1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/mm/memory_hotplug.c
8 #include <linux/stddef.h>
10 #include <linux/sched/signal.h>
11 #include <linux/swap.h>
12 #include <linux/interrupt.h>
13 #include <linux/pagemap.h>
14 #include <linux/compiler.h>
15 #include <linux/export.h>
16 #include <linux/pagevec.h>
17 #include <linux/writeback.h>
18 #include <linux/slab.h>
19 #include <linux/sysctl.h>
20 #include <linux/cpu.h>
21 #include <linux/memory.h>
22 #include <linux/memremap.h>
23 #include <linux/memory_hotplug.h>
24 #include <linux/highmem.h>
25 #include <linux/vmalloc.h>
26 #include <linux/ioport.h>
27 #include <linux/delay.h>
28 #include <linux/migrate.h>
29 #include <linux/page-isolation.h>
30 #include <linux/pfn.h>
31 #include <linux/suspend.h>
32 #include <linux/mm_inline.h>
33 #include <linux/firmware-map.h>
34 #include <linux/stop_machine.h>
35 #include <linux/hugetlb.h>
36 #include <linux/memblock.h>
37 #include <linux/compaction.h>
38 #include <linux/rmap.h>
40 #include <asm/tlbflush.h>
46 * online_page_callback contains pointer to current page onlining function.
47 * Initially it is generic_online_page(). If it is required it could be
48 * changed by calling set_online_page_callback() for callback registration
49 * and restore_online_page_callback() for generic callback restore.
52 static void generic_online_page(struct page
*page
, unsigned int order
);
54 static online_page_callback_t online_page_callback
= generic_online_page
;
55 static DEFINE_MUTEX(online_page_callback_lock
);
57 DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock
);
59 void get_online_mems(void)
61 percpu_down_read(&mem_hotplug_lock
);
64 void put_online_mems(void)
66 percpu_up_read(&mem_hotplug_lock
);
69 bool movable_node_enabled
= false;
71 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
72 bool memhp_auto_online
;
74 bool memhp_auto_online
= true;
76 EXPORT_SYMBOL_GPL(memhp_auto_online
);
78 static int __init
setup_memhp_default_state(char *str
)
80 if (!strcmp(str
, "online"))
81 memhp_auto_online
= true;
82 else if (!strcmp(str
, "offline"))
83 memhp_auto_online
= false;
87 __setup("memhp_default_state=", setup_memhp_default_state
);
89 void mem_hotplug_begin(void)
92 percpu_down_write(&mem_hotplug_lock
);
95 void mem_hotplug_done(void)
97 percpu_up_write(&mem_hotplug_lock
);
101 u64 max_mem_size
= U64_MAX
;
103 /* add this memory to iomem resource */
104 static struct resource
*register_memory_resource(u64 start
, u64 size
)
106 struct resource
*res
;
107 unsigned long flags
= IORESOURCE_SYSTEM_RAM
| IORESOURCE_BUSY
;
108 char *resource_name
= "System RAM";
110 if (start
+ size
> max_mem_size
)
111 return ERR_PTR(-E2BIG
);
114 * Request ownership of the new memory range. This might be
115 * a child of an existing resource that was present but
116 * not marked as busy.
118 res
= __request_region(&iomem_resource
, start
, size
,
119 resource_name
, flags
);
122 pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n",
123 start
, start
+ size
);
124 return ERR_PTR(-EEXIST
);
129 static void release_memory_resource(struct resource
*res
)
133 release_resource(res
);
138 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
139 void get_page_bootmem(unsigned long info
, struct page
*page
,
142 page
->freelist
= (void *)type
;
143 SetPagePrivate(page
);
144 set_page_private(page
, info
);
148 void put_page_bootmem(struct page
*page
)
152 type
= (unsigned long) page
->freelist
;
153 BUG_ON(type
< MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE
||
154 type
> MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE
);
156 if (page_ref_dec_return(page
) == 1) {
157 page
->freelist
= NULL
;
158 ClearPagePrivate(page
);
159 set_page_private(page
, 0);
160 INIT_LIST_HEAD(&page
->lru
);
161 free_reserved_page(page
);
165 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
166 #ifndef CONFIG_SPARSEMEM_VMEMMAP
167 static void register_page_bootmem_info_section(unsigned long start_pfn
)
169 unsigned long *usemap
, mapsize
, section_nr
, i
;
170 struct mem_section
*ms
;
171 struct page
*page
, *memmap
;
173 section_nr
= pfn_to_section_nr(start_pfn
);
174 ms
= __nr_to_section(section_nr
);
176 /* Get section's memmap address */
177 memmap
= sparse_decode_mem_map(ms
->section_mem_map
, section_nr
);
180 * Get page for the memmap's phys address
181 * XXX: need more consideration for sparse_vmemmap...
183 page
= virt_to_page(memmap
);
184 mapsize
= sizeof(struct page
) * PAGES_PER_SECTION
;
185 mapsize
= PAGE_ALIGN(mapsize
) >> PAGE_SHIFT
;
187 /* remember memmap's page */
188 for (i
= 0; i
< mapsize
; i
++, page
++)
189 get_page_bootmem(section_nr
, page
, SECTION_INFO
);
191 usemap
= ms
->pageblock_flags
;
192 page
= virt_to_page(usemap
);
194 mapsize
= PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT
;
196 for (i
= 0; i
< mapsize
; i
++, page
++)
197 get_page_bootmem(section_nr
, page
, MIX_SECTION_INFO
);
200 #else /* CONFIG_SPARSEMEM_VMEMMAP */
201 static void register_page_bootmem_info_section(unsigned long start_pfn
)
203 unsigned long *usemap
, mapsize
, section_nr
, i
;
204 struct mem_section
*ms
;
205 struct page
*page
, *memmap
;
207 section_nr
= pfn_to_section_nr(start_pfn
);
208 ms
= __nr_to_section(section_nr
);
210 memmap
= sparse_decode_mem_map(ms
->section_mem_map
, section_nr
);
212 register_page_bootmem_memmap(section_nr
, memmap
, PAGES_PER_SECTION
);
214 usemap
= ms
->pageblock_flags
;
215 page
= virt_to_page(usemap
);
217 mapsize
= PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT
;
219 for (i
= 0; i
< mapsize
; i
++, page
++)
220 get_page_bootmem(section_nr
, page
, MIX_SECTION_INFO
);
222 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
224 void __init
register_page_bootmem_info_node(struct pglist_data
*pgdat
)
226 unsigned long i
, pfn
, end_pfn
, nr_pages
;
227 int node
= pgdat
->node_id
;
230 nr_pages
= PAGE_ALIGN(sizeof(struct pglist_data
)) >> PAGE_SHIFT
;
231 page
= virt_to_page(pgdat
);
233 for (i
= 0; i
< nr_pages
; i
++, page
++)
234 get_page_bootmem(node
, page
, NODE_INFO
);
236 pfn
= pgdat
->node_start_pfn
;
237 end_pfn
= pgdat_end_pfn(pgdat
);
239 /* register section info */
240 for (; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
242 * Some platforms can assign the same pfn to multiple nodes - on
243 * node0 as well as nodeN. To avoid registering a pfn against
244 * multiple nodes we check that this pfn does not already
245 * reside in some other nodes.
247 if (pfn_valid(pfn
) && (early_pfn_to_nid(pfn
) == node
))
248 register_page_bootmem_info_section(pfn
);
251 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
253 static int __meminit
__add_section(int nid
, unsigned long phys_start_pfn
,
254 struct vmem_altmap
*altmap
)
258 if (pfn_valid(phys_start_pfn
))
261 ret
= sparse_add_one_section(nid
, phys_start_pfn
, altmap
);
262 return ret
< 0 ? ret
: 0;
266 * Reasonably generic function for adding memory. It is
267 * expected that archs that support memory hotplug will
268 * call this function after deciding the zone to which to
271 int __ref
__add_pages(int nid
, unsigned long phys_start_pfn
,
272 unsigned long nr_pages
, struct mhp_restrictions
*restrictions
)
276 int start_sec
, end_sec
;
277 struct vmem_altmap
*altmap
= restrictions
->altmap
;
279 /* during initialize mem_map, align hot-added range to section */
280 start_sec
= pfn_to_section_nr(phys_start_pfn
);
281 end_sec
= pfn_to_section_nr(phys_start_pfn
+ nr_pages
- 1);
285 * Validate altmap is within bounds of the total request
287 if (altmap
->base_pfn
!= phys_start_pfn
288 || vmem_altmap_offset(altmap
) > nr_pages
) {
289 pr_warn_once("memory add fail, invalid altmap\n");
296 for (i
= start_sec
; i
<= end_sec
; i
++) {
297 err
= __add_section(nid
, section_nr_to_pfn(i
), altmap
);
300 * EEXIST is finally dealt with by ioresource collision
301 * check. see add_memory() => register_memory_resource()
302 * Warning will be printed if there is collision.
304 if (err
&& (err
!= -EEXIST
))
309 vmemmap_populate_print_last();
314 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
315 static unsigned long find_smallest_section_pfn(int nid
, struct zone
*zone
,
316 unsigned long start_pfn
,
317 unsigned long end_pfn
)
319 struct mem_section
*ms
;
321 for (; start_pfn
< end_pfn
; start_pfn
+= PAGES_PER_SECTION
) {
322 ms
= __pfn_to_section(start_pfn
);
324 if (unlikely(!valid_section(ms
)))
327 if (unlikely(pfn_to_nid(start_pfn
) != nid
))
330 if (zone
&& zone
!= page_zone(pfn_to_page(start_pfn
)))
339 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
340 static unsigned long find_biggest_section_pfn(int nid
, struct zone
*zone
,
341 unsigned long start_pfn
,
342 unsigned long end_pfn
)
344 struct mem_section
*ms
;
347 /* pfn is the end pfn of a memory section. */
349 for (; pfn
>= start_pfn
; pfn
-= PAGES_PER_SECTION
) {
350 ms
= __pfn_to_section(pfn
);
352 if (unlikely(!valid_section(ms
)))
355 if (unlikely(pfn_to_nid(pfn
) != nid
))
358 if (zone
&& zone
!= page_zone(pfn_to_page(pfn
)))
367 static void shrink_zone_span(struct zone
*zone
, unsigned long start_pfn
,
368 unsigned long end_pfn
)
370 unsigned long zone_start_pfn
= zone
->zone_start_pfn
;
371 unsigned long z
= zone_end_pfn(zone
); /* zone_end_pfn namespace clash */
372 unsigned long zone_end_pfn
= z
;
374 struct mem_section
*ms
;
375 int nid
= zone_to_nid(zone
);
377 zone_span_writelock(zone
);
378 if (zone_start_pfn
== start_pfn
) {
380 * If the section is smallest section in the zone, it need
381 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
382 * In this case, we find second smallest valid mem_section
383 * for shrinking zone.
385 pfn
= find_smallest_section_pfn(nid
, zone
, end_pfn
,
388 zone
->zone_start_pfn
= pfn
;
389 zone
->spanned_pages
= zone_end_pfn
- pfn
;
391 } else if (zone_end_pfn
== end_pfn
) {
393 * If the section is biggest section in the zone, it need
394 * shrink zone->spanned_pages.
395 * In this case, we find second biggest valid mem_section for
398 pfn
= find_biggest_section_pfn(nid
, zone
, zone_start_pfn
,
401 zone
->spanned_pages
= pfn
- zone_start_pfn
+ 1;
405 * The section is not biggest or smallest mem_section in the zone, it
406 * only creates a hole in the zone. So in this case, we need not
407 * change the zone. But perhaps, the zone has only hole data. Thus
408 * it check the zone has only hole or not.
410 pfn
= zone_start_pfn
;
411 for (; pfn
< zone_end_pfn
; pfn
+= PAGES_PER_SECTION
) {
412 ms
= __pfn_to_section(pfn
);
414 if (unlikely(!valid_section(ms
)))
417 if (page_zone(pfn_to_page(pfn
)) != zone
)
420 /* If the section is current section, it continues the loop */
421 if (start_pfn
== pfn
)
424 /* If we find valid section, we have nothing to do */
425 zone_span_writeunlock(zone
);
429 /* The zone has no valid section */
430 zone
->zone_start_pfn
= 0;
431 zone
->spanned_pages
= 0;
432 zone_span_writeunlock(zone
);
435 static void shrink_pgdat_span(struct pglist_data
*pgdat
,
436 unsigned long start_pfn
, unsigned long end_pfn
)
438 unsigned long pgdat_start_pfn
= pgdat
->node_start_pfn
;
439 unsigned long p
= pgdat_end_pfn(pgdat
); /* pgdat_end_pfn namespace clash */
440 unsigned long pgdat_end_pfn
= p
;
442 struct mem_section
*ms
;
443 int nid
= pgdat
->node_id
;
445 if (pgdat_start_pfn
== start_pfn
) {
447 * If the section is smallest section in the pgdat, it need
448 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
449 * In this case, we find second smallest valid mem_section
450 * for shrinking zone.
452 pfn
= find_smallest_section_pfn(nid
, NULL
, end_pfn
,
455 pgdat
->node_start_pfn
= pfn
;
456 pgdat
->node_spanned_pages
= pgdat_end_pfn
- pfn
;
458 } else if (pgdat_end_pfn
== end_pfn
) {
460 * If the section is biggest section in the pgdat, it need
461 * shrink pgdat->node_spanned_pages.
462 * In this case, we find second biggest valid mem_section for
465 pfn
= find_biggest_section_pfn(nid
, NULL
, pgdat_start_pfn
,
468 pgdat
->node_spanned_pages
= pfn
- pgdat_start_pfn
+ 1;
472 * If the section is not biggest or smallest mem_section in the pgdat,
473 * it only creates a hole in the pgdat. So in this case, we need not
475 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
476 * has only hole or not.
478 pfn
= pgdat_start_pfn
;
479 for (; pfn
< pgdat_end_pfn
; pfn
+= PAGES_PER_SECTION
) {
480 ms
= __pfn_to_section(pfn
);
482 if (unlikely(!valid_section(ms
)))
485 if (pfn_to_nid(pfn
) != nid
)
488 /* If the section is current section, it continues the loop */
489 if (start_pfn
== pfn
)
492 /* If we find valid section, we have nothing to do */
496 /* The pgdat has no valid section */
497 pgdat
->node_start_pfn
= 0;
498 pgdat
->node_spanned_pages
= 0;
501 static void __remove_zone(struct zone
*zone
, unsigned long start_pfn
)
503 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
504 int nr_pages
= PAGES_PER_SECTION
;
507 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
508 shrink_zone_span(zone
, start_pfn
, start_pfn
+ nr_pages
);
509 shrink_pgdat_span(pgdat
, start_pfn
, start_pfn
+ nr_pages
);
510 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
513 static void __remove_section(struct zone
*zone
, struct mem_section
*ms
,
514 unsigned long map_offset
,
515 struct vmem_altmap
*altmap
)
517 unsigned long start_pfn
;
520 if (WARN_ON_ONCE(!valid_section(ms
)))
523 unregister_memory_section(ms
);
525 scn_nr
= __section_nr(ms
);
526 start_pfn
= section_nr_to_pfn((unsigned long)scn_nr
);
527 __remove_zone(zone
, start_pfn
);
529 sparse_remove_one_section(zone
, ms
, map_offset
, altmap
);
533 * __remove_pages() - remove sections of pages from a zone
534 * @zone: zone from which pages need to be removed
535 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
536 * @nr_pages: number of pages to remove (must be multiple of section size)
537 * @altmap: alternative device page map or %NULL if default memmap is used
539 * Generic helper function to remove section mappings and sysfs entries
540 * for the section of the memory we are removing. Caller needs to make
541 * sure that pages are marked reserved and zones are adjust properly by
542 * calling offline_pages().
544 void __remove_pages(struct zone
*zone
, unsigned long phys_start_pfn
,
545 unsigned long nr_pages
, struct vmem_altmap
*altmap
)
548 unsigned long map_offset
= 0;
549 int sections_to_remove
;
551 /* In the ZONE_DEVICE case device driver owns the memory region */
552 if (is_dev_zone(zone
))
553 map_offset
= vmem_altmap_offset(altmap
);
555 clear_zone_contiguous(zone
);
558 * We can only remove entire sections
560 BUG_ON(phys_start_pfn
& ~PAGE_SECTION_MASK
);
561 BUG_ON(nr_pages
% PAGES_PER_SECTION
);
563 sections_to_remove
= nr_pages
/ PAGES_PER_SECTION
;
564 for (i
= 0; i
< sections_to_remove
; i
++) {
565 unsigned long pfn
= phys_start_pfn
+ i
*PAGES_PER_SECTION
;
568 __remove_section(zone
, __pfn_to_section(pfn
), map_offset
,
573 set_zone_contiguous(zone
);
576 int set_online_page_callback(online_page_callback_t callback
)
581 mutex_lock(&online_page_callback_lock
);
583 if (online_page_callback
== generic_online_page
) {
584 online_page_callback
= callback
;
588 mutex_unlock(&online_page_callback_lock
);
593 EXPORT_SYMBOL_GPL(set_online_page_callback
);
595 int restore_online_page_callback(online_page_callback_t callback
)
600 mutex_lock(&online_page_callback_lock
);
602 if (online_page_callback
== callback
) {
603 online_page_callback
= generic_online_page
;
607 mutex_unlock(&online_page_callback_lock
);
612 EXPORT_SYMBOL_GPL(restore_online_page_callback
);
614 void __online_page_set_limits(struct page
*page
)
617 EXPORT_SYMBOL_GPL(__online_page_set_limits
);
619 void __online_page_increment_counters(struct page
*page
)
621 adjust_managed_page_count(page
, 1);
623 EXPORT_SYMBOL_GPL(__online_page_increment_counters
);
625 void __online_page_free(struct page
*page
)
627 __free_reserved_page(page
);
629 EXPORT_SYMBOL_GPL(__online_page_free
);
631 static void generic_online_page(struct page
*page
, unsigned int order
)
633 kernel_map_pages(page
, 1 << order
, 1);
634 __free_pages_core(page
, order
);
635 totalram_pages_add(1UL << order
);
636 #ifdef CONFIG_HIGHMEM
637 if (PageHighMem(page
))
638 totalhigh_pages_add(1UL << order
);
642 static int online_pages_blocks(unsigned long start
, unsigned long nr_pages
)
644 unsigned long end
= start
+ nr_pages
;
645 int order
, onlined_pages
= 0;
647 while (start
< end
) {
648 order
= min(MAX_ORDER
- 1,
649 get_order(PFN_PHYS(end
) - PFN_PHYS(start
)));
650 (*online_page_callback
)(pfn_to_page(start
), order
);
652 onlined_pages
+= (1UL << order
);
653 start
+= (1UL << order
);
655 return onlined_pages
;
658 static int online_pages_range(unsigned long start_pfn
, unsigned long nr_pages
,
661 unsigned long onlined_pages
= *(unsigned long *)arg
;
663 if (PageReserved(pfn_to_page(start_pfn
)))
664 onlined_pages
+= online_pages_blocks(start_pfn
, nr_pages
);
666 online_mem_sections(start_pfn
, start_pfn
+ nr_pages
);
668 *(unsigned long *)arg
= onlined_pages
;
672 /* check which state of node_states will be changed when online memory */
673 static void node_states_check_changes_online(unsigned long nr_pages
,
674 struct zone
*zone
, struct memory_notify
*arg
)
676 int nid
= zone_to_nid(zone
);
678 arg
->status_change_nid
= NUMA_NO_NODE
;
679 arg
->status_change_nid_normal
= NUMA_NO_NODE
;
680 arg
->status_change_nid_high
= NUMA_NO_NODE
;
682 if (!node_state(nid
, N_MEMORY
))
683 arg
->status_change_nid
= nid
;
684 if (zone_idx(zone
) <= ZONE_NORMAL
&& !node_state(nid
, N_NORMAL_MEMORY
))
685 arg
->status_change_nid_normal
= nid
;
686 #ifdef CONFIG_HIGHMEM
687 if (zone_idx(zone
) <= ZONE_HIGHMEM
&& !node_state(nid
, N_HIGH_MEMORY
))
688 arg
->status_change_nid_high
= nid
;
692 static void node_states_set_node(int node
, struct memory_notify
*arg
)
694 if (arg
->status_change_nid_normal
>= 0)
695 node_set_state(node
, N_NORMAL_MEMORY
);
697 if (arg
->status_change_nid_high
>= 0)
698 node_set_state(node
, N_HIGH_MEMORY
);
700 if (arg
->status_change_nid
>= 0)
701 node_set_state(node
, N_MEMORY
);
704 static void __meminit
resize_zone_range(struct zone
*zone
, unsigned long start_pfn
,
705 unsigned long nr_pages
)
707 unsigned long old_end_pfn
= zone_end_pfn(zone
);
709 if (zone_is_empty(zone
) || start_pfn
< zone
->zone_start_pfn
)
710 zone
->zone_start_pfn
= start_pfn
;
712 zone
->spanned_pages
= max(start_pfn
+ nr_pages
, old_end_pfn
) - zone
->zone_start_pfn
;
715 static void __meminit
resize_pgdat_range(struct pglist_data
*pgdat
, unsigned long start_pfn
,
716 unsigned long nr_pages
)
718 unsigned long old_end_pfn
= pgdat_end_pfn(pgdat
);
720 if (!pgdat
->node_spanned_pages
|| start_pfn
< pgdat
->node_start_pfn
)
721 pgdat
->node_start_pfn
= start_pfn
;
723 pgdat
->node_spanned_pages
= max(start_pfn
+ nr_pages
, old_end_pfn
) - pgdat
->node_start_pfn
;
726 void __ref
move_pfn_range_to_zone(struct zone
*zone
, unsigned long start_pfn
,
727 unsigned long nr_pages
, struct vmem_altmap
*altmap
)
729 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
730 int nid
= pgdat
->node_id
;
733 clear_zone_contiguous(zone
);
735 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
736 pgdat_resize_lock(pgdat
, &flags
);
737 zone_span_writelock(zone
);
738 if (zone_is_empty(zone
))
739 init_currently_empty_zone(zone
, start_pfn
, nr_pages
);
740 resize_zone_range(zone
, start_pfn
, nr_pages
);
741 zone_span_writeunlock(zone
);
742 resize_pgdat_range(pgdat
, start_pfn
, nr_pages
);
743 pgdat_resize_unlock(pgdat
, &flags
);
746 * TODO now we have a visible range of pages which are not associated
747 * with their zone properly. Not nice but set_pfnblock_flags_mask
748 * expects the zone spans the pfn range. All the pages in the range
749 * are reserved so nobody should be touching them so we should be safe
751 memmap_init_zone(nr_pages
, nid
, zone_idx(zone
), start_pfn
,
752 MEMMAP_HOTPLUG
, altmap
);
754 set_zone_contiguous(zone
);
758 * Returns a default kernel memory zone for the given pfn range.
759 * If no kernel zone covers this pfn range it will automatically go
760 * to the ZONE_NORMAL.
762 static struct zone
*default_kernel_zone_for_pfn(int nid
, unsigned long start_pfn
,
763 unsigned long nr_pages
)
765 struct pglist_data
*pgdat
= NODE_DATA(nid
);
768 for (zid
= 0; zid
<= ZONE_NORMAL
; zid
++) {
769 struct zone
*zone
= &pgdat
->node_zones
[zid
];
771 if (zone_intersects(zone
, start_pfn
, nr_pages
))
775 return &pgdat
->node_zones
[ZONE_NORMAL
];
778 static inline struct zone
*default_zone_for_pfn(int nid
, unsigned long start_pfn
,
779 unsigned long nr_pages
)
781 struct zone
*kernel_zone
= default_kernel_zone_for_pfn(nid
, start_pfn
,
783 struct zone
*movable_zone
= &NODE_DATA(nid
)->node_zones
[ZONE_MOVABLE
];
784 bool in_kernel
= zone_intersects(kernel_zone
, start_pfn
, nr_pages
);
785 bool in_movable
= zone_intersects(movable_zone
, start_pfn
, nr_pages
);
788 * We inherit the existing zone in a simple case where zones do not
789 * overlap in the given range
791 if (in_kernel
^ in_movable
)
792 return (in_kernel
) ? kernel_zone
: movable_zone
;
795 * If the range doesn't belong to any zone or two zones overlap in the
796 * given range then we use movable zone only if movable_node is
797 * enabled because we always online to a kernel zone by default.
799 return movable_node_enabled
? movable_zone
: kernel_zone
;
802 struct zone
* zone_for_pfn_range(int online_type
, int nid
, unsigned start_pfn
,
803 unsigned long nr_pages
)
805 if (online_type
== MMOP_ONLINE_KERNEL
)
806 return default_kernel_zone_for_pfn(nid
, start_pfn
, nr_pages
);
808 if (online_type
== MMOP_ONLINE_MOVABLE
)
809 return &NODE_DATA(nid
)->node_zones
[ZONE_MOVABLE
];
811 return default_zone_for_pfn(nid
, start_pfn
, nr_pages
);
815 * Associates the given pfn range with the given node and the zone appropriate
816 * for the given online type.
818 static struct zone
* __meminit
move_pfn_range(int online_type
, int nid
,
819 unsigned long start_pfn
, unsigned long nr_pages
)
823 zone
= zone_for_pfn_range(online_type
, nid
, start_pfn
, nr_pages
);
824 move_pfn_range_to_zone(zone
, start_pfn
, nr_pages
, NULL
);
828 int __ref
online_pages(unsigned long pfn
, unsigned long nr_pages
, int online_type
)
831 unsigned long onlined_pages
= 0;
833 int need_zonelists_rebuild
= 0;
836 struct memory_notify arg
;
837 struct memory_block
*mem
;
842 * We can't use pfn_to_nid() because nid might be stored in struct page
843 * which is not yet initialized. Instead, we find nid from memory block.
845 mem
= find_memory_block(__pfn_to_section(pfn
));
847 put_device(&mem
->dev
);
849 /* associate pfn range with the zone */
850 zone
= move_pfn_range(online_type
, nid
, pfn
, nr_pages
);
853 arg
.nr_pages
= nr_pages
;
854 node_states_check_changes_online(nr_pages
, zone
, &arg
);
856 ret
= memory_notify(MEM_GOING_ONLINE
, &arg
);
857 ret
= notifier_to_errno(ret
);
859 goto failed_addition
;
862 * If this zone is not populated, then it is not in zonelist.
863 * This means the page allocator ignores this zone.
864 * So, zonelist must be updated after online.
866 if (!populated_zone(zone
)) {
867 need_zonelists_rebuild
= 1;
868 setup_zone_pageset(zone
);
871 ret
= walk_system_ram_range(pfn
, nr_pages
, &onlined_pages
,
874 if (need_zonelists_rebuild
)
875 zone_pcp_reset(zone
);
876 goto failed_addition
;
879 zone
->present_pages
+= onlined_pages
;
881 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
882 zone
->zone_pgdat
->node_present_pages
+= onlined_pages
;
883 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
888 node_states_set_node(nid
, &arg
);
889 if (need_zonelists_rebuild
)
890 build_all_zonelists(NULL
);
892 zone_pcp_update(zone
);
895 init_per_zone_wmark_min();
902 vm_total_pages
= nr_free_pagecache_pages();
904 writeback_set_ratelimit();
907 memory_notify(MEM_ONLINE
, &arg
);
912 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
913 (unsigned long long) pfn
<< PAGE_SHIFT
,
914 (((unsigned long long) pfn
+ nr_pages
) << PAGE_SHIFT
) - 1);
915 memory_notify(MEM_CANCEL_ONLINE
, &arg
);
919 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
921 static void reset_node_present_pages(pg_data_t
*pgdat
)
925 for (z
= pgdat
->node_zones
; z
< pgdat
->node_zones
+ MAX_NR_ZONES
; z
++)
926 z
->present_pages
= 0;
928 pgdat
->node_present_pages
= 0;
931 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
932 static pg_data_t __ref
*hotadd_new_pgdat(int nid
, u64 start
)
934 struct pglist_data
*pgdat
;
935 unsigned long start_pfn
= PFN_DOWN(start
);
937 pgdat
= NODE_DATA(nid
);
939 pgdat
= arch_alloc_nodedata(nid
);
943 arch_refresh_nodedata(nid
, pgdat
);
946 * Reset the nr_zones, order and classzone_idx before reuse.
947 * Note that kswapd will init kswapd_classzone_idx properly
948 * when it starts in the near future.
951 pgdat
->kswapd_order
= 0;
952 pgdat
->kswapd_classzone_idx
= 0;
955 /* we can use NODE_DATA(nid) from here */
957 pgdat
->node_id
= nid
;
958 pgdat
->node_start_pfn
= start_pfn
;
960 /* init node's zones as empty zones, we don't have any present pages.*/
961 free_area_init_core_hotplug(nid
);
962 pgdat
->per_cpu_nodestats
= alloc_percpu(struct per_cpu_nodestat
);
965 * The node we allocated has no zone fallback lists. For avoiding
966 * to access not-initialized zonelist, build here.
968 build_all_zonelists(pgdat
);
971 * When memory is hot-added, all the memory is in offline state. So
972 * clear all zones' present_pages because they will be updated in
973 * online_pages() and offline_pages().
975 reset_node_managed_pages(pgdat
);
976 reset_node_present_pages(pgdat
);
981 static void rollback_node_hotadd(int nid
)
983 pg_data_t
*pgdat
= NODE_DATA(nid
);
985 arch_refresh_nodedata(nid
, NULL
);
986 free_percpu(pgdat
->per_cpu_nodestats
);
987 arch_free_nodedata(pgdat
);
993 * try_online_node - online a node if offlined
995 * @start: start addr of the node
996 * @set_node_online: Whether we want to online the node
997 * called by cpu_up() to online a node without onlined memory.
1000 * 1 -> a new node has been allocated
1001 * 0 -> the node is already online
1002 * -ENOMEM -> the node could not be allocated
1004 static int __try_online_node(int nid
, u64 start
, bool set_node_online
)
1009 if (node_online(nid
))
1012 pgdat
= hotadd_new_pgdat(nid
, start
);
1014 pr_err("Cannot online node %d due to NULL pgdat\n", nid
);
1019 if (set_node_online
) {
1020 node_set_online(nid
);
1021 ret
= register_one_node(nid
);
1029 * Users of this function always want to online/register the node
1031 int try_online_node(int nid
)
1035 mem_hotplug_begin();
1036 ret
= __try_online_node(nid
, 0, true);
1041 static int check_hotplug_memory_range(u64 start
, u64 size
)
1043 /* memory range must be block size aligned */
1044 if (!size
|| !IS_ALIGNED(start
, memory_block_size_bytes()) ||
1045 !IS_ALIGNED(size
, memory_block_size_bytes())) {
1046 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1047 memory_block_size_bytes(), start
, size
);
1054 static int online_memory_block(struct memory_block
*mem
, void *arg
)
1056 return device_online(&mem
->dev
);
1060 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1061 * and online/offline operations (triggered e.g. by sysfs).
1063 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1065 int __ref
add_memory_resource(int nid
, struct resource
*res
)
1067 struct mhp_restrictions restrictions
= {};
1069 bool new_node
= false;
1073 size
= resource_size(res
);
1075 ret
= check_hotplug_memory_range(start
, size
);
1079 mem_hotplug_begin();
1082 * Add new range to memblock so that when hotadd_new_pgdat() is called
1083 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1084 * this new range and calculate total pages correctly. The range will
1085 * be removed at hot-remove time.
1087 memblock_add_node(start
, size
, nid
);
1089 ret
= __try_online_node(nid
, start
, false);
1094 /* call arch's memory hotadd */
1095 ret
= arch_add_memory(nid
, start
, size
, &restrictions
);
1099 /* create memory block devices after memory was added */
1100 ret
= create_memory_block_devices(start
, size
);
1102 arch_remove_memory(nid
, start
, size
, NULL
);
1107 /* If sysfs file of new node can't be created, cpu on the node
1108 * can't be hot-added. There is no rollback way now.
1109 * So, check by BUG_ON() to catch it reluctantly..
1110 * We online node here. We can't roll back from here.
1112 node_set_online(nid
);
1113 ret
= __register_one_node(nid
);
1117 /* link memory sections under this node.*/
1118 ret
= link_mem_sections(nid
, PFN_DOWN(start
), PFN_UP(start
+ size
- 1));
1121 /* create new memmap entry */
1122 firmware_map_add_hotplug(start
, start
+ size
, "System RAM");
1124 /* device_online() will take the lock when calling online_pages() */
1127 /* online pages if requested */
1128 if (memhp_auto_online
)
1129 walk_memory_range(PFN_DOWN(start
), PFN_UP(start
+ size
- 1),
1130 NULL
, online_memory_block
);
1134 /* rollback pgdat allocation and others */
1136 rollback_node_hotadd(nid
);
1137 memblock_remove(start
, size
);
1142 /* requires device_hotplug_lock, see add_memory_resource() */
1143 int __ref
__add_memory(int nid
, u64 start
, u64 size
)
1145 struct resource
*res
;
1148 res
= register_memory_resource(start
, size
);
1150 return PTR_ERR(res
);
1152 ret
= add_memory_resource(nid
, res
);
1154 release_memory_resource(res
);
1158 int add_memory(int nid
, u64 start
, u64 size
)
1162 lock_device_hotplug();
1163 rc
= __add_memory(nid
, start
, size
);
1164 unlock_device_hotplug();
1168 EXPORT_SYMBOL_GPL(add_memory
);
1170 #ifdef CONFIG_MEMORY_HOTREMOVE
1172 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1173 * set and the size of the free page is given by page_order(). Using this,
1174 * the function determines if the pageblock contains only free pages.
1175 * Due to buddy contraints, a free page at least the size of a pageblock will
1176 * be located at the start of the pageblock
1178 static inline int pageblock_free(struct page
*page
)
1180 return PageBuddy(page
) && page_order(page
) >= pageblock_order
;
1183 /* Return the pfn of the start of the next active pageblock after a given pfn */
1184 static unsigned long next_active_pageblock(unsigned long pfn
)
1186 struct page
*page
= pfn_to_page(pfn
);
1188 /* Ensure the starting page is pageblock-aligned */
1189 BUG_ON(pfn
& (pageblock_nr_pages
- 1));
1191 /* If the entire pageblock is free, move to the end of free page */
1192 if (pageblock_free(page
)) {
1194 /* be careful. we don't have locks, page_order can be changed.*/
1195 order
= page_order(page
);
1196 if ((order
< MAX_ORDER
) && (order
>= pageblock_order
))
1197 return pfn
+ (1 << order
);
1200 return pfn
+ pageblock_nr_pages
;
1203 static bool is_pageblock_removable_nolock(unsigned long pfn
)
1205 struct page
*page
= pfn_to_page(pfn
);
1209 * We have to be careful here because we are iterating over memory
1210 * sections which are not zone aware so we might end up outside of
1211 * the zone but still within the section.
1212 * We have to take care about the node as well. If the node is offline
1213 * its NODE_DATA will be NULL - see page_zone.
1215 if (!node_online(page_to_nid(page
)))
1218 zone
= page_zone(page
);
1219 pfn
= page_to_pfn(page
);
1220 if (!zone_spans_pfn(zone
, pfn
))
1223 return !has_unmovable_pages(zone
, page
, 0, MIGRATE_MOVABLE
, SKIP_HWPOISON
);
1226 /* Checks if this range of memory is likely to be hot-removable. */
1227 bool is_mem_section_removable(unsigned long start_pfn
, unsigned long nr_pages
)
1229 unsigned long end_pfn
, pfn
;
1231 end_pfn
= min(start_pfn
+ nr_pages
,
1232 zone_end_pfn(page_zone(pfn_to_page(start_pfn
))));
1234 /* Check the starting page of each pageblock within the range */
1235 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
= next_active_pageblock(pfn
)) {
1236 if (!is_pageblock_removable_nolock(pfn
))
1241 /* All pageblocks in the memory block are likely to be hot-removable */
1246 * Confirm all pages in a range [start, end) belong to the same zone.
1247 * When true, return its valid [start, end).
1249 int test_pages_in_a_zone(unsigned long start_pfn
, unsigned long end_pfn
,
1250 unsigned long *valid_start
, unsigned long *valid_end
)
1252 unsigned long pfn
, sec_end_pfn
;
1253 unsigned long start
, end
;
1254 struct zone
*zone
= NULL
;
1257 for (pfn
= start_pfn
, sec_end_pfn
= SECTION_ALIGN_UP(start_pfn
+ 1);
1259 pfn
= sec_end_pfn
, sec_end_pfn
+= PAGES_PER_SECTION
) {
1260 /* Make sure the memory section is present first */
1261 if (!present_section_nr(pfn_to_section_nr(pfn
)))
1263 for (; pfn
< sec_end_pfn
&& pfn
< end_pfn
;
1264 pfn
+= MAX_ORDER_NR_PAGES
) {
1266 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1267 while ((i
< MAX_ORDER_NR_PAGES
) &&
1268 !pfn_valid_within(pfn
+ i
))
1270 if (i
== MAX_ORDER_NR_PAGES
|| pfn
+ i
>= end_pfn
)
1272 /* Check if we got outside of the zone */
1273 if (zone
&& !zone_spans_pfn(zone
, pfn
+ i
))
1275 page
= pfn_to_page(pfn
+ i
);
1276 if (zone
&& page_zone(page
) != zone
)
1280 zone
= page_zone(page
);
1281 end
= pfn
+ MAX_ORDER_NR_PAGES
;
1286 *valid_start
= start
;
1287 *valid_end
= min(end
, end_pfn
);
1295 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1296 * non-lru movable pages and hugepages). We scan pfn because it's much
1297 * easier than scanning over linked list. This function returns the pfn
1298 * of the first found movable page if it's found, otherwise 0.
1300 static unsigned long scan_movable_pages(unsigned long start
, unsigned long end
)
1304 for (pfn
= start
; pfn
< end
; pfn
++) {
1305 struct page
*page
, *head
;
1308 if (!pfn_valid(pfn
))
1310 page
= pfn_to_page(pfn
);
1313 if (__PageMovable(page
))
1316 if (!PageHuge(page
))
1318 head
= compound_head(page
);
1319 if (page_huge_active(head
))
1321 skip
= (1 << compound_order(head
)) - (page
- head
);
1327 static struct page
*new_node_page(struct page
*page
, unsigned long private)
1329 int nid
= page_to_nid(page
);
1330 nodemask_t nmask
= node_states
[N_MEMORY
];
1333 * try to allocate from a different node but reuse this node if there
1334 * are no other online nodes to be used (e.g. we are offlining a part
1335 * of the only existing node)
1337 node_clear(nid
, nmask
);
1338 if (nodes_empty(nmask
))
1339 node_set(nid
, nmask
);
1341 return new_page_nodemask(page
, nid
, &nmask
);
1345 do_migrate_range(unsigned long start_pfn
, unsigned long end_pfn
)
1352 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
1353 if (!pfn_valid(pfn
))
1355 page
= pfn_to_page(pfn
);
1357 if (PageHuge(page
)) {
1358 struct page
*head
= compound_head(page
);
1359 pfn
= page_to_pfn(head
) + (1<<compound_order(head
)) - 1;
1360 isolate_huge_page(head
, &source
);
1362 } else if (PageTransHuge(page
))
1363 pfn
= page_to_pfn(compound_head(page
))
1364 + hpage_nr_pages(page
) - 1;
1367 * HWPoison pages have elevated reference counts so the migration would
1368 * fail on them. It also doesn't make any sense to migrate them in the
1369 * first place. Still try to unmap such a page in case it is still mapped
1370 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1371 * the unmap as the catch all safety net).
1373 if (PageHWPoison(page
)) {
1374 if (WARN_ON(PageLRU(page
)))
1375 isolate_lru_page(page
);
1376 if (page_mapped(page
))
1377 try_to_unmap(page
, TTU_IGNORE_MLOCK
| TTU_IGNORE_ACCESS
);
1381 if (!get_page_unless_zero(page
))
1384 * We can skip free pages. And we can deal with pages on
1385 * LRU and non-lru movable pages.
1388 ret
= isolate_lru_page(page
);
1390 ret
= isolate_movable_page(page
, ISOLATE_UNEVICTABLE
);
1391 if (!ret
) { /* Success */
1392 list_add_tail(&page
->lru
, &source
);
1393 if (!__PageMovable(page
))
1394 inc_node_page_state(page
, NR_ISOLATED_ANON
+
1395 page_is_file_cache(page
));
1398 pr_warn("failed to isolate pfn %lx\n", pfn
);
1399 dump_page(page
, "isolation failed");
1403 if (!list_empty(&source
)) {
1404 /* Allocate a new page from the nearest neighbor node */
1405 ret
= migrate_pages(&source
, new_node_page
, NULL
, 0,
1406 MIGRATE_SYNC
, MR_MEMORY_HOTPLUG
);
1408 list_for_each_entry(page
, &source
, lru
) {
1409 pr_warn("migrating pfn %lx failed ret:%d ",
1410 page_to_pfn(page
), ret
);
1411 dump_page(page
, "migration failure");
1413 putback_movable_pages(&source
);
1421 * remove from free_area[] and mark all as Reserved.
1424 offline_isolated_pages_cb(unsigned long start
, unsigned long nr_pages
,
1427 unsigned long *offlined_pages
= (unsigned long *)data
;
1429 *offlined_pages
+= __offline_isolated_pages(start
, start
+ nr_pages
);
1434 * Check all pages in range, recoreded as memory resource, are isolated.
1437 check_pages_isolated_cb(unsigned long start_pfn
, unsigned long nr_pages
,
1440 return test_pages_isolated(start_pfn
, start_pfn
+ nr_pages
, true);
1443 static int __init
cmdline_parse_movable_node(char *p
)
1445 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1446 movable_node_enabled
= true;
1448 pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1452 early_param("movable_node", cmdline_parse_movable_node
);
1454 /* check which state of node_states will be changed when offline memory */
1455 static void node_states_check_changes_offline(unsigned long nr_pages
,
1456 struct zone
*zone
, struct memory_notify
*arg
)
1458 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1459 unsigned long present_pages
= 0;
1462 arg
->status_change_nid
= NUMA_NO_NODE
;
1463 arg
->status_change_nid_normal
= NUMA_NO_NODE
;
1464 arg
->status_change_nid_high
= NUMA_NO_NODE
;
1467 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1468 * If the memory to be offline is within the range
1469 * [0..ZONE_NORMAL], and it is the last present memory there,
1470 * the zones in that range will become empty after the offlining,
1471 * thus we can determine that we need to clear the node from
1472 * node_states[N_NORMAL_MEMORY].
1474 for (zt
= 0; zt
<= ZONE_NORMAL
; zt
++)
1475 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1476 if (zone_idx(zone
) <= ZONE_NORMAL
&& nr_pages
>= present_pages
)
1477 arg
->status_change_nid_normal
= zone_to_nid(zone
);
1479 #ifdef CONFIG_HIGHMEM
1481 * node_states[N_HIGH_MEMORY] contains nodes which
1482 * have normal memory or high memory.
1483 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1484 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1485 * we determine that the zones in that range become empty,
1486 * we need to clear the node for N_HIGH_MEMORY.
1488 present_pages
+= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1489 if (zone_idx(zone
) <= ZONE_HIGHMEM
&& nr_pages
>= present_pages
)
1490 arg
->status_change_nid_high
= zone_to_nid(zone
);
1494 * We have accounted the pages from [0..ZONE_NORMAL), and
1495 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1497 * Here we count the possible pages from ZONE_MOVABLE.
1498 * If after having accounted all the pages, we see that the nr_pages
1499 * to be offlined is over or equal to the accounted pages,
1500 * we know that the node will become empty, and so, we can clear
1501 * it for N_MEMORY as well.
1503 present_pages
+= pgdat
->node_zones
[ZONE_MOVABLE
].present_pages
;
1505 if (nr_pages
>= present_pages
)
1506 arg
->status_change_nid
= zone_to_nid(zone
);
1509 static void node_states_clear_node(int node
, struct memory_notify
*arg
)
1511 if (arg
->status_change_nid_normal
>= 0)
1512 node_clear_state(node
, N_NORMAL_MEMORY
);
1514 if (arg
->status_change_nid_high
>= 0)
1515 node_clear_state(node
, N_HIGH_MEMORY
);
1517 if (arg
->status_change_nid
>= 0)
1518 node_clear_state(node
, N_MEMORY
);
1521 static int __ref
__offline_pages(unsigned long start_pfn
,
1522 unsigned long end_pfn
)
1524 unsigned long pfn
, nr_pages
;
1525 unsigned long offlined_pages
= 0;
1526 int ret
, node
, nr_isolate_pageblock
;
1527 unsigned long flags
;
1528 unsigned long valid_start
, valid_end
;
1530 struct memory_notify arg
;
1533 mem_hotplug_begin();
1535 /* This makes hotplug much easier...and readable.
1536 we assume this for now. .*/
1537 if (!test_pages_in_a_zone(start_pfn
, end_pfn
, &valid_start
,
1540 reason
= "multizone range";
1541 goto failed_removal
;
1544 zone
= page_zone(pfn_to_page(valid_start
));
1545 node
= zone_to_nid(zone
);
1546 nr_pages
= end_pfn
- start_pfn
;
1548 /* set above range as isolated */
1549 ret
= start_isolate_page_range(start_pfn
, end_pfn
,
1551 SKIP_HWPOISON
| REPORT_FAILURE
);
1553 reason
= "failure to isolate range";
1554 goto failed_removal
;
1556 nr_isolate_pageblock
= ret
;
1558 arg
.start_pfn
= start_pfn
;
1559 arg
.nr_pages
= nr_pages
;
1560 node_states_check_changes_offline(nr_pages
, zone
, &arg
);
1562 ret
= memory_notify(MEM_GOING_OFFLINE
, &arg
);
1563 ret
= notifier_to_errno(ret
);
1565 reason
= "notifier failure";
1566 goto failed_removal_isolated
;
1570 for (pfn
= start_pfn
; pfn
;) {
1571 if (signal_pending(current
)) {
1573 reason
= "signal backoff";
1574 goto failed_removal_isolated
;
1578 lru_add_drain_all();
1580 pfn
= scan_movable_pages(pfn
, end_pfn
);
1583 * TODO: fatal migration failures should bail
1586 do_migrate_range(pfn
, end_pfn
);
1591 * Dissolve free hugepages in the memory block before doing
1592 * offlining actually in order to make hugetlbfs's object
1593 * counting consistent.
1595 ret
= dissolve_free_huge_pages(start_pfn
, end_pfn
);
1597 reason
= "failure to dissolve huge pages";
1598 goto failed_removal_isolated
;
1601 ret
= walk_system_ram_range(start_pfn
, end_pfn
- start_pfn
,
1602 NULL
, check_pages_isolated_cb
);
1605 /* Ok, all of our target is isolated.
1606 We cannot do rollback at this point. */
1607 walk_system_ram_range(start_pfn
, end_pfn
- start_pfn
,
1608 &offlined_pages
, offline_isolated_pages_cb
);
1609 pr_info("Offlined Pages %ld\n", offlined_pages
);
1611 * Onlining will reset pagetype flags and makes migrate type
1612 * MOVABLE, so just need to decrease the number of isolated
1613 * pageblocks zone counter here.
1615 spin_lock_irqsave(&zone
->lock
, flags
);
1616 zone
->nr_isolate_pageblock
-= nr_isolate_pageblock
;
1617 spin_unlock_irqrestore(&zone
->lock
, flags
);
1619 /* removal success */
1620 adjust_managed_page_count(pfn_to_page(start_pfn
), -offlined_pages
);
1621 zone
->present_pages
-= offlined_pages
;
1623 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
1624 zone
->zone_pgdat
->node_present_pages
-= offlined_pages
;
1625 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
1627 init_per_zone_wmark_min();
1629 if (!populated_zone(zone
)) {
1630 zone_pcp_reset(zone
);
1631 build_all_zonelists(NULL
);
1633 zone_pcp_update(zone
);
1635 node_states_clear_node(node
, &arg
);
1636 if (arg
.status_change_nid
>= 0) {
1638 kcompactd_stop(node
);
1641 vm_total_pages
= nr_free_pagecache_pages();
1642 writeback_set_ratelimit();
1644 memory_notify(MEM_OFFLINE
, &arg
);
1648 failed_removal_isolated
:
1649 undo_isolate_page_range(start_pfn
, end_pfn
, MIGRATE_MOVABLE
);
1650 memory_notify(MEM_CANCEL_OFFLINE
, &arg
);
1652 pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1653 (unsigned long long) start_pfn
<< PAGE_SHIFT
,
1654 ((unsigned long long) end_pfn
<< PAGE_SHIFT
) - 1,
1656 /* pushback to free area */
1661 int offline_pages(unsigned long start_pfn
, unsigned long nr_pages
)
1663 return __offline_pages(start_pfn
, start_pfn
+ nr_pages
);
1665 #endif /* CONFIG_MEMORY_HOTREMOVE */
1668 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1669 * @start_pfn: start pfn of the memory range
1670 * @end_pfn: end pfn of the memory range
1671 * @arg: argument passed to func
1672 * @func: callback for each memory section walked
1674 * This function walks through all present mem sections in range
1675 * [start_pfn, end_pfn) and call func on each mem section.
1677 * Returns the return value of func.
1679 int walk_memory_range(unsigned long start_pfn
, unsigned long end_pfn
,
1680 void *arg
, int (*func
)(struct memory_block
*, void *))
1682 struct memory_block
*mem
= NULL
;
1683 struct mem_section
*section
;
1684 unsigned long pfn
, section_nr
;
1687 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
1688 section_nr
= pfn_to_section_nr(pfn
);
1689 if (!present_section_nr(section_nr
))
1692 section
= __nr_to_section(section_nr
);
1693 /* same memblock? */
1695 if ((section_nr
>= mem
->start_section_nr
) &&
1696 (section_nr
<= mem
->end_section_nr
))
1699 mem
= find_memory_block_hinted(section
, mem
);
1703 ret
= func(mem
, arg
);
1705 kobject_put(&mem
->dev
.kobj
);
1711 kobject_put(&mem
->dev
.kobj
);
1716 #ifdef CONFIG_MEMORY_HOTREMOVE
1717 static int check_memblock_offlined_cb(struct memory_block
*mem
, void *arg
)
1719 int ret
= !is_memblock_offlined(mem
);
1721 if (unlikely(ret
)) {
1722 phys_addr_t beginpa
, endpa
;
1724 beginpa
= PFN_PHYS(section_nr_to_pfn(mem
->start_section_nr
));
1725 endpa
= PFN_PHYS(section_nr_to_pfn(mem
->end_section_nr
+ 1))-1;
1726 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1734 static int check_cpu_on_node(pg_data_t
*pgdat
)
1738 for_each_present_cpu(cpu
) {
1739 if (cpu_to_node(cpu
) == pgdat
->node_id
)
1741 * the cpu on this node isn't removed, and we can't
1742 * offline this node.
1754 * Offline a node if all memory sections and cpus of the node are removed.
1756 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1757 * and online/offline operations before this call.
1759 void try_offline_node(int nid
)
1761 pg_data_t
*pgdat
= NODE_DATA(nid
);
1762 unsigned long start_pfn
= pgdat
->node_start_pfn
;
1763 unsigned long end_pfn
= start_pfn
+ pgdat
->node_spanned_pages
;
1766 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
1767 unsigned long section_nr
= pfn_to_section_nr(pfn
);
1769 if (!present_section_nr(section_nr
))
1772 if (pfn_to_nid(pfn
) != nid
)
1776 * some memory sections of this node are not removed, and we
1777 * can't offline node now.
1782 if (check_cpu_on_node(pgdat
))
1786 * all memory/cpu of this node are removed, we can offline this
1789 node_set_offline(nid
);
1790 unregister_one_node(nid
);
1792 EXPORT_SYMBOL(try_offline_node
);
1794 static void __release_memory_resource(resource_size_t start
,
1795 resource_size_t size
)
1800 * When removing memory in the same granularity as it was added,
1801 * this function never fails. It might only fail if resources
1802 * have to be adjusted or split. We'll ignore the error, as
1803 * removing of memory cannot fail.
1805 ret
= release_mem_region_adjustable(&iomem_resource
, start
, size
);
1807 resource_size_t endres
= start
+ size
- 1;
1809 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
1810 &start
, &endres
, ret
);
1814 static int __ref
try_remove_memory(int nid
, u64 start
, u64 size
)
1818 BUG_ON(check_hotplug_memory_range(start
, size
));
1820 mem_hotplug_begin();
1823 * All memory blocks must be offlined before removing memory. Check
1824 * whether all memory blocks in question are offline and return error
1825 * if this is not the case.
1827 rc
= walk_memory_range(PFN_DOWN(start
), PFN_UP(start
+ size
- 1), NULL
,
1828 check_memblock_offlined_cb
);
1832 /* remove memmap entry */
1833 firmware_map_remove(start
, start
+ size
, "System RAM");
1834 memblock_free(start
, size
);
1835 memblock_remove(start
, size
);
1837 arch_remove_memory(nid
, start
, size
, NULL
);
1838 __release_memory_resource(start
, size
);
1840 try_offline_node(nid
);
1850 * @start: physical address of the region to remove
1851 * @size: size of the region to remove
1853 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1854 * and online/offline operations before this call, as required by
1855 * try_offline_node().
1857 void __remove_memory(int nid
, u64 start
, u64 size
)
1861 * trigger BUG() is some memory is not offlined prior to calling this
1864 if (try_remove_memory(nid
, start
, size
))
1869 * Remove memory if every memory block is offline, otherwise return -EBUSY is
1870 * some memory is not offline
1872 int remove_memory(int nid
, u64 start
, u64 size
)
1876 lock_device_hotplug();
1877 rc
= try_remove_memory(nid
, start
, size
);
1878 unlock_device_hotplug();
1882 EXPORT_SYMBOL_GPL(remove_memory
);
1883 #endif /* CONFIG_MEMORY_HOTREMOVE */