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
,
102 const char *resource_name
)
104 struct resource
*res
;
105 unsigned long flags
= IORESOURCE_SYSTEM_RAM
| IORESOURCE_BUSY
;
107 if (strcmp(resource_name
, "System RAM"))
108 flags
|= IORESOURCE_MEM_DRIVER_MANAGED
;
111 * Make sure value parsed from 'mem=' only restricts memory adding
112 * while booting, so that memory hotplug won't be impacted. Please
113 * refer to document of 'mem=' in kernel-parameters.txt for more
116 if (start
+ size
> max_mem_size
&& system_state
< SYSTEM_RUNNING
)
117 return ERR_PTR(-E2BIG
);
120 * Request ownership of the new memory range. This might be
121 * a child of an existing resource that was present but
122 * not marked as busy.
124 res
= __request_region(&iomem_resource
, start
, size
,
125 resource_name
, flags
);
128 pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n",
129 start
, start
+ size
);
130 return ERR_PTR(-EEXIST
);
135 static void release_memory_resource(struct resource
*res
)
139 release_resource(res
);
143 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
144 void get_page_bootmem(unsigned long info
, struct page
*page
,
147 page
->freelist
= (void *)type
;
148 SetPagePrivate(page
);
149 set_page_private(page
, info
);
153 void put_page_bootmem(struct page
*page
)
157 type
= (unsigned long) page
->freelist
;
158 BUG_ON(type
< MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE
||
159 type
> MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE
);
161 if (page_ref_dec_return(page
) == 1) {
162 page
->freelist
= NULL
;
163 ClearPagePrivate(page
);
164 set_page_private(page
, 0);
165 INIT_LIST_HEAD(&page
->lru
);
166 free_reserved_page(page
);
170 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
171 #ifndef CONFIG_SPARSEMEM_VMEMMAP
172 static void register_page_bootmem_info_section(unsigned long start_pfn
)
174 unsigned long mapsize
, section_nr
, i
;
175 struct mem_section
*ms
;
176 struct page
*page
, *memmap
;
177 struct mem_section_usage
*usage
;
179 section_nr
= pfn_to_section_nr(start_pfn
);
180 ms
= __nr_to_section(section_nr
);
182 /* Get section's memmap address */
183 memmap
= sparse_decode_mem_map(ms
->section_mem_map
, section_nr
);
186 * Get page for the memmap's phys address
187 * XXX: need more consideration for sparse_vmemmap...
189 page
= virt_to_page(memmap
);
190 mapsize
= sizeof(struct page
) * PAGES_PER_SECTION
;
191 mapsize
= PAGE_ALIGN(mapsize
) >> PAGE_SHIFT
;
193 /* remember memmap's page */
194 for (i
= 0; i
< mapsize
; i
++, page
++)
195 get_page_bootmem(section_nr
, page
, SECTION_INFO
);
198 page
= virt_to_page(usage
);
200 mapsize
= PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT
;
202 for (i
= 0; i
< mapsize
; i
++, page
++)
203 get_page_bootmem(section_nr
, page
, MIX_SECTION_INFO
);
206 #else /* CONFIG_SPARSEMEM_VMEMMAP */
207 static void register_page_bootmem_info_section(unsigned long start_pfn
)
209 unsigned long mapsize
, section_nr
, i
;
210 struct mem_section
*ms
;
211 struct page
*page
, *memmap
;
212 struct mem_section_usage
*usage
;
214 section_nr
= pfn_to_section_nr(start_pfn
);
215 ms
= __nr_to_section(section_nr
);
217 memmap
= sparse_decode_mem_map(ms
->section_mem_map
, section_nr
);
219 register_page_bootmem_memmap(section_nr
, memmap
, PAGES_PER_SECTION
);
222 page
= virt_to_page(usage
);
224 mapsize
= PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT
;
226 for (i
= 0; i
< mapsize
; i
++, page
++)
227 get_page_bootmem(section_nr
, page
, MIX_SECTION_INFO
);
229 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
231 void __init
register_page_bootmem_info_node(struct pglist_data
*pgdat
)
233 unsigned long i
, pfn
, end_pfn
, nr_pages
;
234 int node
= pgdat
->node_id
;
237 nr_pages
= PAGE_ALIGN(sizeof(struct pglist_data
)) >> PAGE_SHIFT
;
238 page
= virt_to_page(pgdat
);
240 for (i
= 0; i
< nr_pages
; i
++, page
++)
241 get_page_bootmem(node
, page
, NODE_INFO
);
243 pfn
= pgdat
->node_start_pfn
;
244 end_pfn
= pgdat_end_pfn(pgdat
);
246 /* register section info */
247 for (; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
249 * Some platforms can assign the same pfn to multiple nodes - on
250 * node0 as well as nodeN. To avoid registering a pfn against
251 * multiple nodes we check that this pfn does not already
252 * reside in some other nodes.
254 if (pfn_valid(pfn
) && (early_pfn_to_nid(pfn
) == node
))
255 register_page_bootmem_info_section(pfn
);
258 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
260 static int check_pfn_span(unsigned long pfn
, unsigned long nr_pages
,
264 * Disallow all operations smaller than a sub-section and only
265 * allow operations smaller than a section for
266 * SPARSEMEM_VMEMMAP. Note that check_hotplug_memory_range()
267 * enforces a larger memory_block_size_bytes() granularity for
268 * memory that will be marked online, so this check should only
269 * fire for direct arch_{add,remove}_memory() users outside of
270 * add_memory_resource().
272 unsigned long min_align
;
274 if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP
))
275 min_align
= PAGES_PER_SUBSECTION
;
277 min_align
= PAGES_PER_SECTION
;
278 if (!IS_ALIGNED(pfn
, min_align
)
279 || !IS_ALIGNED(nr_pages
, min_align
)) {
280 WARN(1, "Misaligned __%s_pages start: %#lx end: #%lx\n",
281 reason
, pfn
, pfn
+ nr_pages
- 1);
287 static int check_hotplug_memory_addressable(unsigned long pfn
,
288 unsigned long nr_pages
)
290 const u64 max_addr
= PFN_PHYS(pfn
+ nr_pages
) - 1;
292 if (max_addr
>> MAX_PHYSMEM_BITS
) {
293 const u64 max_allowed
= (1ull << (MAX_PHYSMEM_BITS
+ 1)) - 1;
295 "Hotplugged memory exceeds maximum addressable address, range=%#llx-%#llx, maximum=%#llx\n",
296 (u64
)PFN_PHYS(pfn
), max_addr
, max_allowed
);
304 * Reasonably generic function for adding memory. It is
305 * expected that archs that support memory hotplug will
306 * call this function after deciding the zone to which to
309 int __ref
__add_pages(int nid
, unsigned long pfn
, unsigned long nr_pages
,
310 struct mhp_params
*params
)
312 const unsigned long end_pfn
= pfn
+ nr_pages
;
313 unsigned long cur_nr_pages
;
315 struct vmem_altmap
*altmap
= params
->altmap
;
317 if (WARN_ON_ONCE(!params
->pgprot
.pgprot
))
320 err
= check_hotplug_memory_addressable(pfn
, nr_pages
);
326 * Validate altmap is within bounds of the total request
328 if (altmap
->base_pfn
!= pfn
329 || vmem_altmap_offset(altmap
) > nr_pages
) {
330 pr_warn_once("memory add fail, invalid altmap\n");
336 err
= check_pfn_span(pfn
, nr_pages
, "add");
340 for (; pfn
< end_pfn
; pfn
+= cur_nr_pages
) {
341 /* Select all remaining pages up to the next section boundary */
342 cur_nr_pages
= min(end_pfn
- pfn
,
343 SECTION_ALIGN_UP(pfn
+ 1) - pfn
);
344 err
= sparse_add_section(nid
, pfn
, cur_nr_pages
, altmap
);
349 vmemmap_populate_print_last();
354 int __weak
memory_add_physaddr_to_nid(u64 start
)
356 pr_info_once("Unknown target node for memory at 0x%llx, assuming node 0\n",
360 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid
);
363 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
364 static unsigned long find_smallest_section_pfn(int nid
, struct zone
*zone
,
365 unsigned long start_pfn
,
366 unsigned long end_pfn
)
368 for (; start_pfn
< end_pfn
; start_pfn
+= PAGES_PER_SUBSECTION
) {
369 if (unlikely(!pfn_to_online_page(start_pfn
)))
372 if (unlikely(pfn_to_nid(start_pfn
) != nid
))
375 if (zone
!= page_zone(pfn_to_page(start_pfn
)))
384 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
385 static unsigned long find_biggest_section_pfn(int nid
, struct zone
*zone
,
386 unsigned long start_pfn
,
387 unsigned long end_pfn
)
391 /* pfn is the end pfn of a memory section. */
393 for (; pfn
>= start_pfn
; pfn
-= PAGES_PER_SUBSECTION
) {
394 if (unlikely(!pfn_to_online_page(pfn
)))
397 if (unlikely(pfn_to_nid(pfn
) != nid
))
400 if (zone
!= page_zone(pfn_to_page(pfn
)))
409 static void shrink_zone_span(struct zone
*zone
, unsigned long start_pfn
,
410 unsigned long end_pfn
)
413 int nid
= zone_to_nid(zone
);
415 zone_span_writelock(zone
);
416 if (zone
->zone_start_pfn
== start_pfn
) {
418 * If the section is smallest section in the zone, it need
419 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
420 * In this case, we find second smallest valid mem_section
421 * for shrinking zone.
423 pfn
= find_smallest_section_pfn(nid
, zone
, end_pfn
,
426 zone
->spanned_pages
= zone_end_pfn(zone
) - pfn
;
427 zone
->zone_start_pfn
= pfn
;
429 zone
->zone_start_pfn
= 0;
430 zone
->spanned_pages
= 0;
432 } else if (zone_end_pfn(zone
) == end_pfn
) {
434 * If the section is biggest section in the zone, it need
435 * shrink zone->spanned_pages.
436 * In this case, we find second biggest valid mem_section for
439 pfn
= find_biggest_section_pfn(nid
, zone
, zone
->zone_start_pfn
,
442 zone
->spanned_pages
= pfn
- zone
->zone_start_pfn
+ 1;
444 zone
->zone_start_pfn
= 0;
445 zone
->spanned_pages
= 0;
448 zone_span_writeunlock(zone
);
451 static void update_pgdat_span(struct pglist_data
*pgdat
)
453 unsigned long node_start_pfn
= 0, node_end_pfn
= 0;
456 for (zone
= pgdat
->node_zones
;
457 zone
< pgdat
->node_zones
+ MAX_NR_ZONES
; zone
++) {
458 unsigned long zone_end_pfn
= zone
->zone_start_pfn
+
461 /* No need to lock the zones, they can't change. */
462 if (!zone
->spanned_pages
)
465 node_start_pfn
= zone
->zone_start_pfn
;
466 node_end_pfn
= zone_end_pfn
;
470 if (zone_end_pfn
> node_end_pfn
)
471 node_end_pfn
= zone_end_pfn
;
472 if (zone
->zone_start_pfn
< node_start_pfn
)
473 node_start_pfn
= zone
->zone_start_pfn
;
476 pgdat
->node_start_pfn
= node_start_pfn
;
477 pgdat
->node_spanned_pages
= node_end_pfn
- node_start_pfn
;
480 void __ref
remove_pfn_range_from_zone(struct zone
*zone
,
481 unsigned long start_pfn
,
482 unsigned long nr_pages
)
484 const unsigned long end_pfn
= start_pfn
+ nr_pages
;
485 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
486 unsigned long pfn
, cur_nr_pages
, flags
;
488 /* Poison struct pages because they are now uninitialized again. */
489 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= cur_nr_pages
) {
492 /* Select all remaining pages up to the next section boundary */
494 min(end_pfn
- pfn
, SECTION_ALIGN_UP(pfn
+ 1) - pfn
);
495 page_init_poison(pfn_to_page(pfn
),
496 sizeof(struct page
) * cur_nr_pages
);
499 #ifdef CONFIG_ZONE_DEVICE
501 * Zone shrinking code cannot properly deal with ZONE_DEVICE. So
502 * we will not try to shrink the zones - which is okay as
503 * set_zone_contiguous() cannot deal with ZONE_DEVICE either way.
505 if (zone_idx(zone
) == ZONE_DEVICE
)
509 clear_zone_contiguous(zone
);
511 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
512 shrink_zone_span(zone
, start_pfn
, start_pfn
+ nr_pages
);
513 update_pgdat_span(pgdat
);
514 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
516 set_zone_contiguous(zone
);
519 static void __remove_section(unsigned long pfn
, unsigned long nr_pages
,
520 unsigned long map_offset
,
521 struct vmem_altmap
*altmap
)
523 struct mem_section
*ms
= __pfn_to_section(pfn
);
525 if (WARN_ON_ONCE(!valid_section(ms
)))
528 sparse_remove_section(ms
, pfn
, nr_pages
, map_offset
, altmap
);
532 * __remove_pages() - remove sections of pages
533 * @pfn: starting pageframe (must be aligned to start of a section)
534 * @nr_pages: number of pages to remove (must be multiple of section size)
535 * @altmap: alternative device page map or %NULL if default memmap is used
537 * Generic helper function to remove section mappings and sysfs entries
538 * for the section of the memory we are removing. Caller needs to make
539 * sure that pages are marked reserved and zones are adjust properly by
540 * calling offline_pages().
542 void __remove_pages(unsigned long pfn
, unsigned long nr_pages
,
543 struct vmem_altmap
*altmap
)
545 const unsigned long end_pfn
= pfn
+ nr_pages
;
546 unsigned long cur_nr_pages
;
547 unsigned long map_offset
= 0;
549 map_offset
= vmem_altmap_offset(altmap
);
551 if (check_pfn_span(pfn
, nr_pages
, "remove"))
554 for (; pfn
< end_pfn
; pfn
+= cur_nr_pages
) {
556 /* Select all remaining pages up to the next section boundary */
557 cur_nr_pages
= min(end_pfn
- pfn
,
558 SECTION_ALIGN_UP(pfn
+ 1) - pfn
);
559 __remove_section(pfn
, cur_nr_pages
, map_offset
, altmap
);
564 int set_online_page_callback(online_page_callback_t callback
)
569 mutex_lock(&online_page_callback_lock
);
571 if (online_page_callback
== generic_online_page
) {
572 online_page_callback
= callback
;
576 mutex_unlock(&online_page_callback_lock
);
581 EXPORT_SYMBOL_GPL(set_online_page_callback
);
583 int restore_online_page_callback(online_page_callback_t callback
)
588 mutex_lock(&online_page_callback_lock
);
590 if (online_page_callback
== callback
) {
591 online_page_callback
= generic_online_page
;
595 mutex_unlock(&online_page_callback_lock
);
600 EXPORT_SYMBOL_GPL(restore_online_page_callback
);
602 void generic_online_page(struct page
*page
, unsigned int order
)
605 * Freeing the page with debug_pagealloc enabled will try to unmap it,
606 * so we should map it first. This is better than introducing a special
607 * case in page freeing fast path.
609 if (debug_pagealloc_enabled_static())
610 kernel_map_pages(page
, 1 << order
, 1);
611 __free_pages_core(page
, order
);
612 totalram_pages_add(1UL << order
);
613 #ifdef CONFIG_HIGHMEM
614 if (PageHighMem(page
))
615 totalhigh_pages_add(1UL << order
);
618 EXPORT_SYMBOL_GPL(generic_online_page
);
620 static int online_pages_range(unsigned long start_pfn
, unsigned long nr_pages
,
623 const unsigned long end_pfn
= start_pfn
+ nr_pages
;
628 * Online the pages. The callback might decide to keep some pages
629 * PG_reserved (to add them to the buddy later), but we still account
630 * them as being online/belonging to this zone ("present").
632 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= 1ul << order
) {
633 order
= min(MAX_ORDER
- 1, get_order(PFN_PHYS(end_pfn
- pfn
)));
634 /* __free_pages_core() wants pfns to be aligned to the order */
635 if (WARN_ON_ONCE(!IS_ALIGNED(pfn
, 1ul << order
)))
637 (*online_page_callback
)(pfn_to_page(pfn
), order
);
640 /* mark all involved sections as online */
641 online_mem_sections(start_pfn
, end_pfn
);
643 *(unsigned long *)arg
+= nr_pages
;
647 /* check which state of node_states will be changed when online memory */
648 static void node_states_check_changes_online(unsigned long nr_pages
,
649 struct zone
*zone
, struct memory_notify
*arg
)
651 int nid
= zone_to_nid(zone
);
653 arg
->status_change_nid
= NUMA_NO_NODE
;
654 arg
->status_change_nid_normal
= NUMA_NO_NODE
;
655 arg
->status_change_nid_high
= NUMA_NO_NODE
;
657 if (!node_state(nid
, N_MEMORY
))
658 arg
->status_change_nid
= nid
;
659 if (zone_idx(zone
) <= ZONE_NORMAL
&& !node_state(nid
, N_NORMAL_MEMORY
))
660 arg
->status_change_nid_normal
= nid
;
661 #ifdef CONFIG_HIGHMEM
662 if (zone_idx(zone
) <= ZONE_HIGHMEM
&& !node_state(nid
, N_HIGH_MEMORY
))
663 arg
->status_change_nid_high
= nid
;
667 static void node_states_set_node(int node
, struct memory_notify
*arg
)
669 if (arg
->status_change_nid_normal
>= 0)
670 node_set_state(node
, N_NORMAL_MEMORY
);
672 if (arg
->status_change_nid_high
>= 0)
673 node_set_state(node
, N_HIGH_MEMORY
);
675 if (arg
->status_change_nid
>= 0)
676 node_set_state(node
, N_MEMORY
);
679 static void __meminit
resize_zone_range(struct zone
*zone
, unsigned long start_pfn
,
680 unsigned long nr_pages
)
682 unsigned long old_end_pfn
= zone_end_pfn(zone
);
684 if (zone_is_empty(zone
) || start_pfn
< zone
->zone_start_pfn
)
685 zone
->zone_start_pfn
= start_pfn
;
687 zone
->spanned_pages
= max(start_pfn
+ nr_pages
, old_end_pfn
) - zone
->zone_start_pfn
;
690 static void __meminit
resize_pgdat_range(struct pglist_data
*pgdat
, unsigned long start_pfn
,
691 unsigned long nr_pages
)
693 unsigned long old_end_pfn
= pgdat_end_pfn(pgdat
);
695 if (!pgdat
->node_spanned_pages
|| start_pfn
< pgdat
->node_start_pfn
)
696 pgdat
->node_start_pfn
= start_pfn
;
698 pgdat
->node_spanned_pages
= max(start_pfn
+ nr_pages
, old_end_pfn
) - pgdat
->node_start_pfn
;
702 * Associate the pfn range with the given zone, initializing the memmaps
703 * and resizing the pgdat/zone data to span the added pages. After this
704 * call, all affected pages are PG_reserved.
706 void __ref
move_pfn_range_to_zone(struct zone
*zone
, unsigned long start_pfn
,
707 unsigned long nr_pages
, struct vmem_altmap
*altmap
)
709 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
710 int nid
= pgdat
->node_id
;
713 clear_zone_contiguous(zone
);
715 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
716 pgdat_resize_lock(pgdat
, &flags
);
717 zone_span_writelock(zone
);
718 if (zone_is_empty(zone
))
719 init_currently_empty_zone(zone
, start_pfn
, nr_pages
);
720 resize_zone_range(zone
, start_pfn
, nr_pages
);
721 zone_span_writeunlock(zone
);
722 resize_pgdat_range(pgdat
, start_pfn
, nr_pages
);
723 pgdat_resize_unlock(pgdat
, &flags
);
726 * TODO now we have a visible range of pages which are not associated
727 * with their zone properly. Not nice but set_pfnblock_flags_mask
728 * expects the zone spans the pfn range. All the pages in the range
729 * are reserved so nobody should be touching them so we should be safe
731 memmap_init_zone(nr_pages
, nid
, zone_idx(zone
), start_pfn
,
732 MEMINIT_HOTPLUG
, altmap
);
734 set_zone_contiguous(zone
);
738 * Returns a default kernel memory zone for the given pfn range.
739 * If no kernel zone covers this pfn range it will automatically go
740 * to the ZONE_NORMAL.
742 static struct zone
*default_kernel_zone_for_pfn(int nid
, unsigned long start_pfn
,
743 unsigned long nr_pages
)
745 struct pglist_data
*pgdat
= NODE_DATA(nid
);
748 for (zid
= 0; zid
<= ZONE_NORMAL
; zid
++) {
749 struct zone
*zone
= &pgdat
->node_zones
[zid
];
751 if (zone_intersects(zone
, start_pfn
, nr_pages
))
755 return &pgdat
->node_zones
[ZONE_NORMAL
];
758 static inline struct zone
*default_zone_for_pfn(int nid
, unsigned long start_pfn
,
759 unsigned long nr_pages
)
761 struct zone
*kernel_zone
= default_kernel_zone_for_pfn(nid
, start_pfn
,
763 struct zone
*movable_zone
= &NODE_DATA(nid
)->node_zones
[ZONE_MOVABLE
];
764 bool in_kernel
= zone_intersects(kernel_zone
, start_pfn
, nr_pages
);
765 bool in_movable
= zone_intersects(movable_zone
, start_pfn
, nr_pages
);
768 * We inherit the existing zone in a simple case where zones do not
769 * overlap in the given range
771 if (in_kernel
^ in_movable
)
772 return (in_kernel
) ? kernel_zone
: movable_zone
;
775 * If the range doesn't belong to any zone or two zones overlap in the
776 * given range then we use movable zone only if movable_node is
777 * enabled because we always online to a kernel zone by default.
779 return movable_node_enabled
? movable_zone
: kernel_zone
;
782 struct zone
* zone_for_pfn_range(int online_type
, int nid
, unsigned start_pfn
,
783 unsigned long nr_pages
)
785 if (online_type
== MMOP_ONLINE_KERNEL
)
786 return default_kernel_zone_for_pfn(nid
, start_pfn
, nr_pages
);
788 if (online_type
== MMOP_ONLINE_MOVABLE
)
789 return &NODE_DATA(nid
)->node_zones
[ZONE_MOVABLE
];
791 return default_zone_for_pfn(nid
, start_pfn
, nr_pages
);
794 int __ref
online_pages(unsigned long pfn
, unsigned long nr_pages
,
795 int online_type
, int nid
)
798 unsigned long onlined_pages
= 0;
800 int need_zonelists_rebuild
= 0;
802 struct memory_notify arg
;
806 /* associate pfn range with the zone */
807 zone
= zone_for_pfn_range(online_type
, nid
, pfn
, nr_pages
);
808 move_pfn_range_to_zone(zone
, pfn
, nr_pages
, NULL
);
811 arg
.nr_pages
= nr_pages
;
812 node_states_check_changes_online(nr_pages
, zone
, &arg
);
814 ret
= memory_notify(MEM_GOING_ONLINE
, &arg
);
815 ret
= notifier_to_errno(ret
);
817 goto failed_addition
;
820 * If this zone is not populated, then it is not in zonelist.
821 * This means the page allocator ignores this zone.
822 * So, zonelist must be updated after online.
824 if (!populated_zone(zone
)) {
825 need_zonelists_rebuild
= 1;
826 setup_zone_pageset(zone
);
829 ret
= walk_system_ram_range(pfn
, nr_pages
, &onlined_pages
,
832 /* not a single memory resource was applicable */
833 if (need_zonelists_rebuild
)
834 zone_pcp_reset(zone
);
835 goto failed_addition
;
838 zone
->present_pages
+= onlined_pages
;
840 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
841 zone
->zone_pgdat
->node_present_pages
+= onlined_pages
;
842 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
845 * When exposing larger, physically contiguous memory areas to the
846 * buddy, shuffling in the buddy (when freeing onlined pages, putting
847 * them either to the head or the tail of the freelist) is only helpful
848 * for maintaining the shuffle, but not for creating the initial
849 * shuffle. Shuffle the whole zone to make sure the just onlined pages
850 * are properly distributed across the whole freelist.
854 node_states_set_node(nid
, &arg
);
855 if (need_zonelists_rebuild
)
856 build_all_zonelists(NULL
);
857 zone_pcp_update(zone
);
859 init_per_zone_wmark_min();
864 writeback_set_ratelimit();
866 memory_notify(MEM_ONLINE
, &arg
);
871 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
872 (unsigned long long) pfn
<< PAGE_SHIFT
,
873 (((unsigned long long) pfn
+ nr_pages
) << PAGE_SHIFT
) - 1);
874 memory_notify(MEM_CANCEL_ONLINE
, &arg
);
875 remove_pfn_range_from_zone(zone
, pfn
, nr_pages
);
879 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
881 static void reset_node_present_pages(pg_data_t
*pgdat
)
885 for (z
= pgdat
->node_zones
; z
< pgdat
->node_zones
+ MAX_NR_ZONES
; z
++)
886 z
->present_pages
= 0;
888 pgdat
->node_present_pages
= 0;
891 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
892 static pg_data_t __ref
*hotadd_new_pgdat(int nid
)
894 struct pglist_data
*pgdat
;
896 pgdat
= NODE_DATA(nid
);
898 pgdat
= arch_alloc_nodedata(nid
);
902 pgdat
->per_cpu_nodestats
=
903 alloc_percpu(struct per_cpu_nodestat
);
904 arch_refresh_nodedata(nid
, pgdat
);
908 * Reset the nr_zones, order and highest_zoneidx before reuse.
909 * Note that kswapd will init kswapd_highest_zoneidx properly
910 * when it starts in the near future.
913 pgdat
->kswapd_order
= 0;
914 pgdat
->kswapd_highest_zoneidx
= 0;
915 for_each_online_cpu(cpu
) {
916 struct per_cpu_nodestat
*p
;
918 p
= per_cpu_ptr(pgdat
->per_cpu_nodestats
, cpu
);
919 memset(p
, 0, sizeof(*p
));
923 /* we can use NODE_DATA(nid) from here */
924 pgdat
->node_id
= nid
;
925 pgdat
->node_start_pfn
= 0;
927 /* init node's zones as empty zones, we don't have any present pages.*/
928 free_area_init_core_hotplug(nid
);
931 * The node we allocated has no zone fallback lists. For avoiding
932 * to access not-initialized zonelist, build here.
934 build_all_zonelists(pgdat
);
937 * When memory is hot-added, all the memory is in offline state. So
938 * clear all zones' present_pages because they will be updated in
939 * online_pages() and offline_pages().
941 reset_node_managed_pages(pgdat
);
942 reset_node_present_pages(pgdat
);
947 static void rollback_node_hotadd(int nid
)
949 pg_data_t
*pgdat
= NODE_DATA(nid
);
951 arch_refresh_nodedata(nid
, NULL
);
952 free_percpu(pgdat
->per_cpu_nodestats
);
953 arch_free_nodedata(pgdat
);
958 * try_online_node - online a node if offlined
960 * @set_node_online: Whether we want to online the node
961 * called by cpu_up() to online a node without onlined memory.
964 * 1 -> a new node has been allocated
965 * 0 -> the node is already online
966 * -ENOMEM -> the node could not be allocated
968 static int __try_online_node(int nid
, bool set_node_online
)
973 if (node_online(nid
))
976 pgdat
= hotadd_new_pgdat(nid
);
978 pr_err("Cannot online node %d due to NULL pgdat\n", nid
);
983 if (set_node_online
) {
984 node_set_online(nid
);
985 ret
= register_one_node(nid
);
993 * Users of this function always want to online/register the node
995 int try_online_node(int nid
)
1000 ret
= __try_online_node(nid
, true);
1005 static int check_hotplug_memory_range(u64 start
, u64 size
)
1007 /* memory range must be block size aligned */
1008 if (!size
|| !IS_ALIGNED(start
, memory_block_size_bytes()) ||
1009 !IS_ALIGNED(size
, memory_block_size_bytes())) {
1010 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1011 memory_block_size_bytes(), start
, size
);
1018 static int online_memory_block(struct memory_block
*mem
, void *arg
)
1020 mem
->online_type
= memhp_default_online_type
;
1021 return device_online(&mem
->dev
);
1025 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1026 * and online/offline operations (triggered e.g. by sysfs).
1028 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1030 int __ref
add_memory_resource(int nid
, struct resource
*res
)
1032 struct mhp_params params
= { .pgprot
= PAGE_KERNEL
};
1034 bool new_node
= false;
1038 size
= resource_size(res
);
1040 ret
= check_hotplug_memory_range(start
, size
);
1044 if (!node_possible(nid
)) {
1045 WARN(1, "node %d was absent from the node_possible_map\n", nid
);
1049 mem_hotplug_begin();
1051 if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK
))
1052 memblock_add_node(start
, size
, nid
);
1054 ret
= __try_online_node(nid
, false);
1059 /* call arch's memory hotadd */
1060 ret
= arch_add_memory(nid
, start
, size
, ¶ms
);
1064 /* create memory block devices after memory was added */
1065 ret
= create_memory_block_devices(start
, size
);
1067 arch_remove_memory(nid
, start
, size
, NULL
);
1072 /* If sysfs file of new node can't be created, cpu on the node
1073 * can't be hot-added. There is no rollback way now.
1074 * So, check by BUG_ON() to catch it reluctantly..
1075 * We online node here. We can't roll back from here.
1077 node_set_online(nid
);
1078 ret
= __register_one_node(nid
);
1082 /* link memory sections under this node.*/
1083 ret
= link_mem_sections(nid
, PFN_DOWN(start
), PFN_UP(start
+ size
- 1),
1087 /* create new memmap entry */
1088 if (!strcmp(res
->name
, "System RAM"))
1089 firmware_map_add_hotplug(start
, start
+ size
, "System RAM");
1091 /* device_online() will take the lock when calling online_pages() */
1094 /* online pages if requested */
1095 if (memhp_default_online_type
!= MMOP_OFFLINE
)
1096 walk_memory_blocks(start
, size
, NULL
, online_memory_block
);
1100 /* rollback pgdat allocation and others */
1102 rollback_node_hotadd(nid
);
1103 if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK
))
1104 memblock_remove(start
, size
);
1109 /* requires device_hotplug_lock, see add_memory_resource() */
1110 int __ref
__add_memory(int nid
, u64 start
, u64 size
)
1112 struct resource
*res
;
1115 res
= register_memory_resource(start
, size
, "System RAM");
1117 return PTR_ERR(res
);
1119 ret
= add_memory_resource(nid
, res
);
1121 release_memory_resource(res
);
1125 int add_memory(int nid
, u64 start
, u64 size
)
1129 lock_device_hotplug();
1130 rc
= __add_memory(nid
, start
, size
);
1131 unlock_device_hotplug();
1135 EXPORT_SYMBOL_GPL(add_memory
);
1138 * Add special, driver-managed memory to the system as system RAM. Such
1139 * memory is not exposed via the raw firmware-provided memmap as system
1140 * RAM, instead, it is detected and added by a driver - during cold boot,
1141 * after a reboot, and after kexec.
1143 * Reasons why this memory should not be used for the initial memmap of a
1144 * kexec kernel or for placing kexec images:
1145 * - The booting kernel is in charge of determining how this memory will be
1146 * used (e.g., use persistent memory as system RAM)
1147 * - Coordination with a hypervisor is required before this memory
1148 * can be used (e.g., inaccessible parts).
1150 * For this memory, no entries in /sys/firmware/memmap ("raw firmware-provided
1151 * memory map") are created. Also, the created memory resource is flagged
1152 * with IORESOURCE_MEM_DRIVER_MANAGED, so in-kernel users can special-case
1153 * this memory as well (esp., not place kexec images onto it).
1155 * The resource_name (visible via /proc/iomem) has to have the format
1156 * "System RAM ($DRIVER)".
1158 int add_memory_driver_managed(int nid
, u64 start
, u64 size
,
1159 const char *resource_name
)
1161 struct resource
*res
;
1164 if (!resource_name
||
1165 strstr(resource_name
, "System RAM (") != resource_name
||
1166 resource_name
[strlen(resource_name
) - 1] != ')')
1169 lock_device_hotplug();
1171 res
= register_memory_resource(start
, size
, resource_name
);
1177 rc
= add_memory_resource(nid
, res
);
1179 release_memory_resource(res
);
1182 unlock_device_hotplug();
1185 EXPORT_SYMBOL_GPL(add_memory_driver_managed
);
1187 #ifdef CONFIG_MEMORY_HOTREMOVE
1189 * Confirm all pages in a range [start, end) belong to the same zone (skipping
1190 * memory holes). When true, return the zone.
1192 struct zone
*test_pages_in_a_zone(unsigned long start_pfn
,
1193 unsigned long end_pfn
)
1195 unsigned long pfn
, sec_end_pfn
;
1196 struct zone
*zone
= NULL
;
1199 for (pfn
= start_pfn
, sec_end_pfn
= SECTION_ALIGN_UP(start_pfn
+ 1);
1201 pfn
= sec_end_pfn
, sec_end_pfn
+= PAGES_PER_SECTION
) {
1202 /* Make sure the memory section is present first */
1203 if (!present_section_nr(pfn_to_section_nr(pfn
)))
1205 for (; pfn
< sec_end_pfn
&& pfn
< end_pfn
;
1206 pfn
+= MAX_ORDER_NR_PAGES
) {
1208 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1209 while ((i
< MAX_ORDER_NR_PAGES
) &&
1210 !pfn_valid_within(pfn
+ i
))
1212 if (i
== MAX_ORDER_NR_PAGES
|| pfn
+ i
>= end_pfn
)
1214 /* Check if we got outside of the zone */
1215 if (zone
&& !zone_spans_pfn(zone
, pfn
+ i
))
1217 page
= pfn_to_page(pfn
+ i
);
1218 if (zone
&& page_zone(page
) != zone
)
1220 zone
= page_zone(page
);
1228 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1229 * non-lru movable pages and hugepages). Will skip over most unmovable
1230 * pages (esp., pages that can be skipped when offlining), but bail out on
1231 * definitely unmovable pages.
1234 * 0 in case a movable page is found and movable_pfn was updated.
1235 * -ENOENT in case no movable page was found.
1236 * -EBUSY in case a definitely unmovable page was found.
1238 static int scan_movable_pages(unsigned long start
, unsigned long end
,
1239 unsigned long *movable_pfn
)
1243 for (pfn
= start
; pfn
< end
; pfn
++) {
1244 struct page
*page
, *head
;
1247 if (!pfn_valid(pfn
))
1249 page
= pfn_to_page(pfn
);
1252 if (__PageMovable(page
))
1256 * PageOffline() pages that are not marked __PageMovable() and
1257 * have a reference count > 0 (after MEM_GOING_OFFLINE) are
1258 * definitely unmovable. If their reference count would be 0,
1259 * they could at least be skipped when offlining memory.
1261 if (PageOffline(page
) && page_count(page
))
1264 if (!PageHuge(page
))
1266 head
= compound_head(page
);
1267 if (page_huge_active(head
))
1269 skip
= compound_nr(head
) - (page
- head
);
1278 static struct page
*new_node_page(struct page
*page
, unsigned long private)
1280 nodemask_t nmask
= node_states
[N_MEMORY
];
1281 struct migration_target_control mtc
= {
1282 .nid
= page_to_nid(page
),
1284 .gfp_mask
= GFP_USER
| __GFP_MOVABLE
| __GFP_RETRY_MAYFAIL
,
1288 * try to allocate from a different node but reuse this node if there
1289 * are no other online nodes to be used (e.g. we are offlining a part
1290 * of the only existing node)
1292 node_clear(mtc
.nid
, nmask
);
1293 if (nodes_empty(nmask
))
1294 node_set(mtc
.nid
, nmask
);
1296 return alloc_migration_target(page
, (unsigned long)&mtc
);
1300 do_migrate_range(unsigned long start_pfn
, unsigned long end_pfn
)
1303 struct page
*page
, *head
;
1307 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
1308 if (!pfn_valid(pfn
))
1310 page
= pfn_to_page(pfn
);
1311 head
= compound_head(page
);
1313 if (PageHuge(page
)) {
1314 pfn
= page_to_pfn(head
) + compound_nr(head
) - 1;
1315 isolate_huge_page(head
, &source
);
1317 } else if (PageTransHuge(page
))
1318 pfn
= page_to_pfn(head
) + thp_nr_pages(page
) - 1;
1321 * HWPoison pages have elevated reference counts so the migration would
1322 * fail on them. It also doesn't make any sense to migrate them in the
1323 * first place. Still try to unmap such a page in case it is still mapped
1324 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1325 * the unmap as the catch all safety net).
1327 if (PageHWPoison(page
)) {
1328 if (WARN_ON(PageLRU(page
)))
1329 isolate_lru_page(page
);
1330 if (page_mapped(page
))
1331 try_to_unmap(page
, TTU_IGNORE_MLOCK
| TTU_IGNORE_ACCESS
);
1335 if (!get_page_unless_zero(page
))
1338 * We can skip free pages. And we can deal with pages on
1339 * LRU and non-lru movable pages.
1342 ret
= isolate_lru_page(page
);
1344 ret
= isolate_movable_page(page
, ISOLATE_UNEVICTABLE
);
1345 if (!ret
) { /* Success */
1346 list_add_tail(&page
->lru
, &source
);
1347 if (!__PageMovable(page
))
1348 inc_node_page_state(page
, NR_ISOLATED_ANON
+
1349 page_is_file_lru(page
));
1352 pr_warn("failed to isolate pfn %lx\n", pfn
);
1353 dump_page(page
, "isolation failed");
1357 if (!list_empty(&source
)) {
1358 /* Allocate a new page from the nearest neighbor node */
1359 ret
= migrate_pages(&source
, new_node_page
, NULL
, 0,
1360 MIGRATE_SYNC
, MR_MEMORY_HOTPLUG
);
1362 list_for_each_entry(page
, &source
, lru
) {
1363 pr_warn("migrating pfn %lx failed ret:%d ",
1364 page_to_pfn(page
), ret
);
1365 dump_page(page
, "migration failure");
1367 putback_movable_pages(&source
);
1374 /* Mark all sections offline and remove all free pages from the buddy. */
1376 offline_isolated_pages_cb(unsigned long start
, unsigned long nr_pages
,
1379 unsigned long *offlined_pages
= (unsigned long *)data
;
1381 *offlined_pages
+= __offline_isolated_pages(start
, start
+ nr_pages
);
1386 * Check all pages in range, recorded as memory resource, are isolated.
1389 check_pages_isolated_cb(unsigned long start_pfn
, unsigned long nr_pages
,
1392 return test_pages_isolated(start_pfn
, start_pfn
+ nr_pages
,
1396 static int __init
cmdline_parse_movable_node(char *p
)
1398 movable_node_enabled
= true;
1401 early_param("movable_node", cmdline_parse_movable_node
);
1403 /* check which state of node_states will be changed when offline memory */
1404 static void node_states_check_changes_offline(unsigned long nr_pages
,
1405 struct zone
*zone
, struct memory_notify
*arg
)
1407 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1408 unsigned long present_pages
= 0;
1411 arg
->status_change_nid
= NUMA_NO_NODE
;
1412 arg
->status_change_nid_normal
= NUMA_NO_NODE
;
1413 arg
->status_change_nid_high
= NUMA_NO_NODE
;
1416 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1417 * If the memory to be offline is within the range
1418 * [0..ZONE_NORMAL], and it is the last present memory there,
1419 * the zones in that range will become empty after the offlining,
1420 * thus we can determine that we need to clear the node from
1421 * node_states[N_NORMAL_MEMORY].
1423 for (zt
= 0; zt
<= ZONE_NORMAL
; zt
++)
1424 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1425 if (zone_idx(zone
) <= ZONE_NORMAL
&& nr_pages
>= present_pages
)
1426 arg
->status_change_nid_normal
= zone_to_nid(zone
);
1428 #ifdef CONFIG_HIGHMEM
1430 * node_states[N_HIGH_MEMORY] contains nodes which
1431 * have normal memory or high memory.
1432 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1433 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1434 * we determine that the zones in that range become empty,
1435 * we need to clear the node for N_HIGH_MEMORY.
1437 present_pages
+= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1438 if (zone_idx(zone
) <= ZONE_HIGHMEM
&& nr_pages
>= present_pages
)
1439 arg
->status_change_nid_high
= zone_to_nid(zone
);
1443 * We have accounted the pages from [0..ZONE_NORMAL), and
1444 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1446 * Here we count the possible pages from ZONE_MOVABLE.
1447 * If after having accounted all the pages, we see that the nr_pages
1448 * to be offlined is over or equal to the accounted pages,
1449 * we know that the node will become empty, and so, we can clear
1450 * it for N_MEMORY as well.
1452 present_pages
+= pgdat
->node_zones
[ZONE_MOVABLE
].present_pages
;
1454 if (nr_pages
>= present_pages
)
1455 arg
->status_change_nid
= zone_to_nid(zone
);
1458 static void node_states_clear_node(int node
, struct memory_notify
*arg
)
1460 if (arg
->status_change_nid_normal
>= 0)
1461 node_clear_state(node
, N_NORMAL_MEMORY
);
1463 if (arg
->status_change_nid_high
>= 0)
1464 node_clear_state(node
, N_HIGH_MEMORY
);
1466 if (arg
->status_change_nid
>= 0)
1467 node_clear_state(node
, N_MEMORY
);
1470 static int count_system_ram_pages_cb(unsigned long start_pfn
,
1471 unsigned long nr_pages
, void *data
)
1473 unsigned long *nr_system_ram_pages
= data
;
1475 *nr_system_ram_pages
+= nr_pages
;
1479 static int __ref
__offline_pages(unsigned long start_pfn
,
1480 unsigned long end_pfn
)
1482 unsigned long pfn
, nr_pages
= 0;
1483 unsigned long offlined_pages
= 0;
1484 int ret
, node
, nr_isolate_pageblock
;
1485 unsigned long flags
;
1487 struct memory_notify arg
;
1490 mem_hotplug_begin();
1493 * Don't allow to offline memory blocks that contain holes.
1494 * Consequently, memory blocks with holes can never get onlined
1495 * via the hotplug path - online_pages() - as hotplugged memory has
1496 * no holes. This way, we e.g., don't have to worry about marking
1497 * memory holes PG_reserved, don't need pfn_valid() checks, and can
1498 * avoid using walk_system_ram_range() later.
1500 walk_system_ram_range(start_pfn
, end_pfn
- start_pfn
, &nr_pages
,
1501 count_system_ram_pages_cb
);
1502 if (nr_pages
!= end_pfn
- start_pfn
) {
1504 reason
= "memory holes";
1505 goto failed_removal
;
1508 /* This makes hotplug much easier...and readable.
1509 we assume this for now. .*/
1510 zone
= test_pages_in_a_zone(start_pfn
, end_pfn
);
1513 reason
= "multizone range";
1514 goto failed_removal
;
1516 node
= zone_to_nid(zone
);
1518 /* set above range as isolated */
1519 ret
= start_isolate_page_range(start_pfn
, end_pfn
,
1521 MEMORY_OFFLINE
| REPORT_FAILURE
);
1523 reason
= "failure to isolate range";
1524 goto failed_removal
;
1526 nr_isolate_pageblock
= ret
;
1528 arg
.start_pfn
= start_pfn
;
1529 arg
.nr_pages
= nr_pages
;
1530 node_states_check_changes_offline(nr_pages
, zone
, &arg
);
1532 ret
= memory_notify(MEM_GOING_OFFLINE
, &arg
);
1533 ret
= notifier_to_errno(ret
);
1535 reason
= "notifier failure";
1536 goto failed_removal_isolated
;
1542 if (signal_pending(current
)) {
1544 reason
= "signal backoff";
1545 goto failed_removal_isolated
;
1549 lru_add_drain_all();
1551 ret
= scan_movable_pages(pfn
, end_pfn
, &pfn
);
1554 * TODO: fatal migration failures should bail
1557 do_migrate_range(pfn
, end_pfn
);
1561 if (ret
!= -ENOENT
) {
1562 reason
= "unmovable page";
1563 goto failed_removal_isolated
;
1567 * Dissolve free hugepages in the memory block before doing
1568 * offlining actually in order to make hugetlbfs's object
1569 * counting consistent.
1571 ret
= dissolve_free_huge_pages(start_pfn
, end_pfn
);
1573 reason
= "failure to dissolve huge pages";
1574 goto failed_removal_isolated
;
1577 ret
= walk_system_ram_range(start_pfn
, end_pfn
- start_pfn
,
1578 NULL
, check_pages_isolated_cb
);
1580 * per-cpu pages are drained in start_isolate_page_range, but if
1581 * there are still pages that are not free, make sure that we
1582 * drain again, because when we isolated range we might
1583 * have raced with another thread that was adding pages to pcp
1586 * Forward progress should be still guaranteed because
1587 * pages on the pcp list can only belong to MOVABLE_ZONE
1588 * because has_unmovable_pages explicitly checks for
1589 * PageBuddy on freed pages on other zones.
1592 drain_all_pages(zone
);
1595 /* Ok, all of our target is isolated.
1596 We cannot do rollback at this point. */
1597 walk_system_ram_range(start_pfn
, end_pfn
- start_pfn
,
1598 &offlined_pages
, offline_isolated_pages_cb
);
1599 pr_info("Offlined Pages %ld\n", offlined_pages
);
1601 * Onlining will reset pagetype flags and makes migrate type
1602 * MOVABLE, so just need to decrease the number of isolated
1603 * pageblocks zone counter here.
1605 spin_lock_irqsave(&zone
->lock
, flags
);
1606 zone
->nr_isolate_pageblock
-= nr_isolate_pageblock
;
1607 spin_unlock_irqrestore(&zone
->lock
, flags
);
1609 /* removal success */
1610 adjust_managed_page_count(pfn_to_page(start_pfn
), -offlined_pages
);
1611 zone
->present_pages
-= offlined_pages
;
1613 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
1614 zone
->zone_pgdat
->node_present_pages
-= offlined_pages
;
1615 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
1617 init_per_zone_wmark_min();
1619 if (!populated_zone(zone
)) {
1620 zone_pcp_reset(zone
);
1621 build_all_zonelists(NULL
);
1623 zone_pcp_update(zone
);
1625 node_states_clear_node(node
, &arg
);
1626 if (arg
.status_change_nid
>= 0) {
1628 kcompactd_stop(node
);
1631 writeback_set_ratelimit();
1633 memory_notify(MEM_OFFLINE
, &arg
);
1634 remove_pfn_range_from_zone(zone
, start_pfn
, nr_pages
);
1638 failed_removal_isolated
:
1639 undo_isolate_page_range(start_pfn
, end_pfn
, MIGRATE_MOVABLE
);
1640 memory_notify(MEM_CANCEL_OFFLINE
, &arg
);
1642 pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1643 (unsigned long long) start_pfn
<< PAGE_SHIFT
,
1644 ((unsigned long long) end_pfn
<< PAGE_SHIFT
) - 1,
1646 /* pushback to free area */
1651 int offline_pages(unsigned long start_pfn
, unsigned long nr_pages
)
1653 return __offline_pages(start_pfn
, start_pfn
+ nr_pages
);
1656 static int check_memblock_offlined_cb(struct memory_block
*mem
, void *arg
)
1658 int ret
= !is_memblock_offlined(mem
);
1660 if (unlikely(ret
)) {
1661 phys_addr_t beginpa
, endpa
;
1663 beginpa
= PFN_PHYS(section_nr_to_pfn(mem
->start_section_nr
));
1664 endpa
= beginpa
+ memory_block_size_bytes() - 1;
1665 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1673 static int check_cpu_on_node(pg_data_t
*pgdat
)
1677 for_each_present_cpu(cpu
) {
1678 if (cpu_to_node(cpu
) == pgdat
->node_id
)
1680 * the cpu on this node isn't removed, and we can't
1681 * offline this node.
1689 static int check_no_memblock_for_node_cb(struct memory_block
*mem
, void *arg
)
1691 int nid
= *(int *)arg
;
1694 * If a memory block belongs to multiple nodes, the stored nid is not
1695 * reliable. However, such blocks are always online (e.g., cannot get
1696 * offlined) and, therefore, are still spanned by the node.
1698 return mem
->nid
== nid
? -EEXIST
: 0;
1705 * Offline a node if all memory sections and cpus of the node are removed.
1707 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1708 * and online/offline operations before this call.
1710 void try_offline_node(int nid
)
1712 pg_data_t
*pgdat
= NODE_DATA(nid
);
1716 * If the node still spans pages (especially ZONE_DEVICE), don't
1717 * offline it. A node spans memory after move_pfn_range_to_zone(),
1718 * e.g., after the memory block was onlined.
1720 if (pgdat
->node_spanned_pages
)
1724 * Especially offline memory blocks might not be spanned by the
1725 * node. They will get spanned by the node once they get onlined.
1726 * However, they link to the node in sysfs and can get onlined later.
1728 rc
= for_each_memory_block(&nid
, check_no_memblock_for_node_cb
);
1732 if (check_cpu_on_node(pgdat
))
1736 * all memory/cpu of this node are removed, we can offline this
1739 node_set_offline(nid
);
1740 unregister_one_node(nid
);
1742 EXPORT_SYMBOL(try_offline_node
);
1744 static void __release_memory_resource(resource_size_t start
,
1745 resource_size_t size
)
1750 * When removing memory in the same granularity as it was added,
1751 * this function never fails. It might only fail if resources
1752 * have to be adjusted or split. We'll ignore the error, as
1753 * removing of memory cannot fail.
1755 ret
= release_mem_region_adjustable(&iomem_resource
, start
, size
);
1757 resource_size_t endres
= start
+ size
- 1;
1759 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
1760 &start
, &endres
, ret
);
1764 static int __ref
try_remove_memory(int nid
, u64 start
, u64 size
)
1768 BUG_ON(check_hotplug_memory_range(start
, size
));
1771 * All memory blocks must be offlined before removing memory. Check
1772 * whether all memory blocks in question are offline and return error
1773 * if this is not the case.
1775 rc
= walk_memory_blocks(start
, size
, NULL
, check_memblock_offlined_cb
);
1779 /* remove memmap entry */
1780 firmware_map_remove(start
, start
+ size
, "System RAM");
1783 * Memory block device removal under the device_hotplug_lock is
1784 * a barrier against racing online attempts.
1786 remove_memory_block_devices(start
, size
);
1788 mem_hotplug_begin();
1790 arch_remove_memory(nid
, start
, size
, NULL
);
1792 if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK
)) {
1793 memblock_free(start
, size
);
1794 memblock_remove(start
, size
);
1797 __release_memory_resource(start
, size
);
1799 try_offline_node(nid
);
1808 * @start: physical address of the region to remove
1809 * @size: size of the region to remove
1811 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1812 * and online/offline operations before this call, as required by
1813 * try_offline_node().
1815 void __remove_memory(int nid
, u64 start
, u64 size
)
1819 * trigger BUG() if some memory is not offlined prior to calling this
1822 if (try_remove_memory(nid
, start
, size
))
1827 * Remove memory if every memory block is offline, otherwise return -EBUSY is
1828 * some memory is not offline
1830 int remove_memory(int nid
, u64 start
, u64 size
)
1834 lock_device_hotplug();
1835 rc
= try_remove_memory(nid
, start
, size
);
1836 unlock_device_hotplug();
1840 EXPORT_SYMBOL_GPL(remove_memory
);
1843 * Try to offline and remove a memory block. Might take a long time to
1844 * finish in case memory is still in use. Primarily useful for memory devices
1845 * that logically unplugged all memory (so it's no longer in use) and want to
1846 * offline + remove the memory block.
1848 int offline_and_remove_memory(int nid
, u64 start
, u64 size
)
1850 struct memory_block
*mem
;
1853 if (!IS_ALIGNED(start
, memory_block_size_bytes()) ||
1854 size
!= memory_block_size_bytes())
1857 lock_device_hotplug();
1858 mem
= find_memory_block(__pfn_to_section(PFN_DOWN(start
)));
1860 rc
= device_offline(&mem
->dev
);
1861 /* Ignore if the device is already offline. */
1866 * In case we succeeded to offline the memory block, remove it.
1867 * This cannot fail as it cannot get onlined in the meantime.
1870 rc
= try_remove_memory(nid
, start
, size
);
1873 unlock_device_hotplug();
1877 EXPORT_SYMBOL_GPL(offline_and_remove_memory
);
1878 #endif /* CONFIG_MEMORY_HOTREMOVE */