /* init node's zones as empty zones, we don't have any present pages.*/
900 free_area_init_core_hotplug(nid
);
903 * The node we allocated has no zone fallback lists. For avoiding
904 * to access not-initialized zonelist, build here.
906 build_all_zonelists(pgdat
);
909 * When memory is hot-added, all the memory is in offline state. So
910 * clear all zones' present_pages because they will be updated in
911 * online_pages() and offline_pages().
913 reset_node_managed_pages(pgdat
);
914 reset_node_present_pages(pgdat
);
919 static void rollback_node_hotadd(int nid
)
921 pg_data_t
*pgdat
= NODE_DATA(nid
);
923 arch_refresh_nodedata(nid
, NULL
);
924 free_percpu(pgdat
->per_cpu_nodestats
);
925 arch_free_nodedata(pgdat
);
930 * try_online_node - online a node if offlined
932 * @start: start addr of the node
933 * @set_node_online: Whether we want to online the node
934 * called by cpu_up() to online a node without onlined memory.
937 * 1 -> a new node has been allocated
938 * 0 -> the node is already online
939 * -ENOMEM -> the node could not be allocated
941 static int __try_online_node(int nid
, u64 start
, bool set_node_online
)
946 if (node_online(nid
))
949 pgdat
= hotadd_new_pgdat(nid
, start
);
951 pr_err("Cannot online node %d due to NULL pgdat\n", nid
);
956 if (set_node_online
) {
957 node_set_online(nid
);
958 ret
= register_one_node(nid
);
966 * Users of this function always want to online/register the node
968 int try_online_node(int nid
)
973 ret
= __try_online_node(nid
, 0, true);
978 static int check_hotplug_memory_range(u64 start
, u64 size
)
980 /* memory range must be block size aligned */
981 if (!size
|| !IS_ALIGNED(start
, memory_block_size_bytes()) ||
982 !IS_ALIGNED(size
, memory_block_size_bytes())) {
983 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
984 memory_block_size_bytes(), start
, size
);
991 static int online_memory_block(struct memory_block
*mem
, void *arg
)
993 mem
->online_type
= memhp_default_online_type
;
994 return device_online(&mem
->dev
);
998 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
999 * and online/offline operations (triggered e.g. by sysfs).
1001 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1003 int __ref
add_memory_resource(int nid
, struct resource
*res
)
1005 struct mhp_restrictions restrictions
= {};
1007 bool new_node
= false;
1011 size
= resource_size(res
);
1013 ret
= check_hotplug_memory_range(start
, size
);
1017 mem_hotplug_begin();
1020 * Add new range to memblock so that when hotadd_new_pgdat() is called
1021 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1022 * this new range and calculate total pages correctly. The range will
1023 * be removed at hot-remove time.
1025 memblock_add_node(start
, size
, nid
);
1027 ret
= __try_online_node(nid
, start
, false);
1032 /* call arch's memory hotadd */
1033 ret
= arch_add_memory(nid
, start
, size
, &restrictions
);
1037 /* create memory block devices after memory was added */
1038 ret
= create_memory_block_devices(start
, size
);
1040 arch_remove_memory(nid
, start
, size
, NULL
);
1045 /* If sysfs file of new node can't be created, cpu on the node
1046 * can't be hot-added. There is no rollback way now.
1047 * So, check by BUG_ON() to catch it reluctantly..
1048 * We online node here. We can't roll back from here.
1050 node_set_online(nid
);
1051 ret
= __register_one_node(nid
);
1055 /* link memory sections under this node.*/
1056 ret
= link_mem_sections(nid
, PFN_DOWN(start
), PFN_UP(start
+ size
- 1));
1059 /* create new memmap entry */
1060 firmware_map_add_hotplug(start
, start
+ size
, "System RAM");
1062 /* device_online() will take the lock when calling online_pages() */
1065 /* online pages if requested */
1066 if (memhp_default_online_type
!= MMOP_OFFLINE
)
1067 walk_memory_blocks(start
, size
, NULL
, online_memory_block
);
1071 /* rollback pgdat allocation and others */
1073 rollback_node_hotadd(nid
);
1074 memblock_remove(start
, size
);
1079 /* requires device_hotplug_lock, see add_memory_resource() */
1080 int __ref
__add_memory(int nid
, u64 start
, u64 size
)
1082 struct resource
*res
;
1085 res
= register_memory_resource(start
, size
);
1087 return PTR_ERR(res
);
1089 ret
= add_memory_resource(nid
, res
);
1091 release_memory_resource(res
);
1095 int add_memory(int nid
, u64 start
, u64 size
)
1099 lock_device_hotplug();
1100 rc
= __add_memory(nid
, start
, size
);
1101 unlock_device_hotplug();
1105 EXPORT_SYMBOL_GPL(add_memory
);
1107 #ifdef CONFIG_MEMORY_HOTREMOVE
1109 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1110 * set and the size of the free page is given by page_order(). Using this,
1111 * the function determines if the pageblock contains only free pages.
1112 * Due to buddy contraints, a free page at least the size of a pageblock will
1113 * be located at the start of the pageblock
1115 static inline int pageblock_free(struct page
*page
)
1117 return PageBuddy(page
) && page_order(page
) >= pageblock_order
;
1120 /* Return the pfn of the start of the next active pageblock after a given pfn */
1121 static unsigned long next_active_pageblock(unsigned long pfn
)
1123 struct page
*page
= pfn_to_page(pfn
);
1125 /* Ensure the starting page is pageblock-aligned */
1126 BUG_ON(pfn
& (pageblock_nr_pages
- 1));
1128 /* If the entire pageblock is free, move to the end of free page */
1129 if (pageblock_free(page
)) {
1131 /* be careful. we don't have locks, page_order can be changed.*/
1132 order
= page_order(page
);
1133 if ((order
< MAX_ORDER
) && (order
>= pageblock_order
))
1134 return pfn
+ (1 << order
);
1137 return pfn
+ pageblock_nr_pages
;
1140 static bool is_pageblock_removable_nolock(unsigned long pfn
)
1142 struct page
*page
= pfn_to_page(pfn
);
1146 * We have to be careful here because we are iterating over memory
1147 * sections which are not zone aware so we might end up outside of
1148 * the zone but still within the section.
1149 * We have to take care about the node as well. If the node is offline
1150 * its NODE_DATA will be NULL - see page_zone.
1152 if (!node_online(page_to_nid(page
)))
1155 zone
= page_zone(page
);
1156 pfn
= page_to_pfn(page
);
1157 if (!zone_spans_pfn(zone
, pfn
))
1160 return !has_unmovable_pages(zone
, page
, MIGRATE_MOVABLE
,
1164 /* Checks if this range of memory is likely to be hot-removable. */
1165 bool is_mem_section_removable(unsigned long start_pfn
, unsigned long nr_pages
)
1167 unsigned long end_pfn
, pfn
;
1169 end_pfn
= min(start_pfn
+ nr_pages
,
1170 zone_end_pfn(page_zone(pfn_to_page(start_pfn
))));
1172 /* Check the starting page of each pageblock within the range */
1173 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
= next_active_pageblock(pfn
)) {
1174 if (!is_pageblock_removable_nolock(pfn
))
1179 /* All pageblocks in the memory block are likely to be hot-removable */
1184 * Confirm all pages in a range [start, end) belong to the same zone (skipping
1185 * memory holes). When true, return the zone.
1187 struct zone
*test_pages_in_a_zone(unsigned long start_pfn
,
1188 unsigned long end_pfn
)
1190 unsigned long pfn
, sec_end_pfn
;
1191 struct zone
*zone
= NULL
;
1194 for (pfn
= start_pfn
, sec_end_pfn
= SECTION_ALIGN_UP(start_pfn
+ 1);
1196 pfn
= sec_end_pfn
, sec_end_pfn
+= PAGES_PER_SECTION
) {
1197 /* Make sure the memory section is present first */
1198 if (!present_section_nr(pfn_to_section_nr(pfn
)))
1200 for (; pfn
< sec_end_pfn
&& pfn
< end_pfn
;
1201 pfn
+= MAX_ORDER_NR_PAGES
) {
1203 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1204 while ((i
< MAX_ORDER_NR_PAGES
) &&
1205 !pfn_valid_within(pfn
+ i
))
1207 if (i
== MAX_ORDER_NR_PAGES
|| pfn
+ i
>= end_pfn
)
1209 /* Check if we got outside of the zone */
1210 if (zone
&& !zone_spans_pfn(zone
, pfn
+ i
))
1212 page
= pfn_to_page(pfn
+ i
);
1213 if (zone
&& page_zone(page
) != zone
)
1215 zone
= page_zone(page
);
1223 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1224 * non-lru movable pages and hugepages). We scan pfn because it's much
1225 * easier than scanning over linked list. This function returns the pfn
1226 * of the first found movable page if it's found, otherwise 0.
1228 static unsigned long scan_movable_pages(unsigned long start
, unsigned long end
)
1232 for (pfn
= start
; pfn
< end
; pfn
++) {
1233 struct page
*page
, *head
;
1236 if (!pfn_valid(pfn
))
1238 page
= pfn_to_page(pfn
);
1241 if (__PageMovable(page
))
1244 if (!PageHuge(page
))
1246 head
= compound_head(page
);
1247 if (page_huge_active(head
))
1249 skip
= compound_nr(head
) - (page
- head
);
1255 static struct page
*new_node_page(struct page
*page
, unsigned long private)
1257 int nid
= page_to_nid(page
);
1258 nodemask_t nmask
= node_states
[N_MEMORY
];
1261 * try to allocate from a different node but reuse this node if there
1262 * are no other online nodes to be used (e.g. we are offlining a part
1263 * of the only existing node)
1265 node_clear(nid
, nmask
);
1266 if (nodes_empty(nmask
))
1267 node_set(nid
, nmask
);
1269 return new_page_nodemask(page
, nid
, &nmask
);
1273 do_migrate_range(unsigned long start_pfn
, unsigned long end_pfn
)
1280 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
1281 if (!pfn_valid(pfn
))
1283 page
= pfn_to_page(pfn
);
1285 if (PageHuge(page
)) {
1286 struct page
*head
= compound_head(page
);
1287 pfn
= page_to_pfn(head
) + compound_nr(head
) - 1;
1288 isolate_huge_page(head
, &source
);
1290 } else if (PageTransHuge(page
))
1291 pfn
= page_to_pfn(compound_head(page
))
1292 + hpage_nr_pages(page
) - 1;
1295 * HWPoison pages have elevated reference counts so the migration would
1296 * fail on them. It also doesn't make any sense to migrate them in the
1297 * first place. Still try to unmap such a page in case it is still mapped
1298 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1299 * the unmap as the catch all safety net).
1301 if (PageHWPoison(page
)) {
1302 if (WARN_ON(PageLRU(page
)))
1303 isolate_lru_page(page
);
1304 if (page_mapped(page
))
1305 try_to_unmap(page
, TTU_IGNORE_MLOCK
| TTU_IGNORE_ACCESS
);
1309 if (!get_page_unless_zero(page
))
1312 * We can skip free pages. And we can deal with pages on
1313 * LRU and non-lru movable pages.
1316 ret
= isolate_lru_page(page
);
1318 ret
= isolate_movable_page(page
, ISOLATE_UNEVICTABLE
);
1319 if (!ret
) { /* Success */
1320 list_add_tail(&page
->lru
, &source
);
1321 if (!__PageMovable(page
))
1322 inc_node_page_state(page
, NR_ISOLATED_ANON
+
1323 page_is_file_lru(page
));
1326 pr_warn("failed to isolate pfn %lx\n", pfn
);
1327 dump_page(page
, "isolation failed");
1331 if (!list_empty(&source
)) {
1332 /* Allocate a new page from the nearest neighbor node */
1333 ret
= migrate_pages(&source
, new_node_page
, NULL
, 0,
1334 MIGRATE_SYNC
, MR_MEMORY_HOTPLUG
);
1336 list_for_each_entry(page
, &source
, lru
) {
1337 pr_warn("migrating pfn %lx failed ret:%d ",
1338 page_to_pfn(page
), ret
);
1339 dump_page(page
, "migration failure");
1341 putback_movable_pages(&source
);
1348 /* Mark all sections offline and remove all free pages from the buddy. */
1350 offline_isolated_pages_cb(unsigned long start
, unsigned long nr_pages
,
1353 unsigned long *offlined_pages
= (unsigned long *)data
;
1355 *offlined_pages
+= __offline_isolated_pages(start
, start
+ nr_pages
);
1360 * Check all pages in range, recoreded as memory resource, are isolated.
1363 check_pages_isolated_cb(unsigned long start_pfn
, unsigned long nr_pages
,
1366 return test_pages_isolated(start_pfn
, start_pfn
+ nr_pages
,
1370 static int __init
cmdline_parse_movable_node(char *p
)
1372 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1373 movable_node_enabled
= true;
1375 pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1379 early_param("movable_node", cmdline_parse_movable_node
);
1381 /* check which state of node_states will be changed when offline memory */
1382 static void node_states_check_changes_offline(unsigned long nr_pages
,
1383 struct zone
*zone
, struct memory_notify
*arg
)
1385 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1386 unsigned long present_pages
= 0;
1389 arg
->status_change_nid
= NUMA_NO_NODE
;
1390 arg
->status_change_nid_normal
= NUMA_NO_NODE
;
1391 arg
->status_change_nid_high
= NUMA_NO_NODE
;
1394 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1395 * If the memory to be offline is within the range
1396 * [0..ZONE_NORMAL], and it is the last present memory there,
1397 * the zones in that range will become empty after the offlining,
1398 * thus we can determine that we need to clear the node from
1399 * node_states[N_NORMAL_MEMORY].
1401 for (zt
= 0; zt
<= ZONE_NORMAL
; zt
++)
1402 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1403 if (zone_idx(zone
) <= ZONE_NORMAL
&& nr_pages
>= present_pages
)
1404 arg
->status_change_nid_normal
= zone_to_nid(zone
);
1406 #ifdef CONFIG_HIGHMEM
1408 * node_states[N_HIGH_MEMORY] contains nodes which
1409 * have normal memory or high memory.
1410 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1411 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1412 * we determine that the zones in that range become empty,
1413 * we need to clear the node for N_HIGH_MEMORY.
1415 present_pages
+= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1416 if (zone_idx(zone
) <= ZONE_HIGHMEM
&& nr_pages
>= present_pages
)
1417 arg
->status_change_nid_high
= zone_to_nid(zone
);
1421 * We have accounted the pages from [0..ZONE_NORMAL), and
1422 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1424 * Here we count the possible pages from ZONE_MOVABLE.
1425 * If after having accounted all the pages, we see that the nr_pages
1426 * to be offlined is over or equal to the accounted pages,
1427 * we know that the node will become empty, and so, we can clear
1428 * it for N_MEMORY as well.
1430 present_pages
+= pgdat
->node_zones
[ZONE_MOVABLE
].present_pages
;
1432 if (nr_pages
>= present_pages
)
1433 arg
->status_change_nid
= zone_to_nid(zone
);
1436 static void node_states_clear_node(int node
, struct memory_notify
*arg
)
1438 if (arg
->status_change_nid_normal
>= 0)
1439 node_clear_state(node
, N_NORMAL_MEMORY
);
1441 if (arg
->status_change_nid_high
>= 0)
1442 node_clear_state(node
, N_HIGH_MEMORY
);
1444 if (arg
->status_change_nid
>= 0)
1445 node_clear_state(node
, N_MEMORY
);
1448 static int count_system_ram_pages_cb(unsigned long start_pfn
,
1449 unsigned long nr_pages
, void *data
)
1451 unsigned long *nr_system_ram_pages
= data
;
1453 *nr_system_ram_pages
+= nr_pages
;
1457 static int __ref
__offline_pages(unsigned long start_pfn
,
1458 unsigned long end_pfn
)
1460 unsigned long pfn
, nr_pages
= 0;
1461 unsigned long offlined_pages
= 0;
1462 int ret
, node
, nr_isolate_pageblock
;
1463 unsigned long flags
;
1465 struct memory_notify arg
;
1468 mem_hotplug_begin();
1471 * Don't allow to offline memory blocks that contain holes.
1472 * Consequently, memory blocks with holes can never get onlined
1473 * via the hotplug path - online_pages() - as hotplugged memory has
1474 * no holes. This way, we e.g., don't have to worry about marking
1475 * memory holes PG_reserved, don't need pfn_valid() checks, and can
1476 * avoid using walk_system_ram_range() later.
1478 walk_system_ram_range(start_pfn
, end_pfn
- start_pfn
, &nr_pages
,
1479 count_system_ram_pages_cb
);
1480 if (nr_pages
!= end_pfn
- start_pfn
) {
1482 reason
= "memory holes";
1483 goto failed_removal
;
1486 /* This makes hotplug much easier...and readable.
1487 we assume this for now. .*/
1488 zone
= test_pages_in_a_zone(start_pfn
, end_pfn
);
1491 reason
= "multizone range";
1492 goto failed_removal
;
1494 node
= zone_to_nid(zone
);
1496 /* set above range as isolated */
1497 ret
= start_isolate_page_range(start_pfn
, end_pfn
,
1499 MEMORY_OFFLINE
| REPORT_FAILURE
);
1501 reason
= "failure to isolate range";
1502 goto failed_removal
;
1504 nr_isolate_pageblock
= ret
;
1506 arg
.start_pfn
= start_pfn
;
1507 arg
.nr_pages
= nr_pages
;
1508 node_states_check_changes_offline(nr_pages
, zone
, &arg
);
1510 ret
= memory_notify(MEM_GOING_OFFLINE
, &arg
);
1511 ret
= notifier_to_errno(ret
);
1513 reason
= "notifier failure";
1514 goto failed_removal_isolated
;
1518 for (pfn
= start_pfn
; pfn
;) {
1519 if (signal_pending(current
)) {
1521 reason
= "signal backoff";
1522 goto failed_removal_isolated
;
1526 lru_add_drain_all();
1528 pfn
= scan_movable_pages(pfn
, end_pfn
);
1531 * TODO: fatal migration failures should bail
1534 do_migrate_range(pfn
, end_pfn
);
1539 * Dissolve free hugepages in the memory block before doing
1540 * offlining actually in order to make hugetlbfs's object
1541 * counting consistent.
1543 ret
= dissolve_free_huge_pages(start_pfn
, end_pfn
);
1545 reason
= "failure to dissolve huge pages";
1546 goto failed_removal_isolated
;
1549 ret
= walk_system_ram_range(start_pfn
, end_pfn
- start_pfn
,
1550 NULL
, check_pages_isolated_cb
);
1553 /* Ok, all of our target is isolated.
1554 We cannot do rollback at this point. */
1555 walk_system_ram_range(start_pfn
, end_pfn
- start_pfn
,
1556 &offlined_pages
, offline_isolated_pages_cb
);
1557 pr_info("Offlined Pages %ld\n", offlined_pages
);
1559 * Onlining will reset pagetype flags and makes migrate type
1560 * MOVABLE, so just need to decrease the number of isolated
1561 * pageblocks zone counter here.
1563 spin_lock_irqsave(&zone
->lock
, flags
);
1564 zone
->nr_isolate_pageblock
-= nr_isolate_pageblock
;
1565 spin_unlock_irqrestore(&zone
->lock
, flags
);
1567 /* removal success */
1568 adjust_managed_page_count(pfn_to_page(start_pfn
), -offlined_pages
);
1569 zone
->present_pages
-= offlined_pages
;
1571 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
1572 zone
->zone_pgdat
->node_present_pages
-= offlined_pages
;
1573 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
1575 init_per_zone_wmark_min();
1577 if (!populated_zone(zone
)) {
1578 zone_pcp_reset(zone
);
1579 build_all_zonelists(NULL
);
1581 zone_pcp_update(zone
);
1583 node_states_clear_node(node
, &arg
);
1584 if (arg
.status_change_nid
>= 0) {
1586 kcompactd_stop(node
);
1589 vm_total_pages
= nr_free_pagecache_pages();
1590 writeback_set_ratelimit();
1592 memory_notify(MEM_OFFLINE
, &arg
);
1593 remove_pfn_range_from_zone(zone
, start_pfn
, nr_pages
);
1597 failed_removal_isolated
:
1598 undo_isolate_page_range(start_pfn
, end_pfn
, MIGRATE_MOVABLE
);
1599 memory_notify(MEM_CANCEL_OFFLINE
, &arg
);
1601 pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1602 (unsigned long long) start_pfn
<< PAGE_SHIFT
,
1603 ((unsigned long long) end_pfn
<< PAGE_SHIFT
) - 1,
1605 /* pushback to free area */
1610 int offline_pages(unsigned long start_pfn
, unsigned long nr_pages
)
1612 return __offline_pages(start_pfn
, start_pfn
+ nr_pages
);
1615 static int check_memblock_offlined_cb(struct memory_block
*mem
, void *arg
)
1617 int ret
= !is_memblock_offlined(mem
);
1619 if (unlikely(ret
)) {
1620 phys_addr_t beginpa
, endpa
;
1622 beginpa
= PFN_PHYS(section_nr_to_pfn(mem
->start_section_nr
));
1623 endpa
= beginpa
+ memory_block_size_bytes() - 1;
1624 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1632 static int check_cpu_on_node(pg_data_t
*pgdat
)
1636 for_each_present_cpu(cpu
) {
1637 if (cpu_to_node(cpu
) == pgdat
->node_id
)
1639 * the cpu on this node isn't removed, and we can't
1640 * offline this node.
1648 static int check_no_memblock_for_node_cb(struct memory_block
*mem
, void *arg
)
1650 int nid
= *(int *)arg
;
1653 * If a memory block belongs to multiple nodes, the stored nid is not
1654 * reliable. However, such blocks are always online (e.g., cannot get
1655 * offlined) and, therefore, are still spanned by the node.
1657 return mem
->nid
== nid
? -EEXIST
: 0;
1664 * Offline a node if all memory sections and cpus of the node are removed.
1666 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1667 * and online/offline operations before this call.
1669 void try_offline_node(int nid
)
1671 pg_data_t
*pgdat
= NODE_DATA(nid
);
1675 * If the node still spans pages (especially ZONE_DEVICE), don't
1676 * offline it. A node spans memory after move_pfn_range_to_zone(),
1677 * e.g., after the memory block was onlined.
1679 if (pgdat
->node_spanned_pages
)
1683 * Especially offline memory blocks might not be spanned by the
1684 * node. They will get spanned by the node once they get onlined.
1685 * However, they link to the node in sysfs and can get onlined later.
1687 rc
= for_each_memory_block(&nid
, check_no_memblock_for_node_cb
);
1691 if (check_cpu_on_node(pgdat
))
1695 * all memory/cpu of this node are removed, we can offline this
1698 node_set_offline(nid
);
1699 unregister_one_node(nid
);
1701 EXPORT_SYMBOL(try_offline_node
);
1703 static void __release_memory_resource(resource_size_t start
,
1704 resource_size_t size
)
1709 * When removing memory in the same granularity as it was added,
1710 * this function never fails. It might only fail if resources
1711 * have to be adjusted or split. We'll ignore the error, as
1712 * removing of memory cannot fail.
1714 ret
= release_mem_region_adjustable(&iomem_resource
, start
, size
);
1716 resource_size_t endres
= start
+ size
- 1;
1718 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
1719 &start
, &endres
, ret
);
1723 static int __ref
try_remove_memory(int nid
, u64 start
, u64 size
)
1727 BUG_ON(check_hotplug_memory_range(start
, size
));
1730 * All memory blocks must be offlined before removing memory. Check
1731 * whether all memory blocks in question are offline and return error
1732 * if this is not the case.
1734 rc
= walk_memory_blocks(start
, size
, NULL
, check_memblock_offlined_cb
);
1738 /* remove memmap entry */
1739 firmware_map_remove(start
, start
+ size
, "System RAM");
1742 * Memory block device removal under the device_hotplug_lock is
1743 * a barrier against racing online attempts.
1745 remove_memory_block_devices(start
, size
);
1747 mem_hotplug_begin();
1749 arch_remove_memory(nid
, start
, size
, NULL
);
1750 memblock_free(start
, size
);
1751 memblock_remove(start
, size
);
1752 __release_memory_resource(start
, size
);
1754 try_offline_node(nid
);
1764 * @start: physical address of the region to remove
1765 * @size: size of the region to remove
1767 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1768 * and online/offline operations before this call, as required by
1769 * try_offline_node().
1771 void __remove_memory(int nid
, u64 start
, u64 size
)
1775 * trigger BUG() if some memory is not offlined prior to calling this
1778 if (try_remove_memory(nid
, start
, size
))
1783 * Remove memory if every memory block is offline, otherwise return -EBUSY is
1784 * some memory is not offline
1786 int remove_memory(int nid
, u64 start
, u64 size
)
1790 lock_device_hotplug();
1791 rc
= try_remove_memory(nid
, start
, size
);
1792 unlock_device_hotplug();
1796 EXPORT_SYMBOL_GPL(remove_memory
);
1797 #endif /* CONFIG_MEMORY_HOTREMOVE */