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;
273 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
274 page = virt_to_page(pgdat);
276 for (i = 0; i < nr_pages; i++, page++)
277 get_page_bootmem(node, page, NODE_INFO);
279 zone = &pgdat->node_zones[0];
280 for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
281 if (zone_is_initialized(zone)) {
282 nr_pages = zone->wait_table_hash_nr_entries
283 * sizeof(wait_queue_head_t);
284 nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
285 page = virt_to_page(zone->wait_table);
287 for (i = 0; i < nr_pages; i++, page++)
288 get_page_bootmem(node, page, NODE_INFO);
292 pfn = pgdat->node_start_pfn;
293 end_pfn = pgdat_end_pfn(pgdat);
295 /* register section info */
296 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
298 * Some platforms can assign the same pfn to multiple nodes - on
299 * node0 as well as nodeN. To avoid registering a pfn against
300 * multiple nodes we check that this pfn does not already
301 * reside in some other nodes.
303 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
304 register_page_bootmem_info_section(pfn);
307 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
309 static void __meminit grow_zone_span(struct zone *zone, unsigned long start_pfn,
310 unsigned long end_pfn)
312 unsigned long old_zone_end_pfn;
314 zone_span_writelock(zone);
316 old_zone_end_pfn = zone_end_pfn(zone);
317 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
318 zone->zone_start_pfn = start_pfn;
320 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
321 zone->zone_start_pfn;
323 zone_span_writeunlock(zone);
326 static void resize_zone(struct zone *zone, unsigned long start_pfn,
327 unsigned long end_pfn)
329 zone_span_writelock(zone);
331 if (end_pfn - start_pfn) {
332 zone->zone_start_pfn = start_pfn;
333 zone->spanned_pages = end_pfn - start_pfn;
336 * make it consist as free_area_init_core(),
337 * if spanned_pages = 0, then keep start_pfn = 0
339 zone->zone_start_pfn = 0;
340 zone->spanned_pages = 0;
343 zone_span_writeunlock(zone);
346 static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
347 unsigned long end_pfn)
349 enum zone_type zid = zone_idx(zone);
350 int nid = zone->zone_pgdat->node_id;
353 for (pfn = start_pfn; pfn < end_pfn; pfn++)
354 set_page_links(pfn_to_page(pfn), zid, nid, pfn);
357 /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
358 * alloc_bootmem_node_nopanic()/memblock_virt_alloc_node_nopanic() */
359 static int __ref ensure_zone_is_initialized(struct zone *zone,
360 unsigned long start_pfn, unsigned long num_pages)
362 if (!zone_is_initialized(zone))
363 return init_currently_empty_zone(zone, start_pfn, num_pages);
368 static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
369 unsigned long start_pfn, unsigned long end_pfn)
373 unsigned long z1_start_pfn;
375 ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn);
379 pgdat_resize_lock(z1->zone_pgdat, &flags);
381 /* can't move pfns which are higher than @z2 */
382 if (end_pfn > zone_end_pfn(z2))
384 /* the move out part must be at the left most of @z2 */
385 if (start_pfn > z2->zone_start_pfn)
387 /* must included/overlap */
388 if (end_pfn <= z2->zone_start_pfn)
391 /* use start_pfn for z1's start_pfn if z1 is empty */
392 if (!zone_is_empty(z1))
393 z1_start_pfn = z1->zone_start_pfn;
395 z1_start_pfn = start_pfn;
397 resize_zone(z1, z1_start_pfn, end_pfn);
398 resize_zone(z2, end_pfn, zone_end_pfn(z2));
400 pgdat_resize_unlock(z1->zone_pgdat, &flags);
402 fix_zone_id(z1, start_pfn, end_pfn);
406 pgdat_resize_unlock(z1->zone_pgdat, &flags);
410 static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
411 unsigned long start_pfn, unsigned long end_pfn)
415 unsigned long z2_end_pfn;
417 ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn);
421 pgdat_resize_lock(z1->zone_pgdat, &flags);
423 /* can't move pfns which are lower than @z1 */
424 if (z1->zone_start_pfn > start_pfn)
426 /* the move out part mast at the right most of @z1 */
427 if (zone_end_pfn(z1) > end_pfn)
429 /* must included/overlap */
430 if (start_pfn >= zone_end_pfn(z1))
433 /* use end_pfn for z2's end_pfn if z2 is empty */
434 if (!zone_is_empty(z2))
435 z2_end_pfn = zone_end_pfn(z2);
437 z2_end_pfn = end_pfn;
439 resize_zone(z1, z1->zone_start_pfn, start_pfn);
440 resize_zone(z2, start_pfn, z2_end_pfn);
442 pgdat_resize_unlock(z1->zone_pgdat, &flags);
444 fix_zone_id(z2, start_pfn, end_pfn);
448 pgdat_resize_unlock(z1->zone_pgdat, &flags);
452 static struct zone * __meminit move_pfn_range(int zone_shift,
453 unsigned long start_pfn, unsigned long end_pfn)
455 struct zone *zone = page_zone(pfn_to_page(start_pfn));
459 ret = move_pfn_range_left(zone + zone_shift, zone,
462 ret = move_pfn_range_right(zone, zone + zone_shift,
468 return zone + zone_shift;
471 static void __meminit grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
472 unsigned long end_pfn)
474 unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat);
476 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
477 pgdat->node_start_pfn = start_pfn;
479 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
480 pgdat->node_start_pfn;
483 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
485 struct pglist_data *pgdat = zone->zone_pgdat;
486 int nr_pages = PAGES_PER_SECTION;
487 int nid = pgdat->node_id;
489 unsigned long flags, pfn;
492 zone_type = zone - pgdat->node_zones;
493 ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
497 pgdat_resize_lock(zone->zone_pgdat, &flags);
498 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
499 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
500 phys_start_pfn + nr_pages);
501 pgdat_resize_unlock(zone->zone_pgdat, &flags);
502 memmap_init_zone(nr_pages, nid, zone_type,
503 phys_start_pfn, MEMMAP_HOTPLUG);
505 /* online_page_range is called later and expects pages reserved */
506 for (pfn = phys_start_pfn; pfn < phys_start_pfn + nr_pages; pfn++) {
510 SetPageReserved(pfn_to_page(pfn));
515 static int __meminit __add_section(int nid, struct zone *zone,
516 unsigned long phys_start_pfn)
520 if (pfn_valid(phys_start_pfn))
523 ret = sparse_add_one_section(zone, phys_start_pfn);
528 ret = __add_zone(zone, phys_start_pfn);
533 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
537 * Reasonably generic function for adding memory. It is
538 * expected that archs that support memory hotplug will
539 * call this function after deciding the zone to which to
542 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
543 unsigned long nr_pages)
547 int start_sec, end_sec;
548 struct vmem_altmap *altmap;
550 clear_zone_contiguous(zone);
552 /* during initialize mem_map, align hot-added range to section */
553 start_sec = pfn_to_section_nr(phys_start_pfn);
554 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
556 altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn));
559 * Validate altmap is within bounds of the total request
561 if (altmap->base_pfn != phys_start_pfn
562 || vmem_altmap_offset(altmap) > nr_pages) {
563 pr_warn_once("memory add fail, invalid altmap\n");
570 for (i = start_sec; i <= end_sec; i++) {
571 err = __add_section(nid, zone, section_nr_to_pfn(i));
574 * EEXIST is finally dealt with by ioresource collision
575 * check. see add_memory() => register_memory_resource()
576 * Warning will be printed if there is collision.
578 if (err && (err != -EEXIST))
582 vmemmap_populate_print_last();
584 set_zone_contiguous(zone);
587 EXPORT_SYMBOL_GPL(__add_pages);
589 #ifdef CONFIG_MEMORY_HOTREMOVE
590 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
591 static int find_smallest_section_pfn(int nid, struct zone *zone,
592 unsigned long start_pfn,
593 unsigned long end_pfn)
595 struct mem_section *ms;
597 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
598 ms = __pfn_to_section(start_pfn);
600 if (unlikely(!valid_section(ms)))
603 if (unlikely(pfn_to_nid(start_pfn) != nid))
606 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
615 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
616 static int find_biggest_section_pfn(int nid, struct zone *zone,
617 unsigned long start_pfn,
618 unsigned long end_pfn)
620 struct mem_section *ms;
623 /* pfn is the end pfn of a memory section. */
625 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
626 ms = __pfn_to_section(pfn);
628 if (unlikely(!valid_section(ms)))
631 if (unlikely(pfn_to_nid(pfn) != nid))
634 if (zone && zone != page_zone(pfn_to_page(pfn)))
643 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
644 unsigned long end_pfn)
646 unsigned long zone_start_pfn = zone->zone_start_pfn;
647 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
648 unsigned long zone_end_pfn = z;
650 struct mem_section *ms;
651 int nid = zone_to_nid(zone);
653 zone_span_writelock(zone);
654 if (zone_start_pfn == start_pfn) {
656 * If the section is smallest section in the zone, it need
657 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
658 * In this case, we find second smallest valid mem_section
659 * for shrinking zone.
661 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
664 zone->zone_start_pfn = pfn;
665 zone->spanned_pages = zone_end_pfn - pfn;
667 } else if (zone_end_pfn == end_pfn) {
669 * If the section is biggest section in the zone, it need
670 * shrink zone->spanned_pages.
671 * In this case, we find second biggest valid mem_section for
674 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
677 zone->spanned_pages = pfn - zone_start_pfn + 1;
681 * The section is not biggest or smallest mem_section in the zone, it
682 * only creates a hole in the zone. So in this case, we need not
683 * change the zone. But perhaps, the zone has only hole data. Thus
684 * it check the zone has only hole or not.
686 pfn = zone_start_pfn;
687 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
688 ms = __pfn_to_section(pfn);
690 if (unlikely(!valid_section(ms)))
693 if (page_zone(pfn_to_page(pfn)) != zone)
696 /* If the section is current section, it continues the loop */
697 if (start_pfn == pfn)
700 /* If we find valid section, we have nothing to do */
701 zone_span_writeunlock(zone);
705 /* The zone has no valid section */
706 zone->zone_start_pfn = 0;
707 zone->spanned_pages = 0;
708 zone_span_writeunlock(zone);
711 static void shrink_pgdat_span(struct pglist_data *pgdat,
712 unsigned long start_pfn, unsigned long end_pfn)
714 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
715 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
716 unsigned long pgdat_end_pfn = p;
718 struct mem_section *ms;
719 int nid = pgdat->node_id;
721 if (pgdat_start_pfn == start_pfn) {
723 * If the section is smallest section in the pgdat, it need
724 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
725 * In this case, we find second smallest valid mem_section
726 * for shrinking zone.
728 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
731 pgdat->node_start_pfn = pfn;
732 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
734 } else if (pgdat_end_pfn == end_pfn) {
736 * If the section is biggest section in the pgdat, it need
737 * shrink pgdat->node_spanned_pages.
738 * In this case, we find second biggest valid mem_section for
741 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
744 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
748 * If the section is not biggest or smallest mem_section in the pgdat,
749 * it only creates a hole in the pgdat. So in this case, we need not
751 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
752 * has only hole or not.
754 pfn = pgdat_start_pfn;
755 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
756 ms = __pfn_to_section(pfn);
758 if (unlikely(!valid_section(ms)))
761 if (pfn_to_nid(pfn) != nid)
764 /* If the section is current section, it continues the loop */
765 if (start_pfn == pfn)
768 /* If we find valid section, we have nothing to do */
772 /* The pgdat has no valid section */
773 pgdat->node_start_pfn = 0;
774 pgdat->node_spanned_pages = 0;
777 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
779 struct pglist_data *pgdat = zone->zone_pgdat;
780 int nr_pages = PAGES_PER_SECTION;
784 zone_type = zone - pgdat->node_zones;
786 pgdat_resize_lock(zone->zone_pgdat, &flags);
787 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
788 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
789 pgdat_resize_unlock(zone->zone_pgdat, &flags);
792 static int __remove_section(struct zone *zone, struct mem_section *ms,
793 unsigned long map_offset)
795 unsigned long start_pfn;
799 if (!valid_section(ms))
802 ret = unregister_memory_section(ms);
806 scn_nr = __section_nr(ms);
807 start_pfn = section_nr_to_pfn(scn_nr);
808 __remove_zone(zone, start_pfn);
810 sparse_remove_one_section(zone, ms, map_offset);
815 * __remove_pages() - remove sections of pages from a zone
816 * @zone: zone from which pages need to be removed
817 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
818 * @nr_pages: number of pages to remove (must be multiple of section size)
820 * Generic helper function to remove section mappings and sysfs entries
821 * for the section of the memory we are removing. Caller needs to make
822 * sure that pages are marked reserved and zones are adjust properly by
823 * calling offline_pages().
825 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
826 unsigned long nr_pages)
829 unsigned long map_offset = 0;
830 int sections_to_remove, ret = 0;
832 /* In the ZONE_DEVICE case device driver owns the memory region */
833 if (is_dev_zone(zone)) {
834 struct page *page = pfn_to_page(phys_start_pfn);
835 struct vmem_altmap *altmap;
837 altmap = to_vmem_altmap((unsigned long) page);
839 map_offset = vmem_altmap_offset(altmap);
841 resource_size_t start, size;
843 start = phys_start_pfn << PAGE_SHIFT;
844 size = nr_pages * PAGE_SIZE;
846 ret = release_mem_region_adjustable(&iomem_resource, start,
849 resource_size_t endres = start + size - 1;
851 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
852 &start, &endres, ret);
856 clear_zone_contiguous(zone);
859 * We can only remove entire sections
861 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
862 BUG_ON(nr_pages % PAGES_PER_SECTION);
864 sections_to_remove = nr_pages / PAGES_PER_SECTION;
865 for (i = 0; i < sections_to_remove; i++) {
866 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
868 ret = __remove_section(zone, __pfn_to_section(pfn), map_offset);
874 set_zone_contiguous(zone);
878 EXPORT_SYMBOL_GPL(__remove_pages);
879 #endif /* CONFIG_MEMORY_HOTREMOVE */
881 int set_online_page_callback(online_page_callback_t callback)
886 mutex_lock(&online_page_callback_lock);
888 if (online_page_callback == generic_online_page) {
889 online_page_callback = callback;
893 mutex_unlock(&online_page_callback_lock);
898 EXPORT_SYMBOL_GPL(set_online_page_callback);
900 int restore_online_page_callback(online_page_callback_t callback)
905 mutex_lock(&online_page_callback_lock);
907 if (online_page_callback == callback) {
908 online_page_callback = generic_online_page;
912 mutex_unlock(&online_page_callback_lock);
917 EXPORT_SYMBOL_GPL(restore_online_page_callback);
919 void __online_page_set_limits(struct page *page)
922 EXPORT_SYMBOL_GPL(__online_page_set_limits);
924 void __online_page_increment_counters(struct page *page)
926 adjust_managed_page_count(page, 1);
928 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
930 void __online_page_free(struct page *page)
932 __free_reserved_page(page);
934 EXPORT_SYMBOL_GPL(__online_page_free);
936 static void generic_online_page(struct page *page)
938 __online_page_set_limits(page);
939 __online_page_increment_counters(page);
940 __online_page_free(page);
943 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
947 unsigned long onlined_pages = *(unsigned long *)arg;
949 if (PageReserved(pfn_to_page(start_pfn)))
950 for (i = 0; i < nr_pages; i++) {
951 page = pfn_to_page(start_pfn + i);
952 (*online_page_callback)(page);
955 *(unsigned long *)arg = onlined_pages;
959 #ifdef CONFIG_MOVABLE_NODE
961 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
964 static bool can_online_high_movable(struct zone *zone)
968 #else /* CONFIG_MOVABLE_NODE */
969 /* ensure every online node has NORMAL memory */
970 static bool can_online_high_movable(struct zone *zone)
972 return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
974 #endif /* CONFIG_MOVABLE_NODE */
976 /* check which state of node_states will be changed when online memory */
977 static void node_states_check_changes_online(unsigned long nr_pages,
978 struct zone *zone, struct memory_notify *arg)
980 int nid = zone_to_nid(zone);
981 enum zone_type zone_last = ZONE_NORMAL;
984 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
985 * contains nodes which have zones of 0...ZONE_NORMAL,
986 * set zone_last to ZONE_NORMAL.
988 * If we don't have HIGHMEM nor movable node,
989 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
990 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
992 if (N_MEMORY == N_NORMAL_MEMORY)
993 zone_last = ZONE_MOVABLE;
996 * if the memory to be online is in a zone of 0...zone_last, and
997 * the zones of 0...zone_last don't have memory before online, we will
998 * need to set the node to node_states[N_NORMAL_MEMORY] after
999 * the memory is online.
1001 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
1002 arg->status_change_nid_normal = nid;
1004 arg->status_change_nid_normal = -1;
1006 #ifdef CONFIG_HIGHMEM
1008 * If we have movable node, node_states[N_HIGH_MEMORY]
1009 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1010 * set zone_last to ZONE_HIGHMEM.
1012 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1013 * contains nodes which have zones of 0...ZONE_MOVABLE,
1014 * set zone_last to ZONE_MOVABLE.
1016 zone_last = ZONE_HIGHMEM;
1017 if (N_MEMORY == N_HIGH_MEMORY)
1018 zone_last = ZONE_MOVABLE;
1020 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
1021 arg->status_change_nid_high = nid;
1023 arg->status_change_nid_high = -1;
1025 arg->status_change_nid_high = arg->status_change_nid_normal;
1029 * if the node don't have memory befor online, we will need to
1030 * set the node to node_states[N_MEMORY] after the memory
1033 if (!node_state(nid, N_MEMORY))
1034 arg->status_change_nid = nid;
1036 arg->status_change_nid = -1;
1039 static void node_states_set_node(int node, struct memory_notify *arg)
1041 if (arg->status_change_nid_normal >= 0)
1042 node_set_state(node, N_NORMAL_MEMORY);
1044 if (arg->status_change_nid_high >= 0)
1045 node_set_state(node, N_HIGH_MEMORY);
1047 node_set_state(node, N_MEMORY);
1050 int zone_can_shift(unsigned long pfn, unsigned long nr_pages,
1051 enum zone_type target)
1053 struct zone *zone = page_zone(pfn_to_page(pfn));
1054 enum zone_type idx = zone_idx(zone);
1058 /* pages must be at end of current zone */
1059 if (pfn + nr_pages != zone_end_pfn(zone))
1062 /* no zones in use between current zone and target */
1063 for (i = idx + 1; i < target; i++)
1064 if (zone_is_initialized(zone - idx + i))
1069 /* pages must be at beginning of current zone */
1070 if (pfn != zone->zone_start_pfn)
1073 /* no zones in use between current zone and target */
1074 for (i = target + 1; i < idx; i++)
1075 if (zone_is_initialized(zone - idx + i))
1079 return target - idx;
1082 /* Must be protected by mem_hotplug_begin() */
1083 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
1085 unsigned long flags;
1086 unsigned long onlined_pages = 0;
1088 int need_zonelists_rebuild = 0;
1091 struct memory_notify arg;
1095 * This doesn't need a lock to do pfn_to_page().
1096 * The section can't be removed here because of the
1097 * memory_block->state_mutex.
1099 zone = page_zone(pfn_to_page(pfn));
1101 if ((zone_idx(zone) > ZONE_NORMAL ||
1102 online_type == MMOP_ONLINE_MOVABLE) &&
1103 !can_online_high_movable(zone))
1106 if (online_type == MMOP_ONLINE_KERNEL)
1107 zone_shift = zone_can_shift(pfn, nr_pages, ZONE_NORMAL);
1108 else if (online_type == MMOP_ONLINE_MOVABLE)
1109 zone_shift = zone_can_shift(pfn, nr_pages, ZONE_MOVABLE);
1111 zone = move_pfn_range(zone_shift, pfn, pfn + nr_pages);
1115 arg.start_pfn = pfn;
1116 arg.nr_pages = nr_pages;
1117 node_states_check_changes_online(nr_pages, zone, &arg);
1119 nid = zone_to_nid(zone);
1121 ret = memory_notify(MEM_GOING_ONLINE, &arg);
1122 ret = notifier_to_errno(ret);
1124 goto failed_addition;
1127 * If this zone is not populated, then it is not in zonelist.
1128 * This means the page allocator ignores this zone.
1129 * So, zonelist must be updated after online.
1131 mutex_lock(&zonelists_mutex);
1132 if (!populated_zone(zone)) {
1133 need_zonelists_rebuild = 1;
1134 build_all_zonelists(NULL, zone);
1137 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
1138 online_pages_range);
1140 if (need_zonelists_rebuild)
1141 zone_pcp_reset(zone);
1142 mutex_unlock(&zonelists_mutex);
1143 goto failed_addition;
1146 zone->present_pages += onlined_pages;
1148 pgdat_resize_lock(zone->zone_pgdat, &flags);
1149 zone->zone_pgdat->node_present_pages += onlined_pages;
1150 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1152 if (onlined_pages) {
1153 node_states_set_node(nid, &arg);
1154 if (need_zonelists_rebuild)
1155 build_all_zonelists(NULL, NULL);
1157 zone_pcp_update(zone);
1160 mutex_unlock(&zonelists_mutex);
1162 init_per_zone_wmark_min();
1164 if (onlined_pages) {
1169 vm_total_pages = nr_free_pagecache_pages();
1171 writeback_set_ratelimit();
1174 memory_notify(MEM_ONLINE, &arg);
1178 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
1179 (unsigned long long) pfn << PAGE_SHIFT,
1180 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
1181 memory_notify(MEM_CANCEL_ONLINE, &arg);
1184 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1186 static void reset_node_present_pages(pg_data_t *pgdat)
1190 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
1191 z->present_pages = 0;
1193 pgdat->node_present_pages = 0;
1196 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1197 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1199 struct pglist_data *pgdat;
1200 unsigned long zones_size[MAX_NR_ZONES] = {0};
1201 unsigned long zholes_size[MAX_NR_ZONES] = {0};
1202 unsigned long start_pfn = PFN_DOWN(start);
1204 pgdat = NODE_DATA(nid);
1206 pgdat = arch_alloc_nodedata(nid);
1210 arch_refresh_nodedata(nid, pgdat);
1212 /* Reset the nr_zones, order and classzone_idx before reuse */
1213 pgdat->nr_zones = 0;
1214 pgdat->kswapd_order = 0;
1215 pgdat->kswapd_classzone_idx = 0;
1218 /* we can use NODE_DATA(nid) from here */
1220 /* init node's zones as empty zones, we don't have any present pages.*/
1221 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1224 * The node we allocated has no zone fallback lists. For avoiding
1225 * to access not-initialized zonelist, build here.
1227 mutex_lock(&zonelists_mutex);
1228 build_all_zonelists(pgdat, NULL);
1229 mutex_unlock(&zonelists_mutex);
1232 * zone->managed_pages is set to an approximate value in
1233 * free_area_init_core(), which will cause
1234 * /sys/device/system/node/nodeX/meminfo has wrong data.
1235 * So reset it to 0 before any memory is onlined.
1237 reset_node_managed_pages(pgdat);
1240 * When memory is hot-added, all the memory is in offline state. So
1241 * clear all zones' present_pages because they will be updated in
1242 * online_pages() and offline_pages().
1244 reset_node_present_pages(pgdat);
1249 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1251 arch_refresh_nodedata(nid, NULL);
1252 arch_free_nodedata(pgdat);
1258 * try_online_node - online a node if offlined
1260 * called by cpu_up() to online a node without onlined memory.
1262 int try_online_node(int nid)
1267 if (node_online(nid))
1270 mem_hotplug_begin();
1271 pgdat = hotadd_new_pgdat(nid, 0);
1273 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1277 node_set_online(nid);
1278 ret = register_one_node(nid);
1281 if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
1282 mutex_lock(&zonelists_mutex);
1283 build_all_zonelists(NULL, NULL);
1284 mutex_unlock(&zonelists_mutex);
1292 static int check_hotplug_memory_range(u64 start, u64 size)
1294 u64 start_pfn = PFN_DOWN(start);
1295 u64 nr_pages = size >> PAGE_SHIFT;
1297 /* Memory range must be aligned with section */
1298 if ((start_pfn & ~PAGE_SECTION_MASK) ||
1299 (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1300 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1301 (unsigned long long)start,
1302 (unsigned long long)size);
1310 * If movable zone has already been setup, newly added memory should be check.
1311 * If its address is higher than movable zone, it should be added as movable.
1312 * Without this check, movable zone may overlap with other zone.
1314 static int should_add_memory_movable(int nid, u64 start, u64 size)
1316 unsigned long start_pfn = start >> PAGE_SHIFT;
1317 pg_data_t *pgdat = NODE_DATA(nid);
1318 struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE;
1320 if (zone_is_empty(movable_zone))
1323 if (movable_zone->zone_start_pfn <= start_pfn)
1329 int zone_for_memory(int nid, u64 start, u64 size, int zone_default,
1332 #ifdef CONFIG_ZONE_DEVICE
1336 if (should_add_memory_movable(nid, start, size))
1337 return ZONE_MOVABLE;
1339 return zone_default;
1342 static int online_memory_block(struct memory_block *mem, void *arg)
1344 return memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
1347 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1348 int __ref add_memory_resource(int nid, struct resource *res, bool online)
1351 pg_data_t *pgdat = NULL;
1357 size = resource_size(res);
1359 ret = check_hotplug_memory_range(start, size);
1363 { /* Stupid hack to suppress address-never-null warning */
1364 void *p = NODE_DATA(nid);
1368 mem_hotplug_begin();
1371 * Add new range to memblock so that when hotadd_new_pgdat() is called
1372 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1373 * this new range and calculate total pages correctly. The range will
1374 * be removed at hot-remove time.
1376 memblock_add_node(start, size, nid);
1378 new_node = !node_online(nid);
1380 pgdat = hotadd_new_pgdat(nid, start);
1386 /* call arch's memory hotadd */
1387 ret = arch_add_memory(nid, start, size, false);
1392 /* we online node here. we can't roll back from here. */
1393 node_set_online(nid);
1396 ret = register_one_node(nid);
1398 * If sysfs file of new node can't create, cpu on the node
1399 * can't be hot-added. There is no rollback way now.
1400 * So, check by BUG_ON() to catch it reluctantly..
1405 /* create new memmap entry */
1406 firmware_map_add_hotplug(start, start + size, "System RAM");
1408 /* online pages if requested */
1410 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1411 NULL, online_memory_block);
1416 /* rollback pgdat allocation and others */
1418 rollback_node_hotadd(nid, pgdat);
1419 memblock_remove(start, size);
1425 EXPORT_SYMBOL_GPL(add_memory_resource);
1427 int __ref add_memory(int nid, u64 start, u64 size)
1429 struct resource *res;
1432 res = register_memory_resource(start, size);
1434 return PTR_ERR(res);
1436 ret = add_memory_resource(nid, res, memhp_auto_online);
1438 release_memory_resource(res);
1441 EXPORT_SYMBOL_GPL(add_memory);
1443 #ifdef CONFIG_MEMORY_HOTREMOVE
1445 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1446 * set and the size of the free page is given by page_order(). Using this,
1447 * the function determines if the pageblock contains only free pages.
1448 * Due to buddy contraints, a free page at least the size of a pageblock will
1449 * be located at the start of the pageblock
1451 static inline int pageblock_free(struct page *page)
1453 return PageBuddy(page) && page_order(page) >= pageblock_order;
1456 /* Return the start of the next active pageblock after a given page */
1457 static struct page *next_active_pageblock(struct page *page)
1459 /* Ensure the starting page is pageblock-aligned */
1460 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1462 /* If the entire pageblock is free, move to the end of free page */
1463 if (pageblock_free(page)) {
1465 /* be careful. we don't have locks, page_order can be changed.*/
1466 order = page_order(page);
1467 if ((order < MAX_ORDER) && (order >= pageblock_order))
1468 return page + (1 << order);
1471 return page + pageblock_nr_pages;
1474 /* Checks if this range of memory is likely to be hot-removable. */
1475 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1477 struct page *page = pfn_to_page(start_pfn);
1478 struct page *end_page = page + nr_pages;
1480 /* Check the starting page of each pageblock within the range */
1481 for (; page < end_page; page = next_active_pageblock(page)) {
1482 if (!is_pageblock_removable_nolock(page))
1487 /* All pageblocks in the memory block are likely to be hot-removable */
1492 * Confirm all pages in a range [start, end) is belongs to the same zone.
1494 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
1496 unsigned long pfn, sec_end_pfn;
1497 struct zone *zone = NULL;
1500 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn);
1502 pfn = sec_end_pfn + 1, sec_end_pfn += PAGES_PER_SECTION) {
1503 /* Make sure the memory section is present first */
1504 if (!present_section_nr(pfn_to_section_nr(pfn)))
1506 for (; pfn < sec_end_pfn && pfn < end_pfn;
1507 pfn += MAX_ORDER_NR_PAGES) {
1509 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1510 while ((i < MAX_ORDER_NR_PAGES) &&
1511 !pfn_valid_within(pfn + i))
1513 if (i == MAX_ORDER_NR_PAGES)
1515 page = pfn_to_page(pfn + i);
1516 if (zone && page_zone(page) != zone)
1518 zone = page_zone(page);
1525 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages
1526 * and hugepages). We scan pfn because it's much easier than scanning over
1527 * linked list. This function returns the pfn of the first found movable
1528 * page if it's found, otherwise 0.
1530 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1534 for (pfn = start; pfn < end; pfn++) {
1535 if (pfn_valid(pfn)) {
1536 page = pfn_to_page(pfn);
1539 if (PageHuge(page)) {
1540 if (page_huge_active(page))
1543 pfn = round_up(pfn + 1,
1544 1 << compound_order(page)) - 1;
1551 static struct page *new_node_page(struct page *page, unsigned long private,
1554 gfp_t gfp_mask = GFP_USER | __GFP_MOVABLE;
1555 int nid = page_to_nid(page);
1556 nodemask_t nmask = node_online_map;
1557 struct page *new_page;
1560 * TODO: allocate a destination hugepage from a nearest neighbor node,
1561 * accordance with memory policy of the user process if possible. For
1562 * now as a simple work-around, we use the next node for destination.
1565 return alloc_huge_page_node(page_hstate(compound_head(page)),
1566 next_node_in(nid, nmask));
1568 node_clear(nid, nmask);
1569 if (PageHighMem(page)
1570 || (zone_idx(page_zone(page)) == ZONE_MOVABLE))
1571 gfp_mask |= __GFP_HIGHMEM;
1573 new_page = __alloc_pages_nodemask(gfp_mask, 0,
1574 node_zonelist(nid, gfp_mask), &nmask);
1576 new_page = __alloc_pages(gfp_mask, 0,
1577 node_zonelist(nid, gfp_mask));
1582 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1584 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1588 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1589 int not_managed = 0;
1593 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1594 if (!pfn_valid(pfn))
1596 page = pfn_to_page(pfn);
1598 if (PageHuge(page)) {
1599 struct page *head = compound_head(page);
1600 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1601 if (compound_order(head) > PFN_SECTION_SHIFT) {
1605 if (isolate_huge_page(page, &source))
1606 move_pages -= 1 << compound_order(head);
1610 if (!get_page_unless_zero(page))
1613 * We can skip free pages. And we can only deal with pages on
1616 ret = isolate_lru_page(page);
1617 if (!ret) { /* Success */
1619 list_add_tail(&page->lru, &source);
1621 inc_node_page_state(page, NR_ISOLATED_ANON +
1622 page_is_file_cache(page));
1625 #ifdef CONFIG_DEBUG_VM
1626 pr_alert("removing pfn %lx from LRU failed\n", pfn);
1627 dump_page(page, "failed to remove from LRU");
1630 /* Because we don't have big zone->lock. we should
1631 check this again here. */
1632 if (page_count(page)) {
1639 if (!list_empty(&source)) {
1641 putback_movable_pages(&source);
1645 /* Allocate a new page from the nearest neighbor node */
1646 ret = migrate_pages(&source, new_node_page, NULL, 0,
1647 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1649 putback_movable_pages(&source);
1656 * remove from free_area[] and mark all as Reserved.
1659 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1662 __offline_isolated_pages(start, start + nr_pages);
1667 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1669 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1670 offline_isolated_pages_cb);
1674 * Check all pages in range, recoreded as memory resource, are isolated.
1677 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1681 long offlined = *(long *)data;
1682 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1683 offlined = nr_pages;
1685 *(long *)data += offlined;
1690 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1695 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1696 check_pages_isolated_cb);
1698 offlined = (long)ret;
1702 #ifdef CONFIG_MOVABLE_NODE
1704 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1707 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1711 #else /* CONFIG_MOVABLE_NODE */
1712 /* ensure the node has NORMAL memory if it is still online */
1713 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1715 struct pglist_data *pgdat = zone->zone_pgdat;
1716 unsigned long present_pages = 0;
1719 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1720 present_pages += pgdat->node_zones[zt].present_pages;
1722 if (present_pages > nr_pages)
1726 for (; zt <= ZONE_MOVABLE; zt++)
1727 present_pages += pgdat->node_zones[zt].present_pages;
1730 * we can't offline the last normal memory until all
1731 * higher memory is offlined.
1733 return present_pages == 0;
1735 #endif /* CONFIG_MOVABLE_NODE */
1737 static int __init cmdline_parse_movable_node(char *p)
1739 #ifdef CONFIG_MOVABLE_NODE
1741 * Memory used by the kernel cannot be hot-removed because Linux
1742 * cannot migrate the kernel pages. When memory hotplug is
1743 * enabled, we should prevent memblock from allocating memory
1746 * ACPI SRAT records all hotpluggable memory ranges. But before
1747 * SRAT is parsed, we don't know about it.
1749 * The kernel image is loaded into memory at very early time. We
1750 * cannot prevent this anyway. So on NUMA system, we set any
1751 * node the kernel resides in as un-hotpluggable.
1753 * Since on modern servers, one node could have double-digit
1754 * gigabytes memory, we can assume the memory around the kernel
1755 * image is also un-hotpluggable. So before SRAT is parsed, just
1756 * allocate memory near the kernel image to try the best to keep
1757 * the kernel away from hotpluggable memory.
1759 memblock_set_bottom_up(true);
1760 movable_node_enabled = true;
1762 pr_warn("movable_node option not supported\n");
1766 early_param("movable_node", cmdline_parse_movable_node);
1768 /* check which state of node_states will be changed when offline memory */
1769 static void node_states_check_changes_offline(unsigned long nr_pages,
1770 struct zone *zone, struct memory_notify *arg)
1772 struct pglist_data *pgdat = zone->zone_pgdat;
1773 unsigned long present_pages = 0;
1774 enum zone_type zt, zone_last = ZONE_NORMAL;
1777 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1778 * contains nodes which have zones of 0...ZONE_NORMAL,
1779 * set zone_last to ZONE_NORMAL.
1781 * If we don't have HIGHMEM nor movable node,
1782 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1783 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1785 if (N_MEMORY == N_NORMAL_MEMORY)
1786 zone_last = ZONE_MOVABLE;
1789 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1790 * If the memory to be offline is in a zone of 0...zone_last,
1791 * and it is the last present memory, 0...zone_last will
1792 * become empty after offline , thus we can determind we will
1793 * need to clear the node from node_states[N_NORMAL_MEMORY].
1795 for (zt = 0; zt <= zone_last; zt++)
1796 present_pages += pgdat->node_zones[zt].present_pages;
1797 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1798 arg->status_change_nid_normal = zone_to_nid(zone);
1800 arg->status_change_nid_normal = -1;
1802 #ifdef CONFIG_HIGHMEM
1804 * If we have movable node, node_states[N_HIGH_MEMORY]
1805 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1806 * set zone_last to ZONE_HIGHMEM.
1808 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1809 * contains nodes which have zones of 0...ZONE_MOVABLE,
1810 * set zone_last to ZONE_MOVABLE.
1812 zone_last = ZONE_HIGHMEM;
1813 if (N_MEMORY == N_HIGH_MEMORY)
1814 zone_last = ZONE_MOVABLE;
1816 for (; zt <= zone_last; zt++)
1817 present_pages += pgdat->node_zones[zt].present_pages;
1818 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1819 arg->status_change_nid_high = zone_to_nid(zone);
1821 arg->status_change_nid_high = -1;
1823 arg->status_change_nid_high = arg->status_change_nid_normal;
1827 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1829 zone_last = ZONE_MOVABLE;
1832 * check whether node_states[N_HIGH_MEMORY] will be changed
1833 * If we try to offline the last present @nr_pages from the node,
1834 * we can determind we will need to clear the node from
1835 * node_states[N_HIGH_MEMORY].
1837 for (; zt <= zone_last; zt++)
1838 present_pages += pgdat->node_zones[zt].present_pages;
1839 if (nr_pages >= present_pages)
1840 arg->status_change_nid = zone_to_nid(zone);
1842 arg->status_change_nid = -1;
1845 static void node_states_clear_node(int node, struct memory_notify *arg)
1847 if (arg->status_change_nid_normal >= 0)
1848 node_clear_state(node, N_NORMAL_MEMORY);
1850 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1851 (arg->status_change_nid_high >= 0))
1852 node_clear_state(node, N_HIGH_MEMORY);
1854 if ((N_MEMORY != N_HIGH_MEMORY) &&
1855 (arg->status_change_nid >= 0))
1856 node_clear_state(node, N_MEMORY);
1859 static int __ref __offline_pages(unsigned long start_pfn,
1860 unsigned long end_pfn, unsigned long timeout)
1862 unsigned long pfn, nr_pages, expire;
1863 long offlined_pages;
1864 int ret, drain, retry_max, node;
1865 unsigned long flags;
1867 struct memory_notify arg;
1869 /* at least, alignment against pageblock is necessary */
1870 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1872 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1874 /* This makes hotplug much easier...and readable.
1875 we assume this for now. .*/
1876 if (!test_pages_in_a_zone(start_pfn, end_pfn))
1879 zone = page_zone(pfn_to_page(start_pfn));
1880 node = zone_to_nid(zone);
1881 nr_pages = end_pfn - start_pfn;
1883 if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1886 /* set above range as isolated */
1887 ret = start_isolate_page_range(start_pfn, end_pfn,
1888 MIGRATE_MOVABLE, true);
1892 arg.start_pfn = start_pfn;
1893 arg.nr_pages = nr_pages;
1894 node_states_check_changes_offline(nr_pages, zone, &arg);
1896 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1897 ret = notifier_to_errno(ret);
1899 goto failed_removal;
1902 expire = jiffies + timeout;
1906 /* start memory hot removal */
1908 if (time_after(jiffies, expire))
1909 goto failed_removal;
1911 if (signal_pending(current))
1912 goto failed_removal;
1915 lru_add_drain_all();
1917 drain_all_pages(zone);
1920 pfn = scan_movable_pages(start_pfn, end_pfn);
1921 if (pfn) { /* We have movable pages */
1922 ret = do_migrate_range(pfn, end_pfn);
1928 if (--retry_max == 0)
1929 goto failed_removal;
1935 /* drain all zone's lru pagevec, this is asynchronous... */
1936 lru_add_drain_all();
1938 /* drain pcp pages, this is synchronous. */
1939 drain_all_pages(zone);
1941 * dissolve free hugepages in the memory block before doing offlining
1942 * actually in order to make hugetlbfs's object counting consistent.
1944 dissolve_free_huge_pages(start_pfn, end_pfn);
1946 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1947 if (offlined_pages < 0) {
1949 goto failed_removal;
1951 pr_info("Offlined Pages %ld\n", offlined_pages);
1952 /* Ok, all of our target is isolated.
1953 We cannot do rollback at this point. */
1954 offline_isolated_pages(start_pfn, end_pfn);
1955 /* reset pagetype flags and makes migrate type to be MOVABLE */
1956 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1957 /* removal success */
1958 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1959 zone->present_pages -= offlined_pages;
1961 pgdat_resize_lock(zone->zone_pgdat, &flags);
1962 zone->zone_pgdat->node_present_pages -= offlined_pages;
1963 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1965 init_per_zone_wmark_min();
1967 if (!populated_zone(zone)) {
1968 zone_pcp_reset(zone);
1969 mutex_lock(&zonelists_mutex);
1970 build_all_zonelists(NULL, NULL);
1971 mutex_unlock(&zonelists_mutex);
1973 zone_pcp_update(zone);
1975 node_states_clear_node(node, &arg);
1976 if (arg.status_change_nid >= 0) {
1978 kcompactd_stop(node);
1981 vm_total_pages = nr_free_pagecache_pages();
1982 writeback_set_ratelimit();
1984 memory_notify(MEM_OFFLINE, &arg);
1988 pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n",
1989 (unsigned long long) start_pfn << PAGE_SHIFT,
1990 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1991 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1992 /* pushback to free area */
1993 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1997 /* Must be protected by mem_hotplug_begin() */
1998 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
2000 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
2002 #endif /* CONFIG_MEMORY_HOTREMOVE */
2005 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
2006 * @start_pfn: start pfn of the memory range
2007 * @end_pfn: end pfn of the memory range
2008 * @arg: argument passed to func
2009 * @func: callback for each memory section walked
2011 * This function walks through all present mem sections in range
2012 * [start_pfn, end_pfn) and call func on each mem section.
2014 * Returns the return value of func.
2016 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
2017 void *arg, int (*func)(struct memory_block *, void *))
2019 struct memory_block *mem = NULL;
2020 struct mem_section *section;
2021 unsigned long pfn, section_nr;
2024 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2025 section_nr = pfn_to_section_nr(pfn);
2026 if (!present_section_nr(section_nr))
2029 section = __nr_to_section(section_nr);
2030 /* same memblock? */
2032 if ((section_nr >= mem->start_section_nr) &&
2033 (section_nr <= mem->end_section_nr))
2036 mem = find_memory_block_hinted(section, mem);
2040 ret = func(mem, arg);
2042 kobject_put(&mem->dev.kobj);
2048 kobject_put(&mem->dev.kobj);
2053 #ifdef CONFIG_MEMORY_HOTREMOVE
2054 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
2056 int ret = !is_memblock_offlined(mem);
2058 if (unlikely(ret)) {
2059 phys_addr_t beginpa, endpa;
2061 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
2062 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
2063 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
2070 static int check_cpu_on_node(pg_data_t *pgdat)
2074 for_each_present_cpu(cpu) {
2075 if (cpu_to_node(cpu) == pgdat->node_id)
2077 * the cpu on this node isn't removed, and we can't
2078 * offline this node.
2086 static void unmap_cpu_on_node(pg_data_t *pgdat)
2088 #ifdef CONFIG_ACPI_NUMA
2091 for_each_possible_cpu(cpu)
2092 if (cpu_to_node(cpu) == pgdat->node_id)
2093 numa_clear_node(cpu);
2097 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
2101 ret = check_cpu_on_node(pgdat);
2106 * the node will be offlined when we come here, so we can clear
2107 * the cpu_to_node() now.
2110 unmap_cpu_on_node(pgdat);
2117 * Offline a node if all memory sections and cpus of the node are removed.
2119 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2120 * and online/offline operations before this call.
2122 void try_offline_node(int nid)
2124 pg_data_t *pgdat = NODE_DATA(nid);
2125 unsigned long start_pfn = pgdat->node_start_pfn;
2126 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
2130 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2131 unsigned long section_nr = pfn_to_section_nr(pfn);
2133 if (!present_section_nr(section_nr))
2136 if (pfn_to_nid(pfn) != nid)
2140 * some memory sections of this node are not removed, and we
2141 * can't offline node now.
2146 if (check_and_unmap_cpu_on_node(pgdat))
2150 * all memory/cpu of this node are removed, we can offline this
2153 node_set_offline(nid);
2154 unregister_one_node(nid);
2156 /* free waittable in each zone */
2157 for (i = 0; i < MAX_NR_ZONES; i++) {
2158 struct zone *zone = pgdat->node_zones + i;
2161 * wait_table may be allocated from boot memory,
2162 * here only free if it's allocated by vmalloc.
2164 if (is_vmalloc_addr(zone->wait_table)) {
2165 vfree(zone->wait_table);
2166 zone->wait_table = NULL;
2170 EXPORT_SYMBOL(try_offline_node);
2175 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2176 * and online/offline operations before this call, as required by
2177 * try_offline_node().
2179 void __ref remove_memory(int nid, u64 start, u64 size)
2183 BUG_ON(check_hotplug_memory_range(start, size));
2185 mem_hotplug_begin();
2188 * All memory blocks must be offlined before removing memory. Check
2189 * whether all memory blocks in question are offline and trigger a BUG()
2190 * if this is not the case.
2192 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
2193 check_memblock_offlined_cb);
2197 /* remove memmap entry */
2198 firmware_map_remove(start, start + size, "System RAM");
2199 memblock_free(start, size);
2200 memblock_remove(start, size);
2202 arch_remove_memory(start, size);
2204 try_offline_node(nid);
2208 EXPORT_SYMBOL_GPL(remove_memory);
2209 #endif /* CONFIG_MEMORY_HOTREMOVE */