2 * linux/mm/memory_hotplug.c
7 #include <linux/stddef.h>
9 #include <linux/swap.h>
10 #include <linux/interrupt.h>
11 #include <linux/pagemap.h>
12 #include <linux/compiler.h>
13 #include <linux/export.h>
14 #include <linux/pagevec.h>
15 #include <linux/writeback.h>
16 #include <linux/slab.h>
17 #include <linux/sysctl.h>
18 #include <linux/cpu.h>
19 #include <linux/memory.h>
20 #include <linux/memremap.h>
21 #include <linux/memory_hotplug.h>
22 #include <linux/highmem.h>
23 #include <linux/vmalloc.h>
24 #include <linux/ioport.h>
25 #include <linux/delay.h>
26 #include <linux/migrate.h>
27 #include <linux/page-isolation.h>
28 #include <linux/pfn.h>
29 #include <linux/suspend.h>
30 #include <linux/mm_inline.h>
31 #include <linux/firmware-map.h>
32 #include <linux/stop_machine.h>
33 #include <linux/hugetlb.h>
34 #include <linux/memblock.h>
35 #include <linux/bootmem.h>
36 #include <linux/compaction.h>
38 #include <asm/tlbflush.h>
43 * online_page_callback contains pointer to current page onlining function.
44 * Initially it is generic_online_page(). If it is required it could be
45 * changed by calling set_online_page_callback() for callback registration
46 * and restore_online_page_callback() for generic callback restore.
49 static void generic_online_page(struct page *page);
51 static online_page_callback_t online_page_callback = generic_online_page;
52 static DEFINE_MUTEX(online_page_callback_lock);
54 /* The same as the cpu_hotplug lock, but for memory hotplug. */
56 struct task_struct *active_writer;
57 struct mutex lock; /* Synchronizes accesses to refcount, */
59 * Also blocks the new readers during
60 * an ongoing mem hotplug operation.
64 #ifdef CONFIG_DEBUG_LOCK_ALLOC
65 struct lockdep_map dep_map;
68 .active_writer = NULL,
69 .lock = __MUTEX_INITIALIZER(mem_hotplug.lock),
71 #ifdef CONFIG_DEBUG_LOCK_ALLOC
72 .dep_map = {.name = "mem_hotplug.lock" },
76 /* Lockdep annotations for get/put_online_mems() and mem_hotplug_begin/end() */
77 #define memhp_lock_acquire_read() lock_map_acquire_read(&mem_hotplug.dep_map)
78 #define memhp_lock_acquire() lock_map_acquire(&mem_hotplug.dep_map)
79 #define memhp_lock_release() lock_map_release(&mem_hotplug.dep_map)
81 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
82 bool memhp_auto_online;
84 bool memhp_auto_online = true;
86 EXPORT_SYMBOL_GPL(memhp_auto_online);
88 static int __init setup_memhp_default_state(char *str)
90 if (!strcmp(str, "online"))
91 memhp_auto_online = true;
92 else if (!strcmp(str, "offline"))
93 memhp_auto_online = false;
97 __setup("memhp_default_state=", setup_memhp_default_state);
99 void get_online_mems(void)
102 if (mem_hotplug.active_writer == current)
104 memhp_lock_acquire_read();
105 mutex_lock(&mem_hotplug.lock);
106 mem_hotplug.refcount++;
107 mutex_unlock(&mem_hotplug.lock);
111 void put_online_mems(void)
113 if (mem_hotplug.active_writer == current)
115 mutex_lock(&mem_hotplug.lock);
117 if (WARN_ON(!mem_hotplug.refcount))
118 mem_hotplug.refcount++; /* try to fix things up */
120 if (!--mem_hotplug.refcount && unlikely(mem_hotplug.active_writer))
121 wake_up_process(mem_hotplug.active_writer);
122 mutex_unlock(&mem_hotplug.lock);
123 memhp_lock_release();
127 void mem_hotplug_begin(void)
129 mem_hotplug.active_writer = current;
131 memhp_lock_acquire();
133 mutex_lock(&mem_hotplug.lock);
134 if (likely(!mem_hotplug.refcount))
136 __set_current_state(TASK_UNINTERRUPTIBLE);
137 mutex_unlock(&mem_hotplug.lock);
142 void mem_hotplug_done(void)
144 mem_hotplug.active_writer = NULL;
145 mutex_unlock(&mem_hotplug.lock);
146 memhp_lock_release();
149 /* add this memory to iomem resource */
150 static struct resource *register_memory_resource(u64 start, u64 size)
152 struct resource *res;
153 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
155 return ERR_PTR(-ENOMEM);
157 res->name = "System RAM";
159 res->end = start + size - 1;
160 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
161 if (request_resource(&iomem_resource, res) < 0) {
162 pr_debug("System RAM resource %pR cannot be added\n", res);
164 return ERR_PTR(-EEXIST);
169 static void release_memory_resource(struct resource *res)
173 release_resource(res);
178 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
179 void get_page_bootmem(unsigned long info, struct page *page,
182 page->lru.next = (struct list_head *) type;
183 SetPagePrivate(page);
184 set_page_private(page, info);
188 void put_page_bootmem(struct page *page)
192 type = (unsigned long) page->lru.next;
193 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
194 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
196 if (page_ref_dec_return(page) == 1) {
197 ClearPagePrivate(page);
198 set_page_private(page, 0);
199 INIT_LIST_HEAD(&page->lru);
200 free_reserved_page(page);
204 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
205 #ifndef CONFIG_SPARSEMEM_VMEMMAP
206 static void register_page_bootmem_info_section(unsigned long start_pfn)
208 unsigned long *usemap, mapsize, section_nr, i;
209 struct mem_section *ms;
210 struct page *page, *memmap;
212 section_nr = pfn_to_section_nr(start_pfn);
213 ms = __nr_to_section(section_nr);
215 /* Get section's memmap address */
216 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
219 * Get page for the memmap's phys address
220 * XXX: need more consideration for sparse_vmemmap...
222 page = virt_to_page(memmap);
223 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
224 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
226 /* remember memmap's page */
227 for (i = 0; i < mapsize; i++, page++)
228 get_page_bootmem(section_nr, page, SECTION_INFO);
230 usemap = __nr_to_section(section_nr)->pageblock_flags;
231 page = virt_to_page(usemap);
233 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
235 for (i = 0; i < mapsize; i++, page++)
236 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
239 #else /* CONFIG_SPARSEMEM_VMEMMAP */
240 static void register_page_bootmem_info_section(unsigned long start_pfn)
242 unsigned long *usemap, mapsize, section_nr, i;
243 struct mem_section *ms;
244 struct page *page, *memmap;
246 if (!pfn_valid(start_pfn))
249 section_nr = pfn_to_section_nr(start_pfn);
250 ms = __nr_to_section(section_nr);
252 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
254 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
256 usemap = __nr_to_section(section_nr)->pageblock_flags;
257 page = virt_to_page(usemap);
259 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
261 for (i = 0; i < mapsize; i++, page++)
262 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
264 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
266 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
268 unsigned long i, pfn, end_pfn, nr_pages;
269 int node = pgdat->node_id;
272 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
273 page = virt_to_page(pgdat);
275 for (i = 0; i < nr_pages; i++, page++)
276 get_page_bootmem(node, page, NODE_INFO);
278 pfn = pgdat->node_start_pfn;
279 end_pfn = pgdat_end_pfn(pgdat);
281 /* register section info */
282 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
284 * Some platforms can assign the same pfn to multiple nodes - on
285 * node0 as well as nodeN. To avoid registering a pfn against
286 * multiple nodes we check that this pfn does not already
287 * reside in some other nodes.
289 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
290 register_page_bootmem_info_section(pfn);
293 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
295 static void __meminit grow_zone_span(struct zone *zone, unsigned long start_pfn,
296 unsigned long end_pfn)
298 unsigned long old_zone_end_pfn;
300 zone_span_writelock(zone);
302 old_zone_end_pfn = zone_end_pfn(zone);
303 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
304 zone->zone_start_pfn = start_pfn;
306 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
307 zone->zone_start_pfn;
309 zone_span_writeunlock(zone);
312 static void resize_zone(struct zone *zone, unsigned long start_pfn,
313 unsigned long end_pfn)
315 zone_span_writelock(zone);
317 if (end_pfn - start_pfn) {
318 zone->zone_start_pfn = start_pfn;
319 zone->spanned_pages = end_pfn - start_pfn;
322 * make it consist as free_area_init_core(),
323 * if spanned_pages = 0, then keep start_pfn = 0
325 zone->zone_start_pfn = 0;
326 zone->spanned_pages = 0;
329 zone_span_writeunlock(zone);
332 static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
333 unsigned long end_pfn)
335 enum zone_type zid = zone_idx(zone);
336 int nid = zone->zone_pgdat->node_id;
339 for (pfn = start_pfn; pfn < end_pfn; pfn++)
340 set_page_links(pfn_to_page(pfn), zid, nid, pfn);
343 /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
344 * alloc_bootmem_node_nopanic()/memblock_virt_alloc_node_nopanic() */
345 static int __ref ensure_zone_is_initialized(struct zone *zone,
346 unsigned long start_pfn, unsigned long num_pages)
348 if (!zone_is_initialized(zone))
349 return init_currently_empty_zone(zone, start_pfn, num_pages);
354 static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
355 unsigned long start_pfn, unsigned long end_pfn)
359 unsigned long z1_start_pfn;
361 ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn);
365 pgdat_resize_lock(z1->zone_pgdat, &flags);
367 /* can't move pfns which are higher than @z2 */
368 if (end_pfn > zone_end_pfn(z2))
370 /* the move out part must be at the left most of @z2 */
371 if (start_pfn > z2->zone_start_pfn)
373 /* must included/overlap */
374 if (end_pfn <= z2->zone_start_pfn)
377 /* use start_pfn for z1's start_pfn if z1 is empty */
378 if (!zone_is_empty(z1))
379 z1_start_pfn = z1->zone_start_pfn;
381 z1_start_pfn = start_pfn;
383 resize_zone(z1, z1_start_pfn, end_pfn);
384 resize_zone(z2, end_pfn, zone_end_pfn(z2));
386 pgdat_resize_unlock(z1->zone_pgdat, &flags);
388 fix_zone_id(z1, start_pfn, end_pfn);
392 pgdat_resize_unlock(z1->zone_pgdat, &flags);
396 static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
397 unsigned long start_pfn, unsigned long end_pfn)
401 unsigned long z2_end_pfn;
403 ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn);
407 pgdat_resize_lock(z1->zone_pgdat, &flags);
409 /* can't move pfns which are lower than @z1 */
410 if (z1->zone_start_pfn > start_pfn)
412 /* the move out part mast at the right most of @z1 */
413 if (zone_end_pfn(z1) > end_pfn)
415 /* must included/overlap */
416 if (start_pfn >= zone_end_pfn(z1))
419 /* use end_pfn for z2's end_pfn if z2 is empty */
420 if (!zone_is_empty(z2))
421 z2_end_pfn = zone_end_pfn(z2);
423 z2_end_pfn = end_pfn;
425 resize_zone(z1, z1->zone_start_pfn, start_pfn);
426 resize_zone(z2, start_pfn, z2_end_pfn);
428 pgdat_resize_unlock(z1->zone_pgdat, &flags);
430 fix_zone_id(z2, start_pfn, end_pfn);
434 pgdat_resize_unlock(z1->zone_pgdat, &flags);
438 static struct zone * __meminit move_pfn_range(int zone_shift,
439 unsigned long start_pfn, unsigned long end_pfn)
441 struct zone *zone = page_zone(pfn_to_page(start_pfn));
445 ret = move_pfn_range_left(zone + zone_shift, zone,
448 ret = move_pfn_range_right(zone, zone + zone_shift,
454 return zone + zone_shift;
457 static void __meminit grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
458 unsigned long end_pfn)
460 unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat);
462 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
463 pgdat->node_start_pfn = start_pfn;
465 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
466 pgdat->node_start_pfn;
469 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
471 struct pglist_data *pgdat = zone->zone_pgdat;
472 int nr_pages = PAGES_PER_SECTION;
473 int nid = pgdat->node_id;
475 unsigned long flags, pfn;
478 zone_type = zone - pgdat->node_zones;
479 ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
483 pgdat_resize_lock(zone->zone_pgdat, &flags);
484 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
485 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
486 phys_start_pfn + nr_pages);
487 pgdat_resize_unlock(zone->zone_pgdat, &flags);
488 memmap_init_zone(nr_pages, nid, zone_type,
489 phys_start_pfn, MEMMAP_HOTPLUG);
491 /* online_page_range is called later and expects pages reserved */
492 for (pfn = phys_start_pfn; pfn < phys_start_pfn + nr_pages; pfn++) {
496 SetPageReserved(pfn_to_page(pfn));
501 static int __meminit __add_section(int nid, struct zone *zone,
502 unsigned long phys_start_pfn)
506 if (pfn_valid(phys_start_pfn))
509 ret = sparse_add_one_section(zone, phys_start_pfn);
514 ret = __add_zone(zone, phys_start_pfn);
519 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
523 * Reasonably generic function for adding memory. It is
524 * expected that archs that support memory hotplug will
525 * call this function after deciding the zone to which to
528 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
529 unsigned long nr_pages)
533 int start_sec, end_sec;
534 struct vmem_altmap *altmap;
536 clear_zone_contiguous(zone);
538 /* during initialize mem_map, align hot-added range to section */
539 start_sec = pfn_to_section_nr(phys_start_pfn);
540 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
542 altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn));
545 * Validate altmap is within bounds of the total request
547 if (altmap->base_pfn != phys_start_pfn
548 || vmem_altmap_offset(altmap) > nr_pages) {
549 pr_warn_once("memory add fail, invalid altmap\n");
556 for (i = start_sec; i <= end_sec; i++) {
557 err = __add_section(nid, zone, section_nr_to_pfn(i));
560 * EEXIST is finally dealt with by ioresource collision
561 * check. see add_memory() => register_memory_resource()
562 * Warning will be printed if there is collision.
564 if (err && (err != -EEXIST))
568 vmemmap_populate_print_last();
570 set_zone_contiguous(zone);
573 EXPORT_SYMBOL_GPL(__add_pages);
575 #ifdef CONFIG_MEMORY_HOTREMOVE
576 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
577 static int find_smallest_section_pfn(int nid, struct zone *zone,
578 unsigned long start_pfn,
579 unsigned long end_pfn)
581 struct mem_section *ms;
583 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
584 ms = __pfn_to_section(start_pfn);
586 if (unlikely(!valid_section(ms)))
589 if (unlikely(pfn_to_nid(start_pfn) != nid))
592 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
601 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
602 static int find_biggest_section_pfn(int nid, struct zone *zone,
603 unsigned long start_pfn,
604 unsigned long end_pfn)
606 struct mem_section *ms;
609 /* pfn is the end pfn of a memory section. */
611 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
612 ms = __pfn_to_section(pfn);
614 if (unlikely(!valid_section(ms)))
617 if (unlikely(pfn_to_nid(pfn) != nid))
620 if (zone && zone != page_zone(pfn_to_page(pfn)))
629 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
630 unsigned long end_pfn)
632 unsigned long zone_start_pfn = zone->zone_start_pfn;
633 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
634 unsigned long zone_end_pfn = z;
636 struct mem_section *ms;
637 int nid = zone_to_nid(zone);
639 zone_span_writelock(zone);
640 if (zone_start_pfn == start_pfn) {
642 * If the section is smallest section in the zone, it need
643 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
644 * In this case, we find second smallest valid mem_section
645 * for shrinking zone.
647 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
650 zone->zone_start_pfn = pfn;
651 zone->spanned_pages = zone_end_pfn - pfn;
653 } else if (zone_end_pfn == end_pfn) {
655 * If the section is biggest section in the zone, it need
656 * shrink zone->spanned_pages.
657 * In this case, we find second biggest valid mem_section for
660 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
663 zone->spanned_pages = pfn - zone_start_pfn + 1;
667 * The section is not biggest or smallest mem_section in the zone, it
668 * only creates a hole in the zone. So in this case, we need not
669 * change the zone. But perhaps, the zone has only hole data. Thus
670 * it check the zone has only hole or not.
672 pfn = zone_start_pfn;
673 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
674 ms = __pfn_to_section(pfn);
676 if (unlikely(!valid_section(ms)))
679 if (page_zone(pfn_to_page(pfn)) != zone)
682 /* If the section is current section, it continues the loop */
683 if (start_pfn == pfn)
686 /* If we find valid section, we have nothing to do */
687 zone_span_writeunlock(zone);
691 /* The zone has no valid section */
692 zone->zone_start_pfn = 0;
693 zone->spanned_pages = 0;
694 zone_span_writeunlock(zone);
697 static void shrink_pgdat_span(struct pglist_data *pgdat,
698 unsigned long start_pfn, unsigned long end_pfn)
700 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
701 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
702 unsigned long pgdat_end_pfn = p;
704 struct mem_section *ms;
705 int nid = pgdat->node_id;
707 if (pgdat_start_pfn == start_pfn) {
709 * If the section is smallest section in the pgdat, it need
710 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
711 * In this case, we find second smallest valid mem_section
712 * for shrinking zone.
714 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
717 pgdat->node_start_pfn = pfn;
718 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
720 } else if (pgdat_end_pfn == end_pfn) {
722 * If the section is biggest section in the pgdat, it need
723 * shrink pgdat->node_spanned_pages.
724 * In this case, we find second biggest valid mem_section for
727 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
730 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
734 * If the section is not biggest or smallest mem_section in the pgdat,
735 * it only creates a hole in the pgdat. So in this case, we need not
737 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
738 * has only hole or not.
740 pfn = pgdat_start_pfn;
741 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
742 ms = __pfn_to_section(pfn);
744 if (unlikely(!valid_section(ms)))
747 if (pfn_to_nid(pfn) != nid)
750 /* If the section is current section, it continues the loop */
751 if (start_pfn == pfn)
754 /* If we find valid section, we have nothing to do */
758 /* The pgdat has no valid section */
759 pgdat->node_start_pfn = 0;
760 pgdat->node_spanned_pages = 0;
763 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
765 struct pglist_data *pgdat = zone->zone_pgdat;
766 int nr_pages = PAGES_PER_SECTION;
770 zone_type = zone - pgdat->node_zones;
772 pgdat_resize_lock(zone->zone_pgdat, &flags);
773 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
774 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
775 pgdat_resize_unlock(zone->zone_pgdat, &flags);
778 static int __remove_section(struct zone *zone, struct mem_section *ms,
779 unsigned long map_offset)
781 unsigned long start_pfn;
785 if (!valid_section(ms))
788 ret = unregister_memory_section(ms);
792 scn_nr = __section_nr(ms);
793 start_pfn = section_nr_to_pfn(scn_nr);
794 __remove_zone(zone, start_pfn);
796 sparse_remove_one_section(zone, ms, map_offset);
801 * __remove_pages() - remove sections of pages from a zone
802 * @zone: zone from which pages need to be removed
803 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
804 * @nr_pages: number of pages to remove (must be multiple of section size)
806 * Generic helper function to remove section mappings and sysfs entries
807 * for the section of the memory we are removing. Caller needs to make
808 * sure that pages are marked reserved and zones are adjust properly by
809 * calling offline_pages().
811 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
812 unsigned long nr_pages)
815 unsigned long map_offset = 0;
816 int sections_to_remove, ret = 0;
818 /* In the ZONE_DEVICE case device driver owns the memory region */
819 if (is_dev_zone(zone)) {
820 struct page *page = pfn_to_page(phys_start_pfn);
821 struct vmem_altmap *altmap;
823 altmap = to_vmem_altmap((unsigned long) page);
825 map_offset = vmem_altmap_offset(altmap);
827 resource_size_t start, size;
829 start = phys_start_pfn << PAGE_SHIFT;
830 size = nr_pages * PAGE_SIZE;
832 ret = release_mem_region_adjustable(&iomem_resource, start,
835 resource_size_t endres = start + size - 1;
837 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
838 &start, &endres, ret);
842 clear_zone_contiguous(zone);
845 * We can only remove entire sections
847 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
848 BUG_ON(nr_pages % PAGES_PER_SECTION);
850 sections_to_remove = nr_pages / PAGES_PER_SECTION;
851 for (i = 0; i < sections_to_remove; i++) {
852 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
854 ret = __remove_section(zone, __pfn_to_section(pfn), map_offset);
860 set_zone_contiguous(zone);
864 EXPORT_SYMBOL_GPL(__remove_pages);
865 #endif /* CONFIG_MEMORY_HOTREMOVE */
867 int set_online_page_callback(online_page_callback_t callback)
872 mutex_lock(&online_page_callback_lock);
874 if (online_page_callback == generic_online_page) {
875 online_page_callback = callback;
879 mutex_unlock(&online_page_callback_lock);
884 EXPORT_SYMBOL_GPL(set_online_page_callback);
886 int restore_online_page_callback(online_page_callback_t callback)
891 mutex_lock(&online_page_callback_lock);
893 if (online_page_callback == callback) {
894 online_page_callback = generic_online_page;
898 mutex_unlock(&online_page_callback_lock);
903 EXPORT_SYMBOL_GPL(restore_online_page_callback);
905 void __online_page_set_limits(struct page *page)
908 EXPORT_SYMBOL_GPL(__online_page_set_limits);
910 void __online_page_increment_counters(struct page *page)
912 adjust_managed_page_count(page, 1);
914 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
916 void __online_page_free(struct page *page)
918 __free_reserved_page(page);
920 EXPORT_SYMBOL_GPL(__online_page_free);
922 static void generic_online_page(struct page *page)
924 __online_page_set_limits(page);
925 __online_page_increment_counters(page);
926 __online_page_free(page);
929 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
933 unsigned long onlined_pages = *(unsigned long *)arg;
935 if (PageReserved(pfn_to_page(start_pfn)))
936 for (i = 0; i < nr_pages; i++) {
937 page = pfn_to_page(start_pfn + i);
938 (*online_page_callback)(page);
941 *(unsigned long *)arg = onlined_pages;
945 #ifdef CONFIG_MOVABLE_NODE
947 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
950 static bool can_online_high_movable(struct zone *zone)
954 #else /* CONFIG_MOVABLE_NODE */
955 /* ensure every online node has NORMAL memory */
956 static bool can_online_high_movable(struct zone *zone)
958 return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
960 #endif /* CONFIG_MOVABLE_NODE */
962 /* check which state of node_states will be changed when online memory */
963 static void node_states_check_changes_online(unsigned long nr_pages,
964 struct zone *zone, struct memory_notify *arg)
966 int nid = zone_to_nid(zone);
967 enum zone_type zone_last = ZONE_NORMAL;
970 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
971 * contains nodes which have zones of 0...ZONE_NORMAL,
972 * set zone_last to ZONE_NORMAL.
974 * If we don't have HIGHMEM nor movable node,
975 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
976 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
978 if (N_MEMORY == N_NORMAL_MEMORY)
979 zone_last = ZONE_MOVABLE;
982 * if the memory to be online is in a zone of 0...zone_last, and
983 * the zones of 0...zone_last don't have memory before online, we will
984 * need to set the node to node_states[N_NORMAL_MEMORY] after
985 * the memory is online.
987 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
988 arg->status_change_nid_normal = nid;
990 arg->status_change_nid_normal = -1;
992 #ifdef CONFIG_HIGHMEM
994 * If we have movable node, node_states[N_HIGH_MEMORY]
995 * contains nodes which have zones of 0...ZONE_HIGHMEM,
996 * set zone_last to ZONE_HIGHMEM.
998 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
999 * contains nodes which have zones of 0...ZONE_MOVABLE,
1000 * set zone_last to ZONE_MOVABLE.
1002 zone_last = ZONE_HIGHMEM;
1003 if (N_MEMORY == N_HIGH_MEMORY)
1004 zone_last = ZONE_MOVABLE;
1006 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
1007 arg->status_change_nid_high = nid;
1009 arg->status_change_nid_high = -1;
1011 arg->status_change_nid_high = arg->status_change_nid_normal;
1015 * if the node don't have memory befor online, we will need to
1016 * set the node to node_states[N_MEMORY] after the memory
1019 if (!node_state(nid, N_MEMORY))
1020 arg->status_change_nid = nid;
1022 arg->status_change_nid = -1;
1025 static void node_states_set_node(int node, struct memory_notify *arg)
1027 if (arg->status_change_nid_normal >= 0)
1028 node_set_state(node, N_NORMAL_MEMORY);
1030 if (arg->status_change_nid_high >= 0)
1031 node_set_state(node, N_HIGH_MEMORY);
1033 node_set_state(node, N_MEMORY);
1036 int zone_can_shift(unsigned long pfn, unsigned long nr_pages,
1037 enum zone_type target)
1039 struct zone *zone = page_zone(pfn_to_page(pfn));
1040 enum zone_type idx = zone_idx(zone);
1044 /* pages must be at end of current zone */
1045 if (pfn + nr_pages != zone_end_pfn(zone))
1048 /* no zones in use between current zone and target */
1049 for (i = idx + 1; i < target; i++)
1050 if (zone_is_initialized(zone - idx + i))
1055 /* pages must be at beginning of current zone */
1056 if (pfn != zone->zone_start_pfn)
1059 /* no zones in use between current zone and target */
1060 for (i = target + 1; i < idx; i++)
1061 if (zone_is_initialized(zone - idx + i))
1065 return target - idx;
1068 /* Must be protected by mem_hotplug_begin() */
1069 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
1071 unsigned long flags;
1072 unsigned long onlined_pages = 0;
1074 int need_zonelists_rebuild = 0;
1077 struct memory_notify arg;
1081 * This doesn't need a lock to do pfn_to_page().
1082 * The section can't be removed here because of the
1083 * memory_block->state_mutex.
1085 zone = page_zone(pfn_to_page(pfn));
1087 if ((zone_idx(zone) > ZONE_NORMAL ||
1088 online_type == MMOP_ONLINE_MOVABLE) &&
1089 !can_online_high_movable(zone))
1092 if (online_type == MMOP_ONLINE_KERNEL)
1093 zone_shift = zone_can_shift(pfn, nr_pages, ZONE_NORMAL);
1094 else if (online_type == MMOP_ONLINE_MOVABLE)
1095 zone_shift = zone_can_shift(pfn, nr_pages, ZONE_MOVABLE);
1097 zone = move_pfn_range(zone_shift, pfn, pfn + nr_pages);
1101 arg.start_pfn = pfn;
1102 arg.nr_pages = nr_pages;
1103 node_states_check_changes_online(nr_pages, zone, &arg);
1105 nid = zone_to_nid(zone);
1107 ret = memory_notify(MEM_GOING_ONLINE, &arg);
1108 ret = notifier_to_errno(ret);
1110 goto failed_addition;
1113 * If this zone is not populated, then it is not in zonelist.
1114 * This means the page allocator ignores this zone.
1115 * So, zonelist must be updated after online.
1117 mutex_lock(&zonelists_mutex);
1118 if (!populated_zone(zone)) {
1119 need_zonelists_rebuild = 1;
1120 build_all_zonelists(NULL, zone);
1123 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
1124 online_pages_range);
1126 if (need_zonelists_rebuild)
1127 zone_pcp_reset(zone);
1128 mutex_unlock(&zonelists_mutex);
1129 goto failed_addition;
1132 zone->present_pages += onlined_pages;
1134 pgdat_resize_lock(zone->zone_pgdat, &flags);
1135 zone->zone_pgdat->node_present_pages += onlined_pages;
1136 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1138 if (onlined_pages) {
1139 node_states_set_node(nid, &arg);
1140 if (need_zonelists_rebuild)
1141 build_all_zonelists(NULL, NULL);
1143 zone_pcp_update(zone);
1146 mutex_unlock(&zonelists_mutex);
1148 init_per_zone_wmark_min();
1150 if (onlined_pages) {
1155 vm_total_pages = nr_free_pagecache_pages();
1157 writeback_set_ratelimit();
1160 memory_notify(MEM_ONLINE, &arg);
1164 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
1165 (unsigned long long) pfn << PAGE_SHIFT,
1166 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
1167 memory_notify(MEM_CANCEL_ONLINE, &arg);
1170 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1172 static void reset_node_present_pages(pg_data_t *pgdat)
1176 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
1177 z->present_pages = 0;
1179 pgdat->node_present_pages = 0;
1182 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1183 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1185 struct pglist_data *pgdat;
1186 unsigned long zones_size[MAX_NR_ZONES] = {0};
1187 unsigned long zholes_size[MAX_NR_ZONES] = {0};
1188 unsigned long start_pfn = PFN_DOWN(start);
1190 pgdat = NODE_DATA(nid);
1192 pgdat = arch_alloc_nodedata(nid);
1196 arch_refresh_nodedata(nid, pgdat);
1198 /* Reset the nr_zones, order and classzone_idx before reuse */
1199 pgdat->nr_zones = 0;
1200 pgdat->kswapd_order = 0;
1201 pgdat->kswapd_classzone_idx = 0;
1204 /* we can use NODE_DATA(nid) from here */
1206 /* init node's zones as empty zones, we don't have any present pages.*/
1207 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1208 pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
1211 * The node we allocated has no zone fallback lists. For avoiding
1212 * to access not-initialized zonelist, build here.
1214 mutex_lock(&zonelists_mutex);
1215 build_all_zonelists(pgdat, NULL);
1216 mutex_unlock(&zonelists_mutex);
1219 * zone->managed_pages is set to an approximate value in
1220 * free_area_init_core(), which will cause
1221 * /sys/device/system/node/nodeX/meminfo has wrong data.
1222 * So reset it to 0 before any memory is onlined.
1224 reset_node_managed_pages(pgdat);
1227 * When memory is hot-added, all the memory is in offline state. So
1228 * clear all zones' present_pages because they will be updated in
1229 * online_pages() and offline_pages().
1231 reset_node_present_pages(pgdat);
1236 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1238 arch_refresh_nodedata(nid, NULL);
1239 free_percpu(pgdat->per_cpu_nodestats);
1240 arch_free_nodedata(pgdat);
1246 * try_online_node - online a node if offlined
1248 * called by cpu_up() to online a node without onlined memory.
1250 int try_online_node(int nid)
1255 if (node_online(nid))
1258 mem_hotplug_begin();
1259 pgdat = hotadd_new_pgdat(nid, 0);
1261 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1265 node_set_online(nid);
1266 ret = register_one_node(nid);
1269 if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
1270 mutex_lock(&zonelists_mutex);
1271 build_all_zonelists(NULL, NULL);
1272 mutex_unlock(&zonelists_mutex);
1280 static int check_hotplug_memory_range(u64 start, u64 size)
1282 u64 start_pfn = PFN_DOWN(start);
1283 u64 nr_pages = size >> PAGE_SHIFT;
1285 /* Memory range must be aligned with section */
1286 if ((start_pfn & ~PAGE_SECTION_MASK) ||
1287 (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1288 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1289 (unsigned long long)start,
1290 (unsigned long long)size);
1298 * If movable zone has already been setup, newly added memory should be check.
1299 * If its address is higher than movable zone, it should be added as movable.
1300 * Without this check, movable zone may overlap with other zone.
1302 static int should_add_memory_movable(int nid, u64 start, u64 size)
1304 unsigned long start_pfn = start >> PAGE_SHIFT;
1305 pg_data_t *pgdat = NODE_DATA(nid);
1306 struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE;
1308 if (zone_is_empty(movable_zone))
1311 if (movable_zone->zone_start_pfn <= start_pfn)
1317 int zone_for_memory(int nid, u64 start, u64 size, int zone_default,
1320 #ifdef CONFIG_ZONE_DEVICE
1324 if (should_add_memory_movable(nid, start, size))
1325 return ZONE_MOVABLE;
1327 return zone_default;
1330 static int online_memory_block(struct memory_block *mem, void *arg)
1332 return memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
1335 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1336 int __ref add_memory_resource(int nid, struct resource *res, bool online)
1339 pg_data_t *pgdat = NULL;
1345 size = resource_size(res);
1347 ret = check_hotplug_memory_range(start, size);
1351 { /* Stupid hack to suppress address-never-null warning */
1352 void *p = NODE_DATA(nid);
1356 mem_hotplug_begin();
1359 * Add new range to memblock so that when hotadd_new_pgdat() is called
1360 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1361 * this new range and calculate total pages correctly. The range will
1362 * be removed at hot-remove time.
1364 memblock_add_node(start, size, nid);
1366 new_node = !node_online(nid);
1368 pgdat = hotadd_new_pgdat(nid, start);
1374 /* call arch's memory hotadd */
1375 ret = arch_add_memory(nid, start, size, false);
1380 /* we online node here. we can't roll back from here. */
1381 node_set_online(nid);
1384 ret = register_one_node(nid);
1386 * If sysfs file of new node can't create, cpu on the node
1387 * can't be hot-added. There is no rollback way now.
1388 * So, check by BUG_ON() to catch it reluctantly..
1393 /* create new memmap entry */
1394 firmware_map_add_hotplug(start, start + size, "System RAM");
1396 /* online pages if requested */
1398 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1399 NULL, online_memory_block);
1404 /* rollback pgdat allocation and others */
1406 rollback_node_hotadd(nid, pgdat);
1407 memblock_remove(start, size);
1413 EXPORT_SYMBOL_GPL(add_memory_resource);
1415 int __ref add_memory(int nid, u64 start, u64 size)
1417 struct resource *res;
1420 res = register_memory_resource(start, size);
1422 return PTR_ERR(res);
1424 ret = add_memory_resource(nid, res, memhp_auto_online);
1426 release_memory_resource(res);
1429 EXPORT_SYMBOL_GPL(add_memory);
1431 #ifdef CONFIG_MEMORY_HOTREMOVE
1433 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1434 * set and the size of the free page is given by page_order(). Using this,
1435 * the function determines if the pageblock contains only free pages.
1436 * Due to buddy contraints, a free page at least the size of a pageblock will
1437 * be located at the start of the pageblock
1439 static inline int pageblock_free(struct page *page)
1441 return PageBuddy(page) && page_order(page) >= pageblock_order;
1444 /* Return the start of the next active pageblock after a given page */
1445 static struct page *next_active_pageblock(struct page *page)
1447 /* Ensure the starting page is pageblock-aligned */
1448 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1450 /* If the entire pageblock is free, move to the end of free page */
1451 if (pageblock_free(page)) {
1453 /* be careful. we don't have locks, page_order can be changed.*/
1454 order = page_order(page);
1455 if ((order < MAX_ORDER) && (order >= pageblock_order))
1456 return page + (1 << order);
1459 return page + pageblock_nr_pages;
1462 /* Checks if this range of memory is likely to be hot-removable. */
1463 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1465 struct page *page = pfn_to_page(start_pfn);
1466 struct page *end_page = page + nr_pages;
1468 /* Check the starting page of each pageblock within the range */
1469 for (; page < end_page; page = next_active_pageblock(page)) {
1470 if (!is_pageblock_removable_nolock(page))
1475 /* All pageblocks in the memory block are likely to be hot-removable */
1480 * Confirm all pages in a range [start, end) is belongs to the same zone.
1482 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
1484 unsigned long pfn, sec_end_pfn;
1485 struct zone *zone = NULL;
1488 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn);
1490 pfn = sec_end_pfn + 1, sec_end_pfn += PAGES_PER_SECTION) {
1491 /* Make sure the memory section is present first */
1492 if (!present_section_nr(pfn_to_section_nr(pfn)))
1494 for (; pfn < sec_end_pfn && pfn < end_pfn;
1495 pfn += MAX_ORDER_NR_PAGES) {
1497 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1498 while ((i < MAX_ORDER_NR_PAGES) &&
1499 !pfn_valid_within(pfn + i))
1501 if (i == MAX_ORDER_NR_PAGES)
1503 page = pfn_to_page(pfn + i);
1504 if (zone && page_zone(page) != zone)
1506 zone = page_zone(page);
1513 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages
1514 * and hugepages). We scan pfn because it's much easier than scanning over
1515 * linked list. This function returns the pfn of the first found movable
1516 * page if it's found, otherwise 0.
1518 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1522 for (pfn = start; pfn < end; pfn++) {
1523 if (pfn_valid(pfn)) {
1524 page = pfn_to_page(pfn);
1527 if (PageHuge(page)) {
1528 if (page_huge_active(page))
1531 pfn = round_up(pfn + 1,
1532 1 << compound_order(page)) - 1;
1539 static struct page *new_node_page(struct page *page, unsigned long private,
1542 gfp_t gfp_mask = GFP_USER | __GFP_MOVABLE;
1543 int nid = page_to_nid(page);
1544 nodemask_t nmask = node_states[N_MEMORY];
1545 struct page *new_page = NULL;
1548 * TODO: allocate a destination hugepage from a nearest neighbor node,
1549 * accordance with memory policy of the user process if possible. For
1550 * now as a simple work-around, we use the next node for destination.
1553 return alloc_huge_page_node(page_hstate(compound_head(page)),
1554 next_node_in(nid, nmask));
1556 node_clear(nid, nmask);
1558 if (PageHighMem(page)
1559 || (zone_idx(page_zone(page)) == ZONE_MOVABLE))
1560 gfp_mask |= __GFP_HIGHMEM;
1562 if (!nodes_empty(nmask))
1563 new_page = __alloc_pages_nodemask(gfp_mask, 0,
1564 node_zonelist(nid, gfp_mask), &nmask);
1566 new_page = __alloc_pages(gfp_mask, 0,
1567 node_zonelist(nid, gfp_mask));
1572 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1574 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1578 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1579 int not_managed = 0;
1583 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1584 if (!pfn_valid(pfn))
1586 page = pfn_to_page(pfn);
1588 if (PageHuge(page)) {
1589 struct page *head = compound_head(page);
1590 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1591 if (compound_order(head) > PFN_SECTION_SHIFT) {
1595 if (isolate_huge_page(page, &source))
1596 move_pages -= 1 << compound_order(head);
1600 if (!get_page_unless_zero(page))
1603 * We can skip free pages. And we can only deal with pages on
1606 ret = isolate_lru_page(page);
1607 if (!ret) { /* Success */
1609 list_add_tail(&page->lru, &source);
1611 inc_node_page_state(page, NR_ISOLATED_ANON +
1612 page_is_file_cache(page));
1615 #ifdef CONFIG_DEBUG_VM
1616 pr_alert("removing pfn %lx from LRU failed\n", pfn);
1617 dump_page(page, "failed to remove from LRU");
1620 /* Because we don't have big zone->lock. we should
1621 check this again here. */
1622 if (page_count(page)) {
1629 if (!list_empty(&source)) {
1631 putback_movable_pages(&source);
1635 /* Allocate a new page from the nearest neighbor node */
1636 ret = migrate_pages(&source, new_node_page, NULL, 0,
1637 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1639 putback_movable_pages(&source);
1646 * remove from free_area[] and mark all as Reserved.
1649 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1652 __offline_isolated_pages(start, start + nr_pages);
1657 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1659 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1660 offline_isolated_pages_cb);
1664 * Check all pages in range, recoreded as memory resource, are isolated.
1667 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1671 long offlined = *(long *)data;
1672 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1673 offlined = nr_pages;
1675 *(long *)data += offlined;
1680 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1685 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1686 check_pages_isolated_cb);
1688 offlined = (long)ret;
1692 #ifdef CONFIG_MOVABLE_NODE
1694 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1697 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1701 #else /* CONFIG_MOVABLE_NODE */
1702 /* ensure the node has NORMAL memory if it is still online */
1703 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1705 struct pglist_data *pgdat = zone->zone_pgdat;
1706 unsigned long present_pages = 0;
1709 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1710 present_pages += pgdat->node_zones[zt].present_pages;
1712 if (present_pages > nr_pages)
1716 for (; zt <= ZONE_MOVABLE; zt++)
1717 present_pages += pgdat->node_zones[zt].present_pages;
1720 * we can't offline the last normal memory until all
1721 * higher memory is offlined.
1723 return present_pages == 0;
1725 #endif /* CONFIG_MOVABLE_NODE */
1727 static int __init cmdline_parse_movable_node(char *p)
1729 #ifdef CONFIG_MOVABLE_NODE
1730 movable_node_enabled = true;
1732 pr_warn("movable_node option not supported\n");
1736 early_param("movable_node", cmdline_parse_movable_node);
1738 /* check which state of node_states will be changed when offline memory */
1739 static void node_states_check_changes_offline(unsigned long nr_pages,
1740 struct zone *zone, struct memory_notify *arg)
1742 struct pglist_data *pgdat = zone->zone_pgdat;
1743 unsigned long present_pages = 0;
1744 enum zone_type zt, zone_last = ZONE_NORMAL;
1747 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1748 * contains nodes which have zones of 0...ZONE_NORMAL,
1749 * set zone_last to ZONE_NORMAL.
1751 * If we don't have HIGHMEM nor movable node,
1752 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1753 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1755 if (N_MEMORY == N_NORMAL_MEMORY)
1756 zone_last = ZONE_MOVABLE;
1759 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1760 * If the memory to be offline is in a zone of 0...zone_last,
1761 * and it is the last present memory, 0...zone_last will
1762 * become empty after offline , thus we can determind we will
1763 * need to clear the node from node_states[N_NORMAL_MEMORY].
1765 for (zt = 0; zt <= zone_last; zt++)
1766 present_pages += pgdat->node_zones[zt].present_pages;
1767 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1768 arg->status_change_nid_normal = zone_to_nid(zone);
1770 arg->status_change_nid_normal = -1;
1772 #ifdef CONFIG_HIGHMEM
1774 * If we have movable node, node_states[N_HIGH_MEMORY]
1775 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1776 * set zone_last to ZONE_HIGHMEM.
1778 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1779 * contains nodes which have zones of 0...ZONE_MOVABLE,
1780 * set zone_last to ZONE_MOVABLE.
1782 zone_last = ZONE_HIGHMEM;
1783 if (N_MEMORY == N_HIGH_MEMORY)
1784 zone_last = ZONE_MOVABLE;
1786 for (; zt <= zone_last; zt++)
1787 present_pages += pgdat->node_zones[zt].present_pages;
1788 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1789 arg->status_change_nid_high = zone_to_nid(zone);
1791 arg->status_change_nid_high = -1;
1793 arg->status_change_nid_high = arg->status_change_nid_normal;
1797 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1799 zone_last = ZONE_MOVABLE;
1802 * check whether node_states[N_HIGH_MEMORY] will be changed
1803 * If we try to offline the last present @nr_pages from the node,
1804 * we can determind we will need to clear the node from
1805 * node_states[N_HIGH_MEMORY].
1807 for (; zt <= zone_last; zt++)
1808 present_pages += pgdat->node_zones[zt].present_pages;
1809 if (nr_pages >= present_pages)
1810 arg->status_change_nid = zone_to_nid(zone);
1812 arg->status_change_nid = -1;
1815 static void node_states_clear_node(int node, struct memory_notify *arg)
1817 if (arg->status_change_nid_normal >= 0)
1818 node_clear_state(node, N_NORMAL_MEMORY);
1820 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1821 (arg->status_change_nid_high >= 0))
1822 node_clear_state(node, N_HIGH_MEMORY);
1824 if ((N_MEMORY != N_HIGH_MEMORY) &&
1825 (arg->status_change_nid >= 0))
1826 node_clear_state(node, N_MEMORY);
1829 static int __ref __offline_pages(unsigned long start_pfn,
1830 unsigned long end_pfn, unsigned long timeout)
1832 unsigned long pfn, nr_pages, expire;
1833 long offlined_pages;
1834 int ret, drain, retry_max, node;
1835 unsigned long flags;
1837 struct memory_notify arg;
1839 /* at least, alignment against pageblock is necessary */
1840 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1842 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1844 /* This makes hotplug much easier...and readable.
1845 we assume this for now. .*/
1846 if (!test_pages_in_a_zone(start_pfn, end_pfn))
1849 zone = page_zone(pfn_to_page(start_pfn));
1850 node = zone_to_nid(zone);
1851 nr_pages = end_pfn - start_pfn;
1853 if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1856 /* set above range as isolated */
1857 ret = start_isolate_page_range(start_pfn, end_pfn,
1858 MIGRATE_MOVABLE, true);
1862 arg.start_pfn = start_pfn;
1863 arg.nr_pages = nr_pages;
1864 node_states_check_changes_offline(nr_pages, zone, &arg);
1866 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1867 ret = notifier_to_errno(ret);
1869 goto failed_removal;
1872 expire = jiffies + timeout;
1876 /* start memory hot removal */
1878 if (time_after(jiffies, expire))
1879 goto failed_removal;
1881 if (signal_pending(current))
1882 goto failed_removal;
1885 lru_add_drain_all();
1887 drain_all_pages(zone);
1890 pfn = scan_movable_pages(start_pfn, end_pfn);
1891 if (pfn) { /* We have movable pages */
1892 ret = do_migrate_range(pfn, end_pfn);
1898 if (--retry_max == 0)
1899 goto failed_removal;
1905 /* drain all zone's lru pagevec, this is asynchronous... */
1906 lru_add_drain_all();
1908 /* drain pcp pages, this is synchronous. */
1909 drain_all_pages(zone);
1911 * dissolve free hugepages in the memory block before doing offlining
1912 * actually in order to make hugetlbfs's object counting consistent.
1914 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1916 goto failed_removal;
1918 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1919 if (offlined_pages < 0) {
1921 goto failed_removal;
1923 pr_info("Offlined Pages %ld\n", offlined_pages);
1924 /* Ok, all of our target is isolated.
1925 We cannot do rollback at this point. */
1926 offline_isolated_pages(start_pfn, end_pfn);
1927 /* reset pagetype flags and makes migrate type to be MOVABLE */
1928 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1929 /* removal success */
1930 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1931 zone->present_pages -= offlined_pages;
1933 pgdat_resize_lock(zone->zone_pgdat, &flags);
1934 zone->zone_pgdat->node_present_pages -= offlined_pages;
1935 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1937 init_per_zone_wmark_min();
1939 if (!populated_zone(zone)) {
1940 zone_pcp_reset(zone);
1941 mutex_lock(&zonelists_mutex);
1942 build_all_zonelists(NULL, NULL);
1943 mutex_unlock(&zonelists_mutex);
1945 zone_pcp_update(zone);
1947 node_states_clear_node(node, &arg);
1948 if (arg.status_change_nid >= 0) {
1950 kcompactd_stop(node);
1953 vm_total_pages = nr_free_pagecache_pages();
1954 writeback_set_ratelimit();
1956 memory_notify(MEM_OFFLINE, &arg);
1960 pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n",
1961 (unsigned long long) start_pfn << PAGE_SHIFT,
1962 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1963 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1964 /* pushback to free area */
1965 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1969 /* Must be protected by mem_hotplug_begin() */
1970 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1972 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1974 #endif /* CONFIG_MEMORY_HOTREMOVE */
1977 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1978 * @start_pfn: start pfn of the memory range
1979 * @end_pfn: end pfn of the memory range
1980 * @arg: argument passed to func
1981 * @func: callback for each memory section walked
1983 * This function walks through all present mem sections in range
1984 * [start_pfn, end_pfn) and call func on each mem section.
1986 * Returns the return value of func.
1988 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1989 void *arg, int (*func)(struct memory_block *, void *))
1991 struct memory_block *mem = NULL;
1992 struct mem_section *section;
1993 unsigned long pfn, section_nr;
1996 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1997 section_nr = pfn_to_section_nr(pfn);
1998 if (!present_section_nr(section_nr))
2001 section = __nr_to_section(section_nr);
2002 /* same memblock? */
2004 if ((section_nr >= mem->start_section_nr) &&
2005 (section_nr <= mem->end_section_nr))
2008 mem = find_memory_block_hinted(section, mem);
2012 ret = func(mem, arg);
2014 kobject_put(&mem->dev.kobj);
2020 kobject_put(&mem->dev.kobj);
2025 #ifdef CONFIG_MEMORY_HOTREMOVE
2026 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
2028 int ret = !is_memblock_offlined(mem);
2030 if (unlikely(ret)) {
2031 phys_addr_t beginpa, endpa;
2033 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
2034 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
2035 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
2042 static int check_cpu_on_node(pg_data_t *pgdat)
2046 for_each_present_cpu(cpu) {
2047 if (cpu_to_node(cpu) == pgdat->node_id)
2049 * the cpu on this node isn't removed, and we can't
2050 * offline this node.
2058 static void unmap_cpu_on_node(pg_data_t *pgdat)
2060 #ifdef CONFIG_ACPI_NUMA
2063 for_each_possible_cpu(cpu)
2064 if (cpu_to_node(cpu) == pgdat->node_id)
2065 numa_clear_node(cpu);
2069 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
2073 ret = check_cpu_on_node(pgdat);
2078 * the node will be offlined when we come here, so we can clear
2079 * the cpu_to_node() now.
2082 unmap_cpu_on_node(pgdat);
2089 * Offline a node if all memory sections and cpus of the node are removed.
2091 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2092 * and online/offline operations before this call.
2094 void try_offline_node(int nid)
2096 pg_data_t *pgdat = NODE_DATA(nid);
2097 unsigned long start_pfn = pgdat->node_start_pfn;
2098 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
2101 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2102 unsigned long section_nr = pfn_to_section_nr(pfn);
2104 if (!present_section_nr(section_nr))
2107 if (pfn_to_nid(pfn) != nid)
2111 * some memory sections of this node are not removed, and we
2112 * can't offline node now.
2117 if (check_and_unmap_cpu_on_node(pgdat))
2121 * all memory/cpu of this node are removed, we can offline this
2124 node_set_offline(nid);
2125 unregister_one_node(nid);
2127 EXPORT_SYMBOL(try_offline_node);
2132 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2133 * and online/offline operations before this call, as required by
2134 * try_offline_node().
2136 void __ref remove_memory(int nid, u64 start, u64 size)
2140 BUG_ON(check_hotplug_memory_range(start, size));
2142 mem_hotplug_begin();
2145 * All memory blocks must be offlined before removing memory. Check
2146 * whether all memory blocks in question are offline and trigger a BUG()
2147 * if this is not the case.
2149 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
2150 check_memblock_offlined_cb);
2154 /* remove memmap entry */
2155 firmware_map_remove(start, start + size, "System RAM");
2156 memblock_free(start, size);
2157 memblock_remove(start, size);
2159 arch_remove_memory(start, size);
2161 try_offline_node(nid);
2165 EXPORT_SYMBOL_GPL(remove_memory);
2166 #endif /* CONFIG_MEMORY_HOTREMOVE */