2 * linux/mm/memory_hotplug.c
7 #include <linux/stddef.h>
9 #include <linux/sched/signal.h>
10 #include <linux/swap.h>
11 #include <linux/interrupt.h>
12 #include <linux/pagemap.h>
13 #include <linux/compiler.h>
14 #include <linux/export.h>
15 #include <linux/pagevec.h>
16 #include <linux/writeback.h>
17 #include <linux/slab.h>
18 #include <linux/sysctl.h>
19 #include <linux/cpu.h>
20 #include <linux/memory.h>
21 #include <linux/memremap.h>
22 #include <linux/memory_hotplug.h>
23 #include <linux/highmem.h>
24 #include <linux/vmalloc.h>
25 #include <linux/ioport.h>
26 #include <linux/delay.h>
27 #include <linux/migrate.h>
28 #include <linux/page-isolation.h>
29 #include <linux/pfn.h>
30 #include <linux/suspend.h>
31 #include <linux/mm_inline.h>
32 #include <linux/firmware-map.h>
33 #include <linux/stop_machine.h>
34 #include <linux/hugetlb.h>
35 #include <linux/memblock.h>
36 #include <linux/compaction.h>
37 #include <linux/rmap.h>
39 #include <asm/tlbflush.h>
44 * online_page_callback contains pointer to current page onlining function.
45 * Initially it is generic_online_page(). If it is required it could be
46 * changed by calling set_online_page_callback() for callback registration
47 * and restore_online_page_callback() for generic callback restore.
50 static void generic_online_page(struct page
*page
, unsigned int order
);
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 bool memhp_auto_online
;
72 bool memhp_auto_online
= true;
74 EXPORT_SYMBOL_GPL(memhp_auto_online
);
76 static int __init
setup_memhp_default_state(char *str
)
78 if (!strcmp(str
, "online"))
79 memhp_auto_online
= true;
80 else if (!strcmp(str
, "offline"))
81 memhp_auto_online
= false;
85 __setup("memhp_default_state=", setup_memhp_default_state
);
87 void mem_hotplug_begin(void)
90 percpu_down_write(&mem_hotplug_lock
);
93 void mem_hotplug_done(void)
95 percpu_up_write(&mem_hotplug_lock
);
99 u64 max_mem_size
= U64_MAX
;
101 /* add this memory to iomem resource */
102 static struct resource
*register_memory_resource(u64 start
, u64 size
)
104 struct resource
*res
;
105 unsigned long flags
= IORESOURCE_SYSTEM_RAM
| IORESOURCE_BUSY
;
106 char *resource_name
= "System RAM";
108 if (start
+ size
> max_mem_size
)
109 return ERR_PTR(-E2BIG
);
112 * Request ownership of the new memory range. This might be
113 * a child of an existing resource that was present but
114 * not marked as busy.
116 res
= __request_region(&iomem_resource
, start
, size
,
117 resource_name
, flags
);
120 pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n",
121 start
, start
+ size
);
122 return ERR_PTR(-EEXIST
);
127 static void release_memory_resource(struct resource
*res
)
131 release_resource(res
);
136 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
137 void get_page_bootmem(unsigned long info
, struct page
*page
,
140 page
->freelist
= (void *)type
;
141 SetPagePrivate(page
);
142 set_page_private(page
, info
);
146 void put_page_bootmem(struct page
*page
)
150 type
= (unsigned long) page
->freelist
;
151 BUG_ON(type
< MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE
||
152 type
> MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE
);
154 if (page_ref_dec_return(page
) == 1) {
155 page
->freelist
= NULL
;
156 ClearPagePrivate(page
);
157 set_page_private(page
, 0);
158 INIT_LIST_HEAD(&page
->lru
);
159 free_reserved_page(page
);
163 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
164 #ifndef CONFIG_SPARSEMEM_VMEMMAP
165 static void register_page_bootmem_info_section(unsigned long start_pfn
)
167 unsigned long *usemap
, mapsize
, section_nr
, i
;
168 struct mem_section
*ms
;
169 struct page
*page
, *memmap
;
171 section_nr
= pfn_to_section_nr(start_pfn
);
172 ms
= __nr_to_section(section_nr
);
174 /* Get section's memmap address */
175 memmap
= sparse_decode_mem_map(ms
->section_mem_map
, section_nr
);
178 * Get page for the memmap's phys address
179 * XXX: need more consideration for sparse_vmemmap...
181 page
= virt_to_page(memmap
);
182 mapsize
= sizeof(struct page
) * PAGES_PER_SECTION
;
183 mapsize
= PAGE_ALIGN(mapsize
) >> PAGE_SHIFT
;
185 /* remember memmap's page */
186 for (i
= 0; i
< mapsize
; i
++, page
++)
187 get_page_bootmem(section_nr
, page
, SECTION_INFO
);
189 usemap
= ms
->pageblock_flags
;
190 page
= virt_to_page(usemap
);
192 mapsize
= PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT
;
194 for (i
= 0; i
< mapsize
; i
++, page
++)
195 get_page_bootmem(section_nr
, page
, MIX_SECTION_INFO
);
198 #else /* CONFIG_SPARSEMEM_VMEMMAP */
199 static void register_page_bootmem_info_section(unsigned long start_pfn
)
201 unsigned long *usemap
, mapsize
, section_nr
, i
;
202 struct mem_section
*ms
;
203 struct page
*page
, *memmap
;
205 section_nr
= pfn_to_section_nr(start_pfn
);
206 ms
= __nr_to_section(section_nr
);
208 memmap
= sparse_decode_mem_map(ms
->section_mem_map
, section_nr
);
210 register_page_bootmem_memmap(section_nr
, memmap
, PAGES_PER_SECTION
);
212 usemap
= ms
->pageblock_flags
;
213 page
= virt_to_page(usemap
);
215 mapsize
= PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT
;
217 for (i
= 0; i
< mapsize
; i
++, page
++)
218 get_page_bootmem(section_nr
, page
, MIX_SECTION_INFO
);
220 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
222 void __init
register_page_bootmem_info_node(struct pglist_data
*pgdat
)
224 unsigned long i
, pfn
, end_pfn
, nr_pages
;
225 int node
= pgdat
->node_id
;
228 nr_pages
= PAGE_ALIGN(sizeof(struct pglist_data
)) >> PAGE_SHIFT
;
229 page
= virt_to_page(pgdat
);
231 for (i
= 0; i
< nr_pages
; i
++, page
++)
232 get_page_bootmem(node
, page
, NODE_INFO
);
234 pfn
= pgdat
->node_start_pfn
;
235 end_pfn
= pgdat_end_pfn(pgdat
);
237 /* register section info */
238 for (; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
240 * Some platforms can assign the same pfn to multiple nodes - on
241 * node0 as well as nodeN. To avoid registering a pfn against
242 * multiple nodes we check that this pfn does not already
243 * reside in some other nodes.
245 if (pfn_valid(pfn
) && (early_pfn_to_nid(pfn
) == node
))
246 register_page_bootmem_info_section(pfn
);
249 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
251 static int __meminit
__add_section(int nid
, unsigned long phys_start_pfn
,
252 struct vmem_altmap
*altmap
, bool want_memblock
)
256 if (pfn_valid(phys_start_pfn
))
259 ret
= sparse_add_one_section(nid
, phys_start_pfn
, altmap
);
266 return hotplug_memory_register(nid
, __pfn_to_section(phys_start_pfn
));
270 * Reasonably generic function for adding memory. It is
271 * expected that archs that support memory hotplug will
272 * call this function after deciding the zone to which to
275 int __ref
__add_pages(int nid
, unsigned long phys_start_pfn
,
276 unsigned long nr_pages
, struct vmem_altmap
*altmap
,
281 int start_sec
, end_sec
;
283 /* during initialize mem_map, align hot-added range to section */
284 start_sec
= pfn_to_section_nr(phys_start_pfn
);
285 end_sec
= pfn_to_section_nr(phys_start_pfn
+ nr_pages
- 1);
289 * Validate altmap is within bounds of the total request
291 if (altmap
->base_pfn
!= phys_start_pfn
292 || vmem_altmap_offset(altmap
) > nr_pages
) {
293 pr_warn_once("memory add fail, invalid altmap\n");
300 for (i
= start_sec
; i
<= end_sec
; i
++) {
301 err
= __add_section(nid
, section_nr_to_pfn(i
), altmap
,
305 * EEXIST is finally dealt with by ioresource collision
306 * check. see add_memory() => register_memory_resource()
307 * Warning will be printed if there is collision.
309 if (err
&& (err
!= -EEXIST
))
314 vmemmap_populate_print_last();
319 #ifdef CONFIG_MEMORY_HOTREMOVE
320 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
321 static unsigned long find_smallest_section_pfn(int nid
, struct zone
*zone
,
322 unsigned long start_pfn
,
323 unsigned long end_pfn
)
325 struct mem_section
*ms
;
327 for (; start_pfn
< end_pfn
; start_pfn
+= PAGES_PER_SECTION
) {
328 ms
= __pfn_to_section(start_pfn
);
330 if (unlikely(!valid_section(ms
)))
333 if (unlikely(pfn_to_nid(start_pfn
) != nid
))
336 if (zone
&& zone
!= page_zone(pfn_to_page(start_pfn
)))
345 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
346 static unsigned long find_biggest_section_pfn(int nid
, struct zone
*zone
,
347 unsigned long start_pfn
,
348 unsigned long end_pfn
)
350 struct mem_section
*ms
;
353 /* pfn is the end pfn of a memory section. */
355 for (; pfn
>= start_pfn
; pfn
-= PAGES_PER_SECTION
) {
356 ms
= __pfn_to_section(pfn
);
358 if (unlikely(!valid_section(ms
)))
361 if (unlikely(pfn_to_nid(pfn
) != nid
))
364 if (zone
&& zone
!= page_zone(pfn_to_page(pfn
)))
373 static void shrink_zone_span(struct zone
*zone
, unsigned long start_pfn
,
374 unsigned long end_pfn
)
376 unsigned long zone_start_pfn
= zone
->zone_start_pfn
;
377 unsigned long z
= zone_end_pfn(zone
); /* zone_end_pfn namespace clash */
378 unsigned long zone_end_pfn
= z
;
380 struct mem_section
*ms
;
381 int nid
= zone_to_nid(zone
);
383 zone_span_writelock(zone
);
384 if (zone_start_pfn
== start_pfn
) {
386 * If the section is smallest section in the zone, it need
387 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
388 * In this case, we find second smallest valid mem_section
389 * for shrinking zone.
391 pfn
= find_smallest_section_pfn(nid
, zone
, end_pfn
,
394 zone
->zone_start_pfn
= pfn
;
395 zone
->spanned_pages
= zone_end_pfn
- pfn
;
397 } else if (zone_end_pfn
== end_pfn
) {
399 * If the section is biggest section in the zone, it need
400 * shrink zone->spanned_pages.
401 * In this case, we find second biggest valid mem_section for
404 pfn
= find_biggest_section_pfn(nid
, zone
, zone_start_pfn
,
407 zone
->spanned_pages
= pfn
- zone_start_pfn
+ 1;
411 * The section is not biggest or smallest mem_section in the zone, it
412 * only creates a hole in the zone. So in this case, we need not
413 * change the zone. But perhaps, the zone has only hole data. Thus
414 * it check the zone has only hole or not.
416 pfn
= zone_start_pfn
;
417 for (; pfn
< zone_end_pfn
; pfn
+= PAGES_PER_SECTION
) {
418 ms
= __pfn_to_section(pfn
);
420 if (unlikely(!valid_section(ms
)))
423 if (page_zone(pfn_to_page(pfn
)) != zone
)
426 /* If the section is current section, it continues the loop */
427 if (start_pfn
== pfn
)
430 /* If we find valid section, we have nothing to do */
431 zone_span_writeunlock(zone
);
435 /* The zone has no valid section */
436 zone
->zone_start_pfn
= 0;
437 zone
->spanned_pages
= 0;
438 zone_span_writeunlock(zone
);
441 static void shrink_pgdat_span(struct pglist_data
*pgdat
,
442 unsigned long start_pfn
, unsigned long end_pfn
)
444 unsigned long pgdat_start_pfn
= pgdat
->node_start_pfn
;
445 unsigned long p
= pgdat_end_pfn(pgdat
); /* pgdat_end_pfn namespace clash */
446 unsigned long pgdat_end_pfn
= p
;
448 struct mem_section
*ms
;
449 int nid
= pgdat
->node_id
;
451 if (pgdat_start_pfn
== start_pfn
) {
453 * If the section is smallest section in the pgdat, it need
454 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
455 * In this case, we find second smallest valid mem_section
456 * for shrinking zone.
458 pfn
= find_smallest_section_pfn(nid
, NULL
, end_pfn
,
461 pgdat
->node_start_pfn
= pfn
;
462 pgdat
->node_spanned_pages
= pgdat_end_pfn
- pfn
;
464 } else if (pgdat_end_pfn
== end_pfn
) {
466 * If the section is biggest section in the pgdat, it need
467 * shrink pgdat->node_spanned_pages.
468 * In this case, we find second biggest valid mem_section for
471 pfn
= find_biggest_section_pfn(nid
, NULL
, pgdat_start_pfn
,
474 pgdat
->node_spanned_pages
= pfn
- pgdat_start_pfn
+ 1;
478 * If the section is not biggest or smallest mem_section in the pgdat,
479 * it only creates a hole in the pgdat. So in this case, we need not
481 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
482 * has only hole or not.
484 pfn
= pgdat_start_pfn
;
485 for (; pfn
< pgdat_end_pfn
; pfn
+= PAGES_PER_SECTION
) {
486 ms
= __pfn_to_section(pfn
);
488 if (unlikely(!valid_section(ms
)))
491 if (pfn_to_nid(pfn
) != nid
)
494 /* If the section is current section, it continues the loop */
495 if (start_pfn
== pfn
)
498 /* If we find valid section, we have nothing to do */
502 /* The pgdat has no valid section */
503 pgdat
->node_start_pfn
= 0;
504 pgdat
->node_spanned_pages
= 0;
507 static void __remove_zone(struct zone
*zone
, unsigned long start_pfn
)
509 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
510 int nr_pages
= PAGES_PER_SECTION
;
513 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
514 shrink_zone_span(zone
, start_pfn
, start_pfn
+ nr_pages
);
515 shrink_pgdat_span(pgdat
, start_pfn
, start_pfn
+ nr_pages
);
516 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
519 static int __remove_section(struct zone
*zone
, struct mem_section
*ms
,
520 unsigned long map_offset
, struct vmem_altmap
*altmap
)
522 unsigned long start_pfn
;
526 if (!valid_section(ms
))
529 ret
= unregister_memory_section(ms
);
533 scn_nr
= __section_nr(ms
);
534 start_pfn
= section_nr_to_pfn((unsigned long)scn_nr
);
535 __remove_zone(zone
, start_pfn
);
537 sparse_remove_one_section(zone
, ms
, map_offset
, altmap
);
542 * __remove_pages() - remove sections of pages from a zone
543 * @zone: zone from which pages need to be removed
544 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
545 * @nr_pages: number of pages to remove (must be multiple of section size)
546 * @altmap: alternative device page map or %NULL if default memmap is used
548 * Generic helper function to remove section mappings and sysfs entries
549 * for the section of the memory we are removing. Caller needs to make
550 * sure that pages are marked reserved and zones are adjust properly by
551 * calling offline_pages().
553 int __remove_pages(struct zone
*zone
, unsigned long phys_start_pfn
,
554 unsigned long nr_pages
, struct vmem_altmap
*altmap
)
557 unsigned long map_offset
= 0;
558 int sections_to_remove
, ret
= 0;
560 /* In the ZONE_DEVICE case device driver owns the memory region */
561 if (is_dev_zone(zone
)) {
563 map_offset
= vmem_altmap_offset(altmap
);
565 resource_size_t start
, size
;
567 start
= phys_start_pfn
<< PAGE_SHIFT
;
568 size
= nr_pages
* PAGE_SIZE
;
570 ret
= release_mem_region_adjustable(&iomem_resource
, start
,
573 resource_size_t endres
= start
+ size
- 1;
575 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
576 &start
, &endres
, ret
);
580 clear_zone_contiguous(zone
);
583 * We can only remove entire sections
585 BUG_ON(phys_start_pfn
& ~PAGE_SECTION_MASK
);
586 BUG_ON(nr_pages
% PAGES_PER_SECTION
);
588 sections_to_remove
= nr_pages
/ PAGES_PER_SECTION
;
589 for (i
= 0; i
< sections_to_remove
; i
++) {
590 unsigned long pfn
= phys_start_pfn
+ i
*PAGES_PER_SECTION
;
593 ret
= __remove_section(zone
, __pfn_to_section(pfn
), map_offset
,
600 set_zone_contiguous(zone
);
604 #endif /* CONFIG_MEMORY_HOTREMOVE */
606 int set_online_page_callback(online_page_callback_t callback
)
611 mutex_lock(&online_page_callback_lock
);
613 if (online_page_callback
== generic_online_page
) {
614 online_page_callback
= callback
;
618 mutex_unlock(&online_page_callback_lock
);
623 EXPORT_SYMBOL_GPL(set_online_page_callback
);
625 int restore_online_page_callback(online_page_callback_t callback
)
630 mutex_lock(&online_page_callback_lock
);
632 if (online_page_callback
== callback
) {
633 online_page_callback
= generic_online_page
;
637 mutex_unlock(&online_page_callback_lock
);
642 EXPORT_SYMBOL_GPL(restore_online_page_callback
);
644 void __online_page_set_limits(struct page
*page
)
647 EXPORT_SYMBOL_GPL(__online_page_set_limits
);
649 void __online_page_increment_counters(struct page
*page
)
651 adjust_managed_page_count(page
, 1);
653 EXPORT_SYMBOL_GPL(__online_page_increment_counters
);
655 void __online_page_free(struct page
*page
)
657 __free_reserved_page(page
);
659 EXPORT_SYMBOL_GPL(__online_page_free
);
661 static void generic_online_page(struct page
*page
, unsigned int order
)
663 kernel_map_pages(page
, 1 << order
, 1);
664 __free_pages_core(page
, order
);
665 totalram_pages_add(1UL << order
);
666 #ifdef CONFIG_HIGHMEM
667 if (PageHighMem(page
))
668 totalhigh_pages_add(1UL << order
);
672 static int online_pages_blocks(unsigned long start
, unsigned long nr_pages
)
674 unsigned long end
= start
+ nr_pages
;
675 int order
, onlined_pages
= 0;
677 while (start
< end
) {
678 order
= min(MAX_ORDER
- 1,
679 get_order(PFN_PHYS(end
) - PFN_PHYS(start
)));
680 (*online_page_callback
)(pfn_to_page(start
), order
);
682 onlined_pages
+= (1UL << order
);
683 start
+= (1UL << order
);
685 return onlined_pages
;
688 static int online_pages_range(unsigned long start_pfn
, unsigned long nr_pages
,
691 unsigned long onlined_pages
= *(unsigned long *)arg
;
693 if (PageReserved(pfn_to_page(start_pfn
)))
694 onlined_pages
+= online_pages_blocks(start_pfn
, nr_pages
);
696 online_mem_sections(start_pfn
, start_pfn
+ nr_pages
);
698 *(unsigned long *)arg
= onlined_pages
;
702 /* check which state of node_states will be changed when online memory */
703 static void node_states_check_changes_online(unsigned long nr_pages
,
704 struct zone
*zone
, struct memory_notify
*arg
)
706 int nid
= zone_to_nid(zone
);
708 arg
->status_change_nid
= NUMA_NO_NODE
;
709 arg
->status_change_nid_normal
= NUMA_NO_NODE
;
710 arg
->status_change_nid_high
= NUMA_NO_NODE
;
712 if (!node_state(nid
, N_MEMORY
))
713 arg
->status_change_nid
= nid
;
714 if (zone_idx(zone
) <= ZONE_NORMAL
&& !node_state(nid
, N_NORMAL_MEMORY
))
715 arg
->status_change_nid_normal
= nid
;
716 #ifdef CONFIG_HIGHMEM
717 if (zone_idx(zone
) <= N_HIGH_MEMORY
&& !node_state(nid
, N_HIGH_MEMORY
))
718 arg
->status_change_nid_high
= nid
;
722 static void node_states_set_node(int node
, struct memory_notify
*arg
)
724 if (arg
->status_change_nid_normal
>= 0)
725 node_set_state(node
, N_NORMAL_MEMORY
);
727 if (arg
->status_change_nid_high
>= 0)
728 node_set_state(node
, N_HIGH_MEMORY
);
730 if (arg
->status_change_nid
>= 0)
731 node_set_state(node
, N_MEMORY
);
734 static void __meminit
resize_zone_range(struct zone
*zone
, unsigned long start_pfn
,
735 unsigned long nr_pages
)
737 unsigned long old_end_pfn
= zone_end_pfn(zone
);
739 if (zone_is_empty(zone
) || start_pfn
< zone
->zone_start_pfn
)
740 zone
->zone_start_pfn
= start_pfn
;
742 zone
->spanned_pages
= max(start_pfn
+ nr_pages
, old_end_pfn
) - zone
->zone_start_pfn
;
745 static void __meminit
resize_pgdat_range(struct pglist_data
*pgdat
, unsigned long start_pfn
,
746 unsigned long nr_pages
)
748 unsigned long old_end_pfn
= pgdat_end_pfn(pgdat
);
750 if (!pgdat
->node_spanned_pages
|| start_pfn
< pgdat
->node_start_pfn
)
751 pgdat
->node_start_pfn
= start_pfn
;
753 pgdat
->node_spanned_pages
= max(start_pfn
+ nr_pages
, old_end_pfn
) - pgdat
->node_start_pfn
;
756 void __ref
move_pfn_range_to_zone(struct zone
*zone
, unsigned long start_pfn
,
757 unsigned long nr_pages
, struct vmem_altmap
*altmap
)
759 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
760 int nid
= pgdat
->node_id
;
763 clear_zone_contiguous(zone
);
765 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
766 pgdat_resize_lock(pgdat
, &flags
);
767 zone_span_writelock(zone
);
768 if (zone_is_empty(zone
))
769 init_currently_empty_zone(zone
, start_pfn
, nr_pages
);
770 resize_zone_range(zone
, start_pfn
, nr_pages
);
771 zone_span_writeunlock(zone
);
772 resize_pgdat_range(pgdat
, start_pfn
, nr_pages
);
773 pgdat_resize_unlock(pgdat
, &flags
);
776 * TODO now we have a visible range of pages which are not associated
777 * with their zone properly. Not nice but set_pfnblock_flags_mask
778 * expects the zone spans the pfn range. All the pages in the range
779 * are reserved so nobody should be touching them so we should be safe
781 memmap_init_zone(nr_pages
, nid
, zone_idx(zone
), start_pfn
,
782 MEMMAP_HOTPLUG
, altmap
);
784 set_zone_contiguous(zone
);
788 * Returns a default kernel memory zone for the given pfn range.
789 * If no kernel zone covers this pfn range it will automatically go
790 * to the ZONE_NORMAL.
792 static struct zone
*default_kernel_zone_for_pfn(int nid
, unsigned long start_pfn
,
793 unsigned long nr_pages
)
795 struct pglist_data
*pgdat
= NODE_DATA(nid
);
798 for (zid
= 0; zid
<= ZONE_NORMAL
; zid
++) {
799 struct zone
*zone
= &pgdat
->node_zones
[zid
];
801 if (zone_intersects(zone
, start_pfn
, nr_pages
))
805 return &pgdat
->node_zones
[ZONE_NORMAL
];
808 static inline struct zone
*default_zone_for_pfn(int nid
, unsigned long start_pfn
,
809 unsigned long nr_pages
)
811 struct zone
*kernel_zone
= default_kernel_zone_for_pfn(nid
, start_pfn
,
813 struct zone
*movable_zone
= &NODE_DATA(nid
)->node_zones
[ZONE_MOVABLE
];
814 bool in_kernel
= zone_intersects(kernel_zone
, start_pfn
, nr_pages
);
815 bool in_movable
= zone_intersects(movable_zone
, start_pfn
, nr_pages
);
818 * We inherit the existing zone in a simple case where zones do not
819 * overlap in the given range
821 if (in_kernel
^ in_movable
)
822 return (in_kernel
) ? kernel_zone
: movable_zone
;
825 * If the range doesn't belong to any zone or two zones overlap in the
826 * given range then we use movable zone only if movable_node is
827 * enabled because we always online to a kernel zone by default.
829 return movable_node_enabled
? movable_zone
: kernel_zone
;
832 struct zone
* zone_for_pfn_range(int online_type
, int nid
, unsigned start_pfn
,
833 unsigned long nr_pages
)
835 if (online_type
== MMOP_ONLINE_KERNEL
)
836 return default_kernel_zone_for_pfn(nid
, start_pfn
, nr_pages
);
838 if (online_type
== MMOP_ONLINE_MOVABLE
)
839 return &NODE_DATA(nid
)->node_zones
[ZONE_MOVABLE
];
841 return default_zone_for_pfn(nid
, start_pfn
, nr_pages
);
845 * Associates the given pfn range with the given node and the zone appropriate
846 * for the given online type.
848 static struct zone
* __meminit
move_pfn_range(int online_type
, int nid
,
849 unsigned long start_pfn
, unsigned long nr_pages
)
853 zone
= zone_for_pfn_range(online_type
, nid
, start_pfn
, nr_pages
);
854 move_pfn_range_to_zone(zone
, start_pfn
, nr_pages
, NULL
);
858 int __ref
online_pages(unsigned long pfn
, unsigned long nr_pages
, int online_type
)
861 unsigned long onlined_pages
= 0;
863 int need_zonelists_rebuild
= 0;
866 struct memory_notify arg
;
867 struct memory_block
*mem
;
872 * We can't use pfn_to_nid() because nid might be stored in struct page
873 * which is not yet initialized. Instead, we find nid from memory block.
875 mem
= find_memory_block(__pfn_to_section(pfn
));
877 put_device(&mem
->dev
);
879 /* associate pfn range with the zone */
880 zone
= move_pfn_range(online_type
, nid
, pfn
, nr_pages
);
883 arg
.nr_pages
= nr_pages
;
884 node_states_check_changes_online(nr_pages
, zone
, &arg
);
886 ret
= memory_notify(MEM_GOING_ONLINE
, &arg
);
887 ret
= notifier_to_errno(ret
);
889 goto failed_addition
;
892 * If this zone is not populated, then it is not in zonelist.
893 * This means the page allocator ignores this zone.
894 * So, zonelist must be updated after online.
896 if (!populated_zone(zone
)) {
897 need_zonelists_rebuild
= 1;
898 setup_zone_pageset(zone
);
901 ret
= walk_system_ram_range(pfn
, nr_pages
, &onlined_pages
,
904 if (need_zonelists_rebuild
)
905 zone_pcp_reset(zone
);
906 goto failed_addition
;
909 zone
->present_pages
+= onlined_pages
;
911 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
912 zone
->zone_pgdat
->node_present_pages
+= onlined_pages
;
913 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
916 node_states_set_node(nid
, &arg
);
917 if (need_zonelists_rebuild
)
918 build_all_zonelists(NULL
);
920 zone_pcp_update(zone
);
923 init_per_zone_wmark_min();
930 vm_total_pages
= nr_free_pagecache_pages();
932 writeback_set_ratelimit();
935 memory_notify(MEM_ONLINE
, &arg
);
940 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
941 (unsigned long long) pfn
<< PAGE_SHIFT
,
942 (((unsigned long long) pfn
+ nr_pages
) << PAGE_SHIFT
) - 1);
943 memory_notify(MEM_CANCEL_ONLINE
, &arg
);
947 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
949 static void reset_node_present_pages(pg_data_t
*pgdat
)
953 for (z
= pgdat
->node_zones
; z
< pgdat
->node_zones
+ MAX_NR_ZONES
; z
++)
954 z
->present_pages
= 0;
956 pgdat
->node_present_pages
= 0;
959 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
960 static pg_data_t __ref
*hotadd_new_pgdat(int nid
, u64 start
)
962 struct pglist_data
*pgdat
;
963 unsigned long start_pfn
= PFN_DOWN(start
);
965 pgdat
= NODE_DATA(nid
);
967 pgdat
= arch_alloc_nodedata(nid
);
971 arch_refresh_nodedata(nid
, pgdat
);
974 * Reset the nr_zones, order and classzone_idx before reuse.
975 * Note that kswapd will init kswapd_classzone_idx properly
976 * when it starts in the near future.
979 pgdat
->kswapd_order
= 0;
980 pgdat
->kswapd_classzone_idx
= 0;
983 /* we can use NODE_DATA(nid) from here */
985 pgdat
->node_id
= nid
;
986 pgdat
->node_start_pfn
= start_pfn
;
988 /* init node's zones as empty zones, we don't have any present pages.*/
989 free_area_init_core_hotplug(nid
);
990 pgdat
->per_cpu_nodestats
= alloc_percpu(struct per_cpu_nodestat
);
993 * The node we allocated has no zone fallback lists. For avoiding
994 * to access not-initialized zonelist, build here.
996 build_all_zonelists(pgdat
);
999 * When memory is hot-added, all the memory is in offline state. So
1000 * clear all zones' present_pages because they will be updated in
1001 * online_pages() and offline_pages().
1003 reset_node_managed_pages(pgdat
);
1004 reset_node_present_pages(pgdat
);
1009 static void rollback_node_hotadd(int nid
)
1011 pg_data_t
*pgdat
= NODE_DATA(nid
);
1013 arch_refresh_nodedata(nid
, NULL
);
1014 free_percpu(pgdat
->per_cpu_nodestats
);
1015 arch_free_nodedata(pgdat
);
1021 * try_online_node - online a node if offlined
1023 * @start: start addr of the node
1024 * @set_node_online: Whether we want to online the node
1025 * called by cpu_up() to online a node without onlined memory.
1028 * 1 -> a new node has been allocated
1029 * 0 -> the node is already online
1030 * -ENOMEM -> the node could not be allocated
1032 static int __try_online_node(int nid
, u64 start
, bool set_node_online
)
1037 if (node_online(nid
))
1040 pgdat
= hotadd_new_pgdat(nid
, start
);
1042 pr_err("Cannot online node %d due to NULL pgdat\n", nid
);
1047 if (set_node_online
) {
1048 node_set_online(nid
);
1049 ret
= register_one_node(nid
);
1057 * Users of this function always want to online/register the node
1059 int try_online_node(int nid
)
1063 mem_hotplug_begin();
1064 ret
= __try_online_node(nid
, 0, true);
1069 static int check_hotplug_memory_range(u64 start
, u64 size
)
1071 unsigned long block_sz
= memory_block_size_bytes();
1072 u64 block_nr_pages
= block_sz
>> PAGE_SHIFT
;
1073 u64 nr_pages
= size
>> PAGE_SHIFT
;
1074 u64 start_pfn
= PFN_DOWN(start
);
1076 /* memory range must be block size aligned */
1077 if (!nr_pages
|| !IS_ALIGNED(start_pfn
, block_nr_pages
) ||
1078 !IS_ALIGNED(nr_pages
, block_nr_pages
)) {
1079 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1080 block_sz
, start
, size
);
1087 static int online_memory_block(struct memory_block
*mem
, void *arg
)
1089 return device_online(&mem
->dev
);
1093 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1094 * and online/offline operations (triggered e.g. by sysfs).
1096 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1098 int __ref
add_memory_resource(int nid
, struct resource
*res
)
1101 bool new_node
= false;
1105 size
= resource_size(res
);
1107 ret
= check_hotplug_memory_range(start
, size
);
1111 mem_hotplug_begin();
1114 * Add new range to memblock so that when hotadd_new_pgdat() is called
1115 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1116 * this new range and calculate total pages correctly. The range will
1117 * be removed at hot-remove time.
1119 memblock_add_node(start
, size
, nid
);
1121 ret
= __try_online_node(nid
, start
, false);
1126 /* call arch's memory hotadd */
1127 ret
= arch_add_memory(nid
, start
, size
, NULL
, true);
1132 /* If sysfs file of new node can't be created, cpu on the node
1133 * can't be hot-added. There is no rollback way now.
1134 * So, check by BUG_ON() to catch it reluctantly..
1135 * We online node here. We can't roll back from here.
1137 node_set_online(nid
);
1138 ret
= __register_one_node(nid
);
1142 /* link memory sections under this node.*/
1143 ret
= link_mem_sections(nid
, PFN_DOWN(start
), PFN_UP(start
+ size
- 1));
1146 /* create new memmap entry */
1147 firmware_map_add_hotplug(start
, start
+ size
, "System RAM");
1149 /* device_online() will take the lock when calling online_pages() */
1152 /* online pages if requested */
1153 if (memhp_auto_online
)
1154 walk_memory_range(PFN_DOWN(start
), PFN_UP(start
+ size
- 1),
1155 NULL
, online_memory_block
);
1159 /* rollback pgdat allocation and others */
1161 rollback_node_hotadd(nid
);
1162 memblock_remove(start
, size
);
1167 /* requires device_hotplug_lock, see add_memory_resource() */
1168 int __ref
__add_memory(int nid
, u64 start
, u64 size
)
1170 struct resource
*res
;
1173 res
= register_memory_resource(start
, size
);
1175 return PTR_ERR(res
);
1177 ret
= add_memory_resource(nid
, res
);
1179 release_memory_resource(res
);
1183 int add_memory(int nid
, u64 start
, u64 size
)
1187 lock_device_hotplug();
1188 rc
= __add_memory(nid
, start
, size
);
1189 unlock_device_hotplug();
1193 EXPORT_SYMBOL_GPL(add_memory
);
1195 #ifdef CONFIG_MEMORY_HOTREMOVE
1197 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1198 * set and the size of the free page is given by page_order(). Using this,
1199 * the function determines if the pageblock contains only free pages.
1200 * Due to buddy contraints, a free page at least the size of a pageblock will
1201 * be located at the start of the pageblock
1203 static inline int pageblock_free(struct page
*page
)
1205 return PageBuddy(page
) && page_order(page
) >= pageblock_order
;
1208 /* Return the pfn of the start of the next active pageblock after a given pfn */
1209 static unsigned long next_active_pageblock(unsigned long pfn
)
1211 struct page
*page
= pfn_to_page(pfn
);
1213 /* Ensure the starting page is pageblock-aligned */
1214 BUG_ON(pfn
& (pageblock_nr_pages
- 1));
1216 /* If the entire pageblock is free, move to the end of free page */
1217 if (pageblock_free(page
)) {
1219 /* be careful. we don't have locks, page_order can be changed.*/
1220 order
= page_order(page
);
1221 if ((order
< MAX_ORDER
) && (order
>= pageblock_order
))
1222 return pfn
+ (1 << order
);
1225 return pfn
+ pageblock_nr_pages
;
1228 static bool is_pageblock_removable_nolock(unsigned long pfn
)
1230 struct page
*page
= pfn_to_page(pfn
);
1234 * We have to be careful here because we are iterating over memory
1235 * sections which are not zone aware so we might end up outside of
1236 * the zone but still within the section.
1237 * We have to take care about the node as well. If the node is offline
1238 * its NODE_DATA will be NULL - see page_zone.
1240 if (!node_online(page_to_nid(page
)))
1243 zone
= page_zone(page
);
1244 pfn
= page_to_pfn(page
);
1245 if (!zone_spans_pfn(zone
, pfn
))
1248 return !has_unmovable_pages(zone
, page
, 0, MIGRATE_MOVABLE
, SKIP_HWPOISON
);
1251 /* Checks if this range of memory is likely to be hot-removable. */
1252 bool is_mem_section_removable(unsigned long start_pfn
, unsigned long nr_pages
)
1254 unsigned long end_pfn
, pfn
;
1256 end_pfn
= min(start_pfn
+ nr_pages
,
1257 zone_end_pfn(page_zone(pfn_to_page(start_pfn
))));
1259 /* Check the starting page of each pageblock within the range */
1260 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
= next_active_pageblock(pfn
)) {
1261 if (!is_pageblock_removable_nolock(pfn
))
1266 /* All pageblocks in the memory block are likely to be hot-removable */
1271 * Confirm all pages in a range [start, end) belong to the same zone.
1272 * When true, return its valid [start, end).
1274 int test_pages_in_a_zone(unsigned long start_pfn
, unsigned long end_pfn
,
1275 unsigned long *valid_start
, unsigned long *valid_end
)
1277 unsigned long pfn
, sec_end_pfn
;
1278 unsigned long start
, end
;
1279 struct zone
*zone
= NULL
;
1282 for (pfn
= start_pfn
, sec_end_pfn
= SECTION_ALIGN_UP(start_pfn
+ 1);
1284 pfn
= sec_end_pfn
, sec_end_pfn
+= PAGES_PER_SECTION
) {
1285 /* Make sure the memory section is present first */
1286 if (!present_section_nr(pfn_to_section_nr(pfn
)))
1288 for (; pfn
< sec_end_pfn
&& pfn
< end_pfn
;
1289 pfn
+= MAX_ORDER_NR_PAGES
) {
1291 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1292 while ((i
< MAX_ORDER_NR_PAGES
) &&
1293 !pfn_valid_within(pfn
+ i
))
1295 if (i
== MAX_ORDER_NR_PAGES
|| pfn
+ i
>= end_pfn
)
1297 /* Check if we got outside of the zone */
1298 if (zone
&& !zone_spans_pfn(zone
, pfn
+ i
))
1300 page
= pfn_to_page(pfn
+ i
);
1301 if (zone
&& page_zone(page
) != zone
)
1305 zone
= page_zone(page
);
1306 end
= pfn
+ MAX_ORDER_NR_PAGES
;
1311 *valid_start
= start
;
1312 *valid_end
= min(end
, end_pfn
);
1320 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1321 * non-lru movable pages and hugepages). We scan pfn because it's much
1322 * easier than scanning over linked list. This function returns the pfn
1323 * of the first found movable page if it's found, otherwise 0.
1325 static unsigned long scan_movable_pages(unsigned long start
, unsigned long end
)
1329 for (pfn
= start
; pfn
< end
; pfn
++) {
1330 struct page
*page
, *head
;
1333 if (!pfn_valid(pfn
))
1335 page
= pfn_to_page(pfn
);
1338 if (__PageMovable(page
))
1341 if (!PageHuge(page
))
1343 head
= compound_head(page
);
1344 if (hugepage_migration_supported(page_hstate(head
)) &&
1345 page_huge_active(head
))
1347 skip
= (1 << compound_order(head
)) - (page
- head
);
1353 static struct page
*new_node_page(struct page
*page
, unsigned long private)
1355 int nid
= page_to_nid(page
);
1356 nodemask_t nmask
= node_states
[N_MEMORY
];
1359 * try to allocate from a different node but reuse this node if there
1360 * are no other online nodes to be used (e.g. we are offlining a part
1361 * of the only existing node)
1363 node_clear(nid
, nmask
);
1364 if (nodes_empty(nmask
))
1365 node_set(nid
, nmask
);
1367 return new_page_nodemask(page
, nid
, &nmask
);
1371 do_migrate_range(unsigned long start_pfn
, unsigned long end_pfn
)
1378 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
1379 if (!pfn_valid(pfn
))
1381 page
= pfn_to_page(pfn
);
1383 if (PageHuge(page
)) {
1384 struct page
*head
= compound_head(page
);
1385 if (compound_order(head
) > PFN_SECTION_SHIFT
) {
1389 pfn
= page_to_pfn(head
) + (1<<compound_order(head
)) - 1;
1390 isolate_huge_page(head
, &source
);
1392 } else if (PageTransHuge(page
))
1393 pfn
= page_to_pfn(compound_head(page
))
1394 + hpage_nr_pages(page
) - 1;
1397 * HWPoison pages have elevated reference counts so the migration would
1398 * fail on them. It also doesn't make any sense to migrate them in the
1399 * first place. Still try to unmap such a page in case it is still mapped
1400 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1401 * the unmap as the catch all safety net).
1403 if (PageHWPoison(page
)) {
1404 if (WARN_ON(PageLRU(page
)))
1405 isolate_lru_page(page
);
1406 if (page_mapped(page
))
1407 try_to_unmap(page
, TTU_IGNORE_MLOCK
| TTU_IGNORE_ACCESS
);
1411 if (!get_page_unless_zero(page
))
1414 * We can skip free pages. And we can deal with pages on
1415 * LRU and non-lru movable pages.
1418 ret
= isolate_lru_page(page
);
1420 ret
= isolate_movable_page(page
, ISOLATE_UNEVICTABLE
);
1421 if (!ret
) { /* Success */
1422 list_add_tail(&page
->lru
, &source
);
1423 if (!__PageMovable(page
))
1424 inc_node_page_state(page
, NR_ISOLATED_ANON
+
1425 page_is_file_cache(page
));
1428 pr_warn("failed to isolate pfn %lx\n", pfn
);
1429 dump_page(page
, "isolation failed");
1433 if (!list_empty(&source
)) {
1434 /* Allocate a new page from the nearest neighbor node */
1435 ret
= migrate_pages(&source
, new_node_page
, NULL
, 0,
1436 MIGRATE_SYNC
, MR_MEMORY_HOTPLUG
);
1438 list_for_each_entry(page
, &source
, lru
) {
1439 pr_warn("migrating pfn %lx failed ret:%d ",
1440 page_to_pfn(page
), ret
);
1441 dump_page(page
, "migration failure");
1443 putback_movable_pages(&source
);
1451 * remove from free_area[] and mark all as Reserved.
1454 offline_isolated_pages_cb(unsigned long start
, unsigned long nr_pages
,
1457 __offline_isolated_pages(start
, start
+ nr_pages
);
1462 offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
1464 walk_system_ram_range(start_pfn
, end_pfn
- start_pfn
, NULL
,
1465 offline_isolated_pages_cb
);
1469 * Check all pages in range, recoreded as memory resource, are isolated.
1472 check_pages_isolated_cb(unsigned long start_pfn
, unsigned long nr_pages
,
1476 long offlined
= *(long *)data
;
1477 ret
= test_pages_isolated(start_pfn
, start_pfn
+ nr_pages
, true);
1478 offlined
= nr_pages
;
1480 *(long *)data
+= offlined
;
1485 check_pages_isolated(unsigned long start_pfn
, unsigned long end_pfn
)
1490 ret
= walk_system_ram_range(start_pfn
, end_pfn
- start_pfn
, &offlined
,
1491 check_pages_isolated_cb
);
1493 offlined
= (long)ret
;
1497 static int __init
cmdline_parse_movable_node(char *p
)
1499 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1500 movable_node_enabled
= true;
1502 pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1506 early_param("movable_node", cmdline_parse_movable_node
);
1508 /* check which state of node_states will be changed when offline memory */
1509 static void node_states_check_changes_offline(unsigned long nr_pages
,
1510 struct zone
*zone
, struct memory_notify
*arg
)
1512 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1513 unsigned long present_pages
= 0;
1516 arg
->status_change_nid
= NUMA_NO_NODE
;
1517 arg
->status_change_nid_normal
= NUMA_NO_NODE
;
1518 arg
->status_change_nid_high
= NUMA_NO_NODE
;
1521 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1522 * If the memory to be offline is within the range
1523 * [0..ZONE_NORMAL], and it is the last present memory there,
1524 * the zones in that range will become empty after the offlining,
1525 * thus we can determine that we need to clear the node from
1526 * node_states[N_NORMAL_MEMORY].
1528 for (zt
= 0; zt
<= ZONE_NORMAL
; zt
++)
1529 present_pages
+= pgdat
->node_zones
[zt
].present_pages
;
1530 if (zone_idx(zone
) <= ZONE_NORMAL
&& nr_pages
>= present_pages
)
1531 arg
->status_change_nid_normal
= zone_to_nid(zone
);
1533 #ifdef CONFIG_HIGHMEM
1535 * node_states[N_HIGH_MEMORY] contains nodes which
1536 * have normal memory or high memory.
1537 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1538 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1539 * we determine that the zones in that range become empty,
1540 * we need to clear the node for N_HIGH_MEMORY.
1542 present_pages
+= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1543 if (zone_idx(zone
) <= ZONE_HIGHMEM
&& nr_pages
>= present_pages
)
1544 arg
->status_change_nid_high
= zone_to_nid(zone
);
1548 * We have accounted the pages from [0..ZONE_NORMAL), and
1549 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1551 * Here we count the possible pages from ZONE_MOVABLE.
1552 * If after having accounted all the pages, we see that the nr_pages
1553 * to be offlined is over or equal to the accounted pages,
1554 * we know that the node will become empty, and so, we can clear
1555 * it for N_MEMORY as well.
1557 present_pages
+= pgdat
->node_zones
[ZONE_MOVABLE
].present_pages
;
1559 if (nr_pages
>= present_pages
)
1560 arg
->status_change_nid
= zone_to_nid(zone
);
1563 static void node_states_clear_node(int node
, struct memory_notify
*arg
)
1565 if (arg
->status_change_nid_normal
>= 0)
1566 node_clear_state(node
, N_NORMAL_MEMORY
);
1568 if (arg
->status_change_nid_high
>= 0)
1569 node_clear_state(node
, N_HIGH_MEMORY
);
1571 if (arg
->status_change_nid
>= 0)
1572 node_clear_state(node
, N_MEMORY
);
1575 static int __ref
__offline_pages(unsigned long start_pfn
,
1576 unsigned long end_pfn
)
1578 unsigned long pfn
, nr_pages
;
1579 long offlined_pages
;
1580 int ret
, node
, nr_isolate_pageblock
;
1581 unsigned long flags
;
1582 unsigned long valid_start
, valid_end
;
1584 struct memory_notify arg
;
1587 mem_hotplug_begin();
1589 /* This makes hotplug much easier...and readable.
1590 we assume this for now. .*/
1591 if (!test_pages_in_a_zone(start_pfn
, end_pfn
, &valid_start
,
1594 reason
= "multizone range";
1595 goto failed_removal
;
1598 zone
= page_zone(pfn_to_page(valid_start
));
1599 node
= zone_to_nid(zone
);
1600 nr_pages
= end_pfn
- start_pfn
;
1602 /* set above range as isolated */
1603 ret
= start_isolate_page_range(start_pfn
, end_pfn
,
1605 SKIP_HWPOISON
| REPORT_FAILURE
);
1607 reason
= "failure to isolate range";
1608 goto failed_removal
;
1610 nr_isolate_pageblock
= ret
;
1612 arg
.start_pfn
= start_pfn
;
1613 arg
.nr_pages
= nr_pages
;
1614 node_states_check_changes_offline(nr_pages
, zone
, &arg
);
1616 ret
= memory_notify(MEM_GOING_OFFLINE
, &arg
);
1617 ret
= notifier_to_errno(ret
);
1619 reason
= "notifier failure";
1620 goto failed_removal_isolated
;
1624 for (pfn
= start_pfn
; pfn
;) {
1625 if (signal_pending(current
)) {
1627 reason
= "signal backoff";
1628 goto failed_removal_isolated
;
1632 lru_add_drain_all();
1634 pfn
= scan_movable_pages(pfn
, end_pfn
);
1637 * TODO: fatal migration failures should bail
1640 do_migrate_range(pfn
, end_pfn
);
1645 * Dissolve free hugepages in the memory block before doing
1646 * offlining actually in order to make hugetlbfs's object
1647 * counting consistent.
1649 ret
= dissolve_free_huge_pages(start_pfn
, end_pfn
);
1651 reason
= "failure to dissolve huge pages";
1652 goto failed_removal_isolated
;
1655 offlined_pages
= check_pages_isolated(start_pfn
, end_pfn
);
1656 } while (offlined_pages
< 0);
1658 pr_info("Offlined Pages %ld\n", offlined_pages
);
1659 /* Ok, all of our target is isolated.
1660 We cannot do rollback at this point. */
1661 offline_isolated_pages(start_pfn
, end_pfn
);
1664 * Onlining will reset pagetype flags and makes migrate type
1665 * MOVABLE, so just need to decrease the number of isolated
1666 * pageblocks zone counter here.
1668 spin_lock_irqsave(&zone
->lock
, flags
);
1669 zone
->nr_isolate_pageblock
-= nr_isolate_pageblock
;
1670 spin_unlock_irqrestore(&zone
->lock
, flags
);
1672 /* removal success */
1673 adjust_managed_page_count(pfn_to_page(start_pfn
), -offlined_pages
);
1674 zone
->present_pages
-= offlined_pages
;
1676 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
1677 zone
->zone_pgdat
->node_present_pages
-= offlined_pages
;
1678 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
1680 init_per_zone_wmark_min();
1682 if (!populated_zone(zone
)) {
1683 zone_pcp_reset(zone
);
1684 build_all_zonelists(NULL
);
1686 zone_pcp_update(zone
);
1688 node_states_clear_node(node
, &arg
);
1689 if (arg
.status_change_nid
>= 0) {
1691 kcompactd_stop(node
);
1694 vm_total_pages
= nr_free_pagecache_pages();
1695 writeback_set_ratelimit();
1697 memory_notify(MEM_OFFLINE
, &arg
);
1701 failed_removal_isolated
:
1702 undo_isolate_page_range(start_pfn
, end_pfn
, MIGRATE_MOVABLE
);
1703 memory_notify(MEM_CANCEL_OFFLINE
, &arg
);
1705 pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1706 (unsigned long long) start_pfn
<< PAGE_SHIFT
,
1707 ((unsigned long long) end_pfn
<< PAGE_SHIFT
) - 1,
1709 /* pushback to free area */
1714 int offline_pages(unsigned long start_pfn
, unsigned long nr_pages
)
1716 return __offline_pages(start_pfn
, start_pfn
+ nr_pages
);
1718 #endif /* CONFIG_MEMORY_HOTREMOVE */
1721 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1722 * @start_pfn: start pfn of the memory range
1723 * @end_pfn: end pfn of the memory range
1724 * @arg: argument passed to func
1725 * @func: callback for each memory section walked
1727 * This function walks through all present mem sections in range
1728 * [start_pfn, end_pfn) and call func on each mem section.
1730 * Returns the return value of func.
1732 int walk_memory_range(unsigned long start_pfn
, unsigned long end_pfn
,
1733 void *arg
, int (*func
)(struct memory_block
*, void *))
1735 struct memory_block
*mem
= NULL
;
1736 struct mem_section
*section
;
1737 unsigned long pfn
, section_nr
;
1740 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
1741 section_nr
= pfn_to_section_nr(pfn
);
1742 if (!present_section_nr(section_nr
))
1745 section
= __nr_to_section(section_nr
);
1746 /* same memblock? */
1748 if ((section_nr
>= mem
->start_section_nr
) &&
1749 (section_nr
<= mem
->end_section_nr
))
1752 mem
= find_memory_block_hinted(section
, mem
);
1756 ret
= func(mem
, arg
);
1758 kobject_put(&mem
->dev
.kobj
);
1764 kobject_put(&mem
->dev
.kobj
);
1769 #ifdef CONFIG_MEMORY_HOTREMOVE
1770 static int check_memblock_offlined_cb(struct memory_block
*mem
, void *arg
)
1772 int ret
= !is_memblock_offlined(mem
);
1774 if (unlikely(ret
)) {
1775 phys_addr_t beginpa
, endpa
;
1777 beginpa
= PFN_PHYS(section_nr_to_pfn(mem
->start_section_nr
));
1778 endpa
= PFN_PHYS(section_nr_to_pfn(mem
->end_section_nr
+ 1))-1;
1779 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1786 static int check_cpu_on_node(pg_data_t
*pgdat
)
1790 for_each_present_cpu(cpu
) {
1791 if (cpu_to_node(cpu
) == pgdat
->node_id
)
1793 * the cpu on this node isn't removed, and we can't
1794 * offline this node.
1806 * Offline a node if all memory sections and cpus of the node are removed.
1808 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1809 * and online/offline operations before this call.
1811 void try_offline_node(int nid
)
1813 pg_data_t
*pgdat
= NODE_DATA(nid
);
1814 unsigned long start_pfn
= pgdat
->node_start_pfn
;
1815 unsigned long end_pfn
= start_pfn
+ pgdat
->node_spanned_pages
;
1818 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
1819 unsigned long section_nr
= pfn_to_section_nr(pfn
);
1821 if (!present_section_nr(section_nr
))
1824 if (pfn_to_nid(pfn
) != nid
)
1828 * some memory sections of this node are not removed, and we
1829 * can't offline node now.
1834 if (check_cpu_on_node(pgdat
))
1838 * all memory/cpu of this node are removed, we can offline this
1841 node_set_offline(nid
);
1842 unregister_one_node(nid
);
1844 EXPORT_SYMBOL(try_offline_node
);
1849 * @start: physical address of the region to remove
1850 * @size: size of the region to remove
1852 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1853 * and online/offline operations before this call, as required by
1854 * try_offline_node().
1856 void __ref
__remove_memory(int nid
, u64 start
, u64 size
)
1860 BUG_ON(check_hotplug_memory_range(start
, size
));
1862 mem_hotplug_begin();
1865 * All memory blocks must be offlined before removing memory. Check
1866 * whether all memory blocks in question are offline and trigger a BUG()
1867 * if this is not the case.
1869 ret
= walk_memory_range(PFN_DOWN(start
), PFN_UP(start
+ size
- 1), NULL
,
1870 check_memblock_offlined_cb
);
1874 /* remove memmap entry */
1875 firmware_map_remove(start
, start
+ size
, "System RAM");
1876 memblock_free(start
, size
);
1877 memblock_remove(start
, size
);
1879 arch_remove_memory(nid
, start
, size
, NULL
);
1881 try_offline_node(nid
);
1886 void remove_memory(int nid
, u64 start
, u64 size
)
1888 lock_device_hotplug();
1889 __remove_memory(nid
, start
, size
);
1890 unlock_device_hotplug();
1892 EXPORT_SYMBOL_GPL(remove_memory
);
1893 #endif /* CONFIG_MEMORY_HOTREMOVE */