2 * linux/mm/memory_hotplug.c
7 #include <linux/stddef.h>
9 #include <linux/sched/signal.h>
10 #include <linux/swap.h>
11 #include <linux/interrupt.h>
12 #include <linux/pagemap.h>
13 #include <linux/compiler.h>
14 #include <linux/export.h>
15 #include <linux/pagevec.h>
16 #include <linux/writeback.h>
17 #include <linux/slab.h>
18 #include <linux/sysctl.h>
19 #include <linux/cpu.h>
20 #include <linux/memory.h>
21 #include <linux/memremap.h>
22 #include <linux/memory_hotplug.h>
23 #include <linux/highmem.h>
24 #include <linux/vmalloc.h>
25 #include <linux/ioport.h>
26 #include <linux/delay.h>
27 #include <linux/migrate.h>
28 #include <linux/page-isolation.h>
29 #include <linux/pfn.h>
30 #include <linux/suspend.h>
31 #include <linux/mm_inline.h>
32 #include <linux/firmware-map.h>
33 #include <linux/stop_machine.h>
34 #include <linux/hugetlb.h>
35 #include <linux/memblock.h>
36 #include <linux/bootmem.h>
37 #include <linux/compaction.h>
39 #include <asm/tlbflush.h>
44 * online_page_callback contains pointer to current page onlining function.
45 * Initially it is generic_online_page(). If it is required it could be
46 * changed by calling set_online_page_callback() for callback registration
47 * and restore_online_page_callback() for generic callback restore.
50 static void generic_online_page(struct page *page);
52 static online_page_callback_t online_page_callback = generic_online_page;
53 static DEFINE_MUTEX(online_page_callback_lock);
55 /* The same as the cpu_hotplug lock, but for memory hotplug. */
57 struct task_struct *active_writer;
58 struct mutex lock; /* Synchronizes accesses to refcount, */
60 * Also blocks the new readers during
61 * an ongoing mem hotplug operation.
65 #ifdef CONFIG_DEBUG_LOCK_ALLOC
66 struct lockdep_map dep_map;
69 .active_writer = NULL,
70 .lock = __MUTEX_INITIALIZER(mem_hotplug.lock),
72 #ifdef CONFIG_DEBUG_LOCK_ALLOC
73 .dep_map = {.name = "mem_hotplug.lock" },
77 /* Lockdep annotations for get/put_online_mems() and mem_hotplug_begin/end() */
78 #define memhp_lock_acquire_read() lock_map_acquire_read(&mem_hotplug.dep_map)
79 #define memhp_lock_acquire() lock_map_acquire(&mem_hotplug.dep_map)
80 #define memhp_lock_release() lock_map_release(&mem_hotplug.dep_map)
82 bool movable_node_enabled = false;
84 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
85 bool memhp_auto_online;
87 bool memhp_auto_online = true;
89 EXPORT_SYMBOL_GPL(memhp_auto_online);
91 static int __init setup_memhp_default_state(char *str)
93 if (!strcmp(str, "online"))
94 memhp_auto_online = true;
95 else if (!strcmp(str, "offline"))
96 memhp_auto_online = false;
100 __setup("memhp_default_state=", setup_memhp_default_state);
102 void get_online_mems(void)
105 if (mem_hotplug.active_writer == current)
107 memhp_lock_acquire_read();
108 mutex_lock(&mem_hotplug.lock);
109 mem_hotplug.refcount++;
110 mutex_unlock(&mem_hotplug.lock);
114 void put_online_mems(void)
116 if (mem_hotplug.active_writer == current)
118 mutex_lock(&mem_hotplug.lock);
120 if (WARN_ON(!mem_hotplug.refcount))
121 mem_hotplug.refcount++; /* try to fix things up */
123 if (!--mem_hotplug.refcount && unlikely(mem_hotplug.active_writer))
124 wake_up_process(mem_hotplug.active_writer);
125 mutex_unlock(&mem_hotplug.lock);
126 memhp_lock_release();
130 /* Serializes write accesses to mem_hotplug.active_writer. */
131 static DEFINE_MUTEX(memory_add_remove_lock);
133 void mem_hotplug_begin(void)
135 mutex_lock(&memory_add_remove_lock);
137 mem_hotplug.active_writer = current;
139 memhp_lock_acquire();
141 mutex_lock(&mem_hotplug.lock);
142 if (likely(!mem_hotplug.refcount))
144 __set_current_state(TASK_UNINTERRUPTIBLE);
145 mutex_unlock(&mem_hotplug.lock);
150 void mem_hotplug_done(void)
152 mem_hotplug.active_writer = NULL;
153 mutex_unlock(&mem_hotplug.lock);
154 memhp_lock_release();
155 mutex_unlock(&memory_add_remove_lock);
158 /* add this memory to iomem resource */
159 static struct resource *register_memory_resource(u64 start, u64 size)
161 struct resource *res;
162 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
164 return ERR_PTR(-ENOMEM);
166 res->name = "System RAM";
168 res->end = start + size - 1;
169 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
170 if (request_resource(&iomem_resource, res) < 0) {
171 pr_debug("System RAM resource %pR cannot be added\n", res);
173 return ERR_PTR(-EEXIST);
178 static void release_memory_resource(struct resource *res)
182 release_resource(res);
187 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
188 void get_page_bootmem(unsigned long info, struct page *page,
191 page->freelist = (void *)type;
192 SetPagePrivate(page);
193 set_page_private(page, info);
197 void put_page_bootmem(struct page *page)
201 type = (unsigned long) page->freelist;
202 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
203 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
205 if (page_ref_dec_return(page) == 1) {
206 page->freelist = NULL;
207 ClearPagePrivate(page);
208 set_page_private(page, 0);
209 INIT_LIST_HEAD(&page->lru);
210 free_reserved_page(page);
214 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
215 #ifndef CONFIG_SPARSEMEM_VMEMMAP
216 static void register_page_bootmem_info_section(unsigned long start_pfn)
218 unsigned long *usemap, mapsize, section_nr, i;
219 struct mem_section *ms;
220 struct page *page, *memmap;
222 section_nr = pfn_to_section_nr(start_pfn);
223 ms = __nr_to_section(section_nr);
225 /* Get section's memmap address */
226 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
229 * Get page for the memmap's phys address
230 * XXX: need more consideration for sparse_vmemmap...
232 page = virt_to_page(memmap);
233 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
234 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
236 /* remember memmap's page */
237 for (i = 0; i < mapsize; i++, page++)
238 get_page_bootmem(section_nr, page, SECTION_INFO);
240 usemap = __nr_to_section(section_nr)->pageblock_flags;
241 page = virt_to_page(usemap);
243 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
245 for (i = 0; i < mapsize; i++, page++)
246 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
249 #else /* CONFIG_SPARSEMEM_VMEMMAP */
250 static void register_page_bootmem_info_section(unsigned long start_pfn)
252 unsigned long *usemap, mapsize, section_nr, i;
253 struct mem_section *ms;
254 struct page *page, *memmap;
256 if (!pfn_valid(start_pfn))
259 section_nr = pfn_to_section_nr(start_pfn);
260 ms = __nr_to_section(section_nr);
262 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
264 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
266 usemap = __nr_to_section(section_nr)->pageblock_flags;
267 page = virt_to_page(usemap);
269 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
271 for (i = 0; i < mapsize; i++, page++)
272 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
274 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
276 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
278 unsigned long i, pfn, end_pfn, nr_pages;
279 int node = pgdat->node_id;
282 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
283 page = virt_to_page(pgdat);
285 for (i = 0; i < nr_pages; i++, page++)
286 get_page_bootmem(node, page, NODE_INFO);
288 pfn = pgdat->node_start_pfn;
289 end_pfn = pgdat_end_pfn(pgdat);
291 /* register section info */
292 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
294 * Some platforms can assign the same pfn to multiple nodes - on
295 * node0 as well as nodeN. To avoid registering a pfn against
296 * multiple nodes we check that this pfn does not already
297 * reside in some other nodes.
299 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
300 register_page_bootmem_info_section(pfn);
303 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
305 static int __meminit __add_section(int nid, unsigned long phys_start_pfn,
311 if (pfn_valid(phys_start_pfn))
314 ret = sparse_add_one_section(NODE_DATA(nid), phys_start_pfn);
319 * Make all the pages reserved so that nobody will stumble over half
321 * FIXME: We also have to associate it with a node because pfn_to_node
322 * relies on having page with the proper node.
324 for (i = 0; i < PAGES_PER_SECTION; i++) {
325 unsigned long pfn = phys_start_pfn + i;
330 page = pfn_to_page(pfn);
331 set_page_node(page, nid);
332 SetPageReserved(page);
338 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
342 * Reasonably generic function for adding memory. It is
343 * expected that archs that support memory hotplug will
344 * call this function after deciding the zone to which to
347 int __ref __add_pages(int nid, unsigned long phys_start_pfn,
348 unsigned long nr_pages, bool want_memblock)
352 int start_sec, end_sec;
353 struct vmem_altmap *altmap;
355 /* during initialize mem_map, align hot-added range to section */
356 start_sec = pfn_to_section_nr(phys_start_pfn);
357 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
359 altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn));
362 * Validate altmap is within bounds of the total request
364 if (altmap->base_pfn != phys_start_pfn
365 || vmem_altmap_offset(altmap) > nr_pages) {
366 pr_warn_once("memory add fail, invalid altmap\n");
373 for (i = start_sec; i <= end_sec; i++) {
374 err = __add_section(nid, section_nr_to_pfn(i), want_memblock);
377 * EEXIST is finally dealt with by ioresource collision
378 * check. see add_memory() => register_memory_resource()
379 * Warning will be printed if there is collision.
381 if (err && (err != -EEXIST))
385 vmemmap_populate_print_last();
389 EXPORT_SYMBOL_GPL(__add_pages);
391 #ifdef CONFIG_MEMORY_HOTREMOVE
392 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
393 static int find_smallest_section_pfn(int nid, struct zone *zone,
394 unsigned long start_pfn,
395 unsigned long end_pfn)
397 struct mem_section *ms;
399 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
400 ms = __pfn_to_section(start_pfn);
402 if (unlikely(!valid_section(ms)))
405 if (unlikely(pfn_to_nid(start_pfn) != nid))
408 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
417 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
418 static int find_biggest_section_pfn(int nid, struct zone *zone,
419 unsigned long start_pfn,
420 unsigned long end_pfn)
422 struct mem_section *ms;
425 /* pfn is the end pfn of a memory section. */
427 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
428 ms = __pfn_to_section(pfn);
430 if (unlikely(!valid_section(ms)))
433 if (unlikely(pfn_to_nid(pfn) != nid))
436 if (zone && zone != page_zone(pfn_to_page(pfn)))
445 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
446 unsigned long end_pfn)
448 unsigned long zone_start_pfn = zone->zone_start_pfn;
449 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
450 unsigned long zone_end_pfn = z;
452 struct mem_section *ms;
453 int nid = zone_to_nid(zone);
455 zone_span_writelock(zone);
456 if (zone_start_pfn == start_pfn) {
458 * If the section is smallest section in the zone, it need
459 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
460 * In this case, we find second smallest valid mem_section
461 * for shrinking zone.
463 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
466 zone->zone_start_pfn = pfn;
467 zone->spanned_pages = zone_end_pfn - pfn;
469 } else if (zone_end_pfn == end_pfn) {
471 * If the section is biggest section in the zone, it need
472 * shrink zone->spanned_pages.
473 * In this case, we find second biggest valid mem_section for
476 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
479 zone->spanned_pages = pfn - zone_start_pfn + 1;
483 * The section is not biggest or smallest mem_section in the zone, it
484 * only creates a hole in the zone. So in this case, we need not
485 * change the zone. But perhaps, the zone has only hole data. Thus
486 * it check the zone has only hole or not.
488 pfn = zone_start_pfn;
489 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
490 ms = __pfn_to_section(pfn);
492 if (unlikely(!valid_section(ms)))
495 if (page_zone(pfn_to_page(pfn)) != zone)
498 /* If the section is current section, it continues the loop */
499 if (start_pfn == pfn)
502 /* If we find valid section, we have nothing to do */
503 zone_span_writeunlock(zone);
507 /* The zone has no valid section */
508 zone->zone_start_pfn = 0;
509 zone->spanned_pages = 0;
510 zone_span_writeunlock(zone);
513 static void shrink_pgdat_span(struct pglist_data *pgdat,
514 unsigned long start_pfn, unsigned long end_pfn)
516 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
517 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
518 unsigned long pgdat_end_pfn = p;
520 struct mem_section *ms;
521 int nid = pgdat->node_id;
523 if (pgdat_start_pfn == start_pfn) {
525 * If the section is smallest section in the pgdat, it need
526 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
527 * In this case, we find second smallest valid mem_section
528 * for shrinking zone.
530 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
533 pgdat->node_start_pfn = pfn;
534 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
536 } else if (pgdat_end_pfn == end_pfn) {
538 * If the section is biggest section in the pgdat, it need
539 * shrink pgdat->node_spanned_pages.
540 * In this case, we find second biggest valid mem_section for
543 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
546 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
550 * If the section is not biggest or smallest mem_section in the pgdat,
551 * it only creates a hole in the pgdat. So in this case, we need not
553 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
554 * has only hole or not.
556 pfn = pgdat_start_pfn;
557 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
558 ms = __pfn_to_section(pfn);
560 if (unlikely(!valid_section(ms)))
563 if (pfn_to_nid(pfn) != nid)
566 /* If the section is current section, it continues the loop */
567 if (start_pfn == pfn)
570 /* If we find valid section, we have nothing to do */
574 /* The pgdat has no valid section */
575 pgdat->node_start_pfn = 0;
576 pgdat->node_spanned_pages = 0;
579 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
581 struct pglist_data *pgdat = zone->zone_pgdat;
582 int nr_pages = PAGES_PER_SECTION;
585 pgdat_resize_lock(zone->zone_pgdat, &flags);
586 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
587 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
588 pgdat_resize_unlock(zone->zone_pgdat, &flags);
591 static int __remove_section(struct zone *zone, struct mem_section *ms,
592 unsigned long map_offset)
594 unsigned long start_pfn;
598 if (!valid_section(ms))
601 ret = unregister_memory_section(ms);
605 scn_nr = __section_nr(ms);
606 start_pfn = section_nr_to_pfn(scn_nr);
607 __remove_zone(zone, start_pfn);
609 sparse_remove_one_section(zone, ms, map_offset);
614 * __remove_pages() - remove sections of pages from a zone
615 * @zone: zone from which pages need to be removed
616 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
617 * @nr_pages: number of pages to remove (must be multiple of section size)
619 * Generic helper function to remove section mappings and sysfs entries
620 * for the section of the memory we are removing. Caller needs to make
621 * sure that pages are marked reserved and zones are adjust properly by
622 * calling offline_pages().
624 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
625 unsigned long nr_pages)
628 unsigned long map_offset = 0;
629 int sections_to_remove, ret = 0;
631 /* In the ZONE_DEVICE case device driver owns the memory region */
632 if (is_dev_zone(zone)) {
633 struct page *page = pfn_to_page(phys_start_pfn);
634 struct vmem_altmap *altmap;
636 altmap = to_vmem_altmap((unsigned long) page);
638 map_offset = vmem_altmap_offset(altmap);
640 resource_size_t start, size;
642 start = phys_start_pfn << PAGE_SHIFT;
643 size = nr_pages * PAGE_SIZE;
645 ret = release_mem_region_adjustable(&iomem_resource, start,
648 resource_size_t endres = start + size - 1;
650 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
651 &start, &endres, ret);
655 clear_zone_contiguous(zone);
658 * We can only remove entire sections
660 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
661 BUG_ON(nr_pages % PAGES_PER_SECTION);
663 sections_to_remove = nr_pages / PAGES_PER_SECTION;
664 for (i = 0; i < sections_to_remove; i++) {
665 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
667 ret = __remove_section(zone, __pfn_to_section(pfn), map_offset);
673 set_zone_contiguous(zone);
677 #endif /* CONFIG_MEMORY_HOTREMOVE */
679 int set_online_page_callback(online_page_callback_t callback)
684 mutex_lock(&online_page_callback_lock);
686 if (online_page_callback == generic_online_page) {
687 online_page_callback = callback;
691 mutex_unlock(&online_page_callback_lock);
696 EXPORT_SYMBOL_GPL(set_online_page_callback);
698 int restore_online_page_callback(online_page_callback_t callback)
703 mutex_lock(&online_page_callback_lock);
705 if (online_page_callback == callback) {
706 online_page_callback = generic_online_page;
710 mutex_unlock(&online_page_callback_lock);
715 EXPORT_SYMBOL_GPL(restore_online_page_callback);
717 void __online_page_set_limits(struct page *page)
720 EXPORT_SYMBOL_GPL(__online_page_set_limits);
722 void __online_page_increment_counters(struct page *page)
724 adjust_managed_page_count(page, 1);
726 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
728 void __online_page_free(struct page *page)
730 __free_reserved_page(page);
732 EXPORT_SYMBOL_GPL(__online_page_free);
734 static void generic_online_page(struct page *page)
736 __online_page_set_limits(page);
737 __online_page_increment_counters(page);
738 __online_page_free(page);
741 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
745 unsigned long onlined_pages = *(unsigned long *)arg;
748 if (PageReserved(pfn_to_page(start_pfn)))
749 for (i = 0; i < nr_pages; i++) {
750 page = pfn_to_page(start_pfn + i);
751 (*online_page_callback)(page);
755 online_mem_sections(start_pfn, start_pfn + nr_pages);
757 *(unsigned long *)arg = onlined_pages;
761 /* check which state of node_states will be changed when online memory */
762 static void node_states_check_changes_online(unsigned long nr_pages,
763 struct zone *zone, struct memory_notify *arg)
765 int nid = zone_to_nid(zone);
766 enum zone_type zone_last = ZONE_NORMAL;
769 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
770 * contains nodes which have zones of 0...ZONE_NORMAL,
771 * set zone_last to ZONE_NORMAL.
773 * If we don't have HIGHMEM nor movable node,
774 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
775 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
777 if (N_MEMORY == N_NORMAL_MEMORY)
778 zone_last = ZONE_MOVABLE;
781 * if the memory to be online is in a zone of 0...zone_last, and
782 * the zones of 0...zone_last don't have memory before online, we will
783 * need to set the node to node_states[N_NORMAL_MEMORY] after
784 * the memory is online.
786 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
787 arg->status_change_nid_normal = nid;
789 arg->status_change_nid_normal = -1;
791 #ifdef CONFIG_HIGHMEM
793 * If we have movable node, node_states[N_HIGH_MEMORY]
794 * contains nodes which have zones of 0...ZONE_HIGHMEM,
795 * set zone_last to ZONE_HIGHMEM.
797 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
798 * contains nodes which have zones of 0...ZONE_MOVABLE,
799 * set zone_last to ZONE_MOVABLE.
801 zone_last = ZONE_HIGHMEM;
802 if (N_MEMORY == N_HIGH_MEMORY)
803 zone_last = ZONE_MOVABLE;
805 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
806 arg->status_change_nid_high = nid;
808 arg->status_change_nid_high = -1;
810 arg->status_change_nid_high = arg->status_change_nid_normal;
814 * if the node don't have memory befor online, we will need to
815 * set the node to node_states[N_MEMORY] after the memory
818 if (!node_state(nid, N_MEMORY))
819 arg->status_change_nid = nid;
821 arg->status_change_nid = -1;
824 static void node_states_set_node(int node, struct memory_notify *arg)
826 if (arg->status_change_nid_normal >= 0)
827 node_set_state(node, N_NORMAL_MEMORY);
829 if (arg->status_change_nid_high >= 0)
830 node_set_state(node, N_HIGH_MEMORY);
832 node_set_state(node, N_MEMORY);
835 bool allow_online_pfn_range(int nid, unsigned long pfn, unsigned long nr_pages, int online_type)
837 struct pglist_data *pgdat = NODE_DATA(nid);
838 struct zone *movable_zone = &pgdat->node_zones[ZONE_MOVABLE];
839 struct zone *default_zone = default_zone_for_pfn(nid, pfn, nr_pages);
842 * TODO there shouldn't be any inherent reason to have ZONE_NORMAL
843 * physically before ZONE_MOVABLE. All we need is they do not
844 * overlap. Historically we didn't allow ZONE_NORMAL after ZONE_MOVABLE
845 * though so let's stick with it for simplicity for now.
846 * TODO make sure we do not overlap with ZONE_DEVICE
848 if (online_type == MMOP_ONLINE_KERNEL) {
849 if (zone_is_empty(movable_zone))
851 return movable_zone->zone_start_pfn >= pfn + nr_pages;
852 } else if (online_type == MMOP_ONLINE_MOVABLE) {
853 return zone_end_pfn(default_zone) <= pfn;
856 /* MMOP_ONLINE_KEEP will always succeed and inherits the current zone */
857 return online_type == MMOP_ONLINE_KEEP;
860 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
861 unsigned long nr_pages)
863 unsigned long old_end_pfn = zone_end_pfn(zone);
865 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
866 zone->zone_start_pfn = start_pfn;
868 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
871 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
872 unsigned long nr_pages)
874 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
876 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
877 pgdat->node_start_pfn = start_pfn;
879 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
882 void __ref move_pfn_range_to_zone(struct zone *zone,
883 unsigned long start_pfn, unsigned long nr_pages)
885 struct pglist_data *pgdat = zone->zone_pgdat;
886 int nid = pgdat->node_id;
889 if (zone_is_empty(zone))
890 init_currently_empty_zone(zone, start_pfn, nr_pages);
892 clear_zone_contiguous(zone);
894 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
895 pgdat_resize_lock(pgdat, &flags);
896 zone_span_writelock(zone);
897 resize_zone_range(zone, start_pfn, nr_pages);
898 zone_span_writeunlock(zone);
899 resize_pgdat_range(pgdat, start_pfn, nr_pages);
900 pgdat_resize_unlock(pgdat, &flags);
903 * TODO now we have a visible range of pages which are not associated
904 * with their zone properly. Not nice but set_pfnblock_flags_mask
905 * expects the zone spans the pfn range. All the pages in the range
906 * are reserved so nobody should be touching them so we should be safe
908 memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn, MEMMAP_HOTPLUG);
910 set_zone_contiguous(zone);
914 * Returns a default kernel memory zone for the given pfn range.
915 * If no kernel zone covers this pfn range it will automatically go
916 * to the ZONE_NORMAL.
918 struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
919 unsigned long nr_pages)
921 struct pglist_data *pgdat = NODE_DATA(nid);
924 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
925 struct zone *zone = &pgdat->node_zones[zid];
927 if (zone_intersects(zone, start_pfn, nr_pages))
931 return &pgdat->node_zones[ZONE_NORMAL];
934 static inline bool movable_pfn_range(int nid, struct zone *default_zone,
935 unsigned long start_pfn, unsigned long nr_pages)
937 if (!allow_online_pfn_range(nid, start_pfn, nr_pages,
941 if (!movable_node_is_enabled())
944 return !zone_intersects(default_zone, start_pfn, nr_pages);
948 * Associates the given pfn range with the given node and the zone appropriate
949 * for the given online type.
951 static struct zone * __meminit move_pfn_range(int online_type, int nid,
952 unsigned long start_pfn, unsigned long nr_pages)
954 struct pglist_data *pgdat = NODE_DATA(nid);
955 struct zone *zone = default_zone_for_pfn(nid, start_pfn, nr_pages);
957 if (online_type == MMOP_ONLINE_KEEP) {
958 struct zone *movable_zone = &pgdat->node_zones[ZONE_MOVABLE];
960 * MMOP_ONLINE_KEEP defaults to MMOP_ONLINE_KERNEL but use
961 * movable zone if that is not possible (e.g. we are within
962 * or past the existing movable zone). movable_node overrides
963 * this default and defaults to movable zone
965 if (movable_pfn_range(nid, zone, start_pfn, nr_pages))
967 } else if (online_type == MMOP_ONLINE_MOVABLE) {
968 zone = &pgdat->node_zones[ZONE_MOVABLE];
971 move_pfn_range_to_zone(zone, start_pfn, nr_pages);
975 /* Must be protected by mem_hotplug_begin() */
976 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
979 unsigned long onlined_pages = 0;
981 int need_zonelists_rebuild = 0;
984 struct memory_notify arg;
986 nid = pfn_to_nid(pfn);
987 if (!allow_online_pfn_range(nid, pfn, nr_pages, online_type))
990 /* associate pfn range with the zone */
991 zone = move_pfn_range(online_type, nid, pfn, nr_pages);
994 arg.nr_pages = nr_pages;
995 node_states_check_changes_online(nr_pages, zone, &arg);
997 ret = memory_notify(MEM_GOING_ONLINE, &arg);
998 ret = notifier_to_errno(ret);
1000 goto failed_addition;
1003 * If this zone is not populated, then it is not in zonelist.
1004 * This means the page allocator ignores this zone.
1005 * So, zonelist must be updated after online.
1007 mutex_lock(&zonelists_mutex);
1008 if (!populated_zone(zone)) {
1009 need_zonelists_rebuild = 1;
1010 build_all_zonelists(NULL, zone);
1013 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
1014 online_pages_range);
1016 if (need_zonelists_rebuild)
1017 zone_pcp_reset(zone);
1018 mutex_unlock(&zonelists_mutex);
1019 goto failed_addition;
1022 zone->present_pages += onlined_pages;
1024 pgdat_resize_lock(zone->zone_pgdat, &flags);
1025 zone->zone_pgdat->node_present_pages += onlined_pages;
1026 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1028 if (onlined_pages) {
1029 node_states_set_node(nid, &arg);
1030 if (need_zonelists_rebuild)
1031 build_all_zonelists(NULL, NULL);
1033 zone_pcp_update(zone);
1036 mutex_unlock(&zonelists_mutex);
1038 init_per_zone_wmark_min();
1040 if (onlined_pages) {
1045 vm_total_pages = nr_free_pagecache_pages();
1047 writeback_set_ratelimit();
1050 memory_notify(MEM_ONLINE, &arg);
1054 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
1055 (unsigned long long) pfn << PAGE_SHIFT,
1056 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
1057 memory_notify(MEM_CANCEL_ONLINE, &arg);
1060 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1062 static void reset_node_present_pages(pg_data_t *pgdat)
1066 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
1067 z->present_pages = 0;
1069 pgdat->node_present_pages = 0;
1072 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1073 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1075 struct pglist_data *pgdat;
1076 unsigned long zones_size[MAX_NR_ZONES] = {0};
1077 unsigned long zholes_size[MAX_NR_ZONES] = {0};
1078 unsigned long start_pfn = PFN_DOWN(start);
1080 pgdat = NODE_DATA(nid);
1082 pgdat = arch_alloc_nodedata(nid);
1086 arch_refresh_nodedata(nid, pgdat);
1089 * Reset the nr_zones, order and classzone_idx before reuse.
1090 * Note that kswapd will init kswapd_classzone_idx properly
1091 * when it starts in the near future.
1093 pgdat->nr_zones = 0;
1094 pgdat->kswapd_order = 0;
1095 pgdat->kswapd_classzone_idx = 0;
1098 /* we can use NODE_DATA(nid) from here */
1100 /* init node's zones as empty zones, we don't have any present pages.*/
1101 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1102 pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
1105 * The node we allocated has no zone fallback lists. For avoiding
1106 * to access not-initialized zonelist, build here.
1108 mutex_lock(&zonelists_mutex);
1109 build_all_zonelists(pgdat, NULL);
1110 mutex_unlock(&zonelists_mutex);
1113 * zone->managed_pages is set to an approximate value in
1114 * free_area_init_core(), which will cause
1115 * /sys/device/system/node/nodeX/meminfo has wrong data.
1116 * So reset it to 0 before any memory is onlined.
1118 reset_node_managed_pages(pgdat);
1121 * When memory is hot-added, all the memory is in offline state. So
1122 * clear all zones' present_pages because they will be updated in
1123 * online_pages() and offline_pages().
1125 reset_node_present_pages(pgdat);
1130 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1132 arch_refresh_nodedata(nid, NULL);
1133 free_percpu(pgdat->per_cpu_nodestats);
1134 arch_free_nodedata(pgdat);
1140 * try_online_node - online a node if offlined
1142 * called by cpu_up() to online a node without onlined memory.
1144 int try_online_node(int nid)
1149 if (node_online(nid))
1152 mem_hotplug_begin();
1153 pgdat = hotadd_new_pgdat(nid, 0);
1155 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1159 node_set_online(nid);
1160 ret = register_one_node(nid);
1163 if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
1164 mutex_lock(&zonelists_mutex);
1165 build_all_zonelists(NULL, NULL);
1166 mutex_unlock(&zonelists_mutex);
1174 static int check_hotplug_memory_range(u64 start, u64 size)
1176 u64 start_pfn = PFN_DOWN(start);
1177 u64 nr_pages = size >> PAGE_SHIFT;
1179 /* Memory range must be aligned with section */
1180 if ((start_pfn & ~PAGE_SECTION_MASK) ||
1181 (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1182 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1183 (unsigned long long)start,
1184 (unsigned long long)size);
1191 static int online_memory_block(struct memory_block *mem, void *arg)
1193 return device_online(&mem->dev);
1196 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1197 int __ref add_memory_resource(int nid, struct resource *res, bool online)
1200 pg_data_t *pgdat = NULL;
1206 size = resource_size(res);
1208 ret = check_hotplug_memory_range(start, size);
1212 { /* Stupid hack to suppress address-never-null warning */
1213 void *p = NODE_DATA(nid);
1217 mem_hotplug_begin();
1220 * Add new range to memblock so that when hotadd_new_pgdat() is called
1221 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1222 * this new range and calculate total pages correctly. The range will
1223 * be removed at hot-remove time.
1225 memblock_add_node(start, size, nid);
1227 new_node = !node_online(nid);
1229 pgdat = hotadd_new_pgdat(nid, start);
1235 /* call arch's memory hotadd */
1236 ret = arch_add_memory(nid, start, size, true);
1241 /* we online node here. we can't roll back from here. */
1242 node_set_online(nid);
1245 unsigned long start_pfn = start >> PAGE_SHIFT;
1246 unsigned long nr_pages = size >> PAGE_SHIFT;
1248 ret = __register_one_node(nid);
1253 * link memory sections under this node. This is already
1254 * done when creatig memory section in register_new_memory
1255 * but that depends to have the node registered so offline
1256 * nodes have to go through register_node.
1257 * TODO clean up this mess.
1259 ret = link_mem_sections(nid, start_pfn, nr_pages);
1262 * If sysfs file of new node can't create, cpu on the node
1263 * can't be hot-added. There is no rollback way now.
1264 * So, check by BUG_ON() to catch it reluctantly..
1269 /* create new memmap entry */
1270 firmware_map_add_hotplug(start, start + size, "System RAM");
1272 /* online pages if requested */
1274 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1275 NULL, online_memory_block);
1280 /* rollback pgdat allocation and others */
1281 if (new_pgdat && pgdat)
1282 rollback_node_hotadd(nid, pgdat);
1283 memblock_remove(start, size);
1289 EXPORT_SYMBOL_GPL(add_memory_resource);
1291 int __ref add_memory(int nid, u64 start, u64 size)
1293 struct resource *res;
1296 res = register_memory_resource(start, size);
1298 return PTR_ERR(res);
1300 ret = add_memory_resource(nid, res, memhp_auto_online);
1302 release_memory_resource(res);
1305 EXPORT_SYMBOL_GPL(add_memory);
1307 #ifdef CONFIG_MEMORY_HOTREMOVE
1309 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1310 * set and the size of the free page is given by page_order(). Using this,
1311 * the function determines if the pageblock contains only free pages.
1312 * Due to buddy contraints, a free page at least the size of a pageblock will
1313 * be located at the start of the pageblock
1315 static inline int pageblock_free(struct page *page)
1317 return PageBuddy(page) && page_order(page) >= pageblock_order;
1320 /* Return the start of the next active pageblock after a given page */
1321 static struct page *next_active_pageblock(struct page *page)
1323 /* Ensure the starting page is pageblock-aligned */
1324 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1326 /* If the entire pageblock is free, move to the end of free page */
1327 if (pageblock_free(page)) {
1329 /* be careful. we don't have locks, page_order can be changed.*/
1330 order = page_order(page);
1331 if ((order < MAX_ORDER) && (order >= pageblock_order))
1332 return page + (1 << order);
1335 return page + pageblock_nr_pages;
1338 /* Checks if this range of memory is likely to be hot-removable. */
1339 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1341 struct page *page = pfn_to_page(start_pfn);
1342 struct page *end_page = page + nr_pages;
1344 /* Check the starting page of each pageblock within the range */
1345 for (; page < end_page; page = next_active_pageblock(page)) {
1346 if (!is_pageblock_removable_nolock(page))
1351 /* All pageblocks in the memory block are likely to be hot-removable */
1356 * Confirm all pages in a range [start, end) belong to the same zone.
1357 * When true, return its valid [start, end).
1359 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1360 unsigned long *valid_start, unsigned long *valid_end)
1362 unsigned long pfn, sec_end_pfn;
1363 unsigned long start, end;
1364 struct zone *zone = NULL;
1367 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1369 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1370 /* Make sure the memory section is present first */
1371 if (!present_section_nr(pfn_to_section_nr(pfn)))
1373 for (; pfn < sec_end_pfn && pfn < end_pfn;
1374 pfn += MAX_ORDER_NR_PAGES) {
1376 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1377 while ((i < MAX_ORDER_NR_PAGES) &&
1378 !pfn_valid_within(pfn + i))
1380 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1382 page = pfn_to_page(pfn + i);
1383 if (zone && page_zone(page) != zone)
1387 zone = page_zone(page);
1388 end = pfn + MAX_ORDER_NR_PAGES;
1393 *valid_start = start;
1394 *valid_end = min(end, end_pfn);
1402 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1403 * non-lru movable pages and hugepages). We scan pfn because it's much
1404 * easier than scanning over linked list. This function returns the pfn
1405 * of the first found movable page if it's found, otherwise 0.
1407 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1411 for (pfn = start; pfn < end; pfn++) {
1412 if (pfn_valid(pfn)) {
1413 page = pfn_to_page(pfn);
1416 if (__PageMovable(page))
1418 if (PageHuge(page)) {
1419 if (page_huge_active(page))
1422 pfn = round_up(pfn + 1,
1423 1 << compound_order(page)) - 1;
1430 static struct page *new_node_page(struct page *page, unsigned long private,
1433 int nid = page_to_nid(page);
1434 nodemask_t nmask = node_states[N_MEMORY];
1437 * try to allocate from a different node but reuse this node if there
1438 * are no other online nodes to be used (e.g. we are offlining a part
1439 * of the only existing node)
1441 node_clear(nid, nmask);
1442 if (nodes_empty(nmask))
1443 node_set(nid, nmask);
1445 return new_page_nodemask(page, nid, &nmask);
1448 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1450 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1454 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1455 int not_managed = 0;
1459 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1460 if (!pfn_valid(pfn))
1462 page = pfn_to_page(pfn);
1464 if (PageHuge(page)) {
1465 struct page *head = compound_head(page);
1466 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1467 if (compound_order(head) > PFN_SECTION_SHIFT) {
1471 if (isolate_huge_page(page, &source))
1472 move_pages -= 1 << compound_order(head);
1476 if (!get_page_unless_zero(page))
1479 * We can skip free pages. And we can deal with pages on
1480 * LRU and non-lru movable pages.
1483 ret = isolate_lru_page(page);
1485 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1486 if (!ret) { /* Success */
1488 list_add_tail(&page->lru, &source);
1490 if (!__PageMovable(page))
1491 inc_node_page_state(page, NR_ISOLATED_ANON +
1492 page_is_file_cache(page));
1495 #ifdef CONFIG_DEBUG_VM
1496 pr_alert("failed to isolate pfn %lx\n", pfn);
1497 dump_page(page, "isolation failed");
1500 /* Because we don't have big zone->lock. we should
1501 check this again here. */
1502 if (page_count(page)) {
1509 if (!list_empty(&source)) {
1511 putback_movable_pages(&source);
1515 /* Allocate a new page from the nearest neighbor node */
1516 ret = migrate_pages(&source, new_node_page, NULL, 0,
1517 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1519 putback_movable_pages(&source);
1526 * remove from free_area[] and mark all as Reserved.
1529 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1532 __offline_isolated_pages(start, start + nr_pages);
1537 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1539 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1540 offline_isolated_pages_cb);
1544 * Check all pages in range, recoreded as memory resource, are isolated.
1547 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1551 long offlined = *(long *)data;
1552 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1553 offlined = nr_pages;
1555 *(long *)data += offlined;
1560 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1565 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1566 check_pages_isolated_cb);
1568 offlined = (long)ret;
1572 static int __init cmdline_parse_movable_node(char *p)
1574 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1575 movable_node_enabled = true;
1577 pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1581 early_param("movable_node", cmdline_parse_movable_node);
1583 /* check which state of node_states will be changed when offline memory */
1584 static void node_states_check_changes_offline(unsigned long nr_pages,
1585 struct zone *zone, struct memory_notify *arg)
1587 struct pglist_data *pgdat = zone->zone_pgdat;
1588 unsigned long present_pages = 0;
1589 enum zone_type zt, zone_last = ZONE_NORMAL;
1592 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1593 * contains nodes which have zones of 0...ZONE_NORMAL,
1594 * set zone_last to ZONE_NORMAL.
1596 * If we don't have HIGHMEM nor movable node,
1597 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1598 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1600 if (N_MEMORY == N_NORMAL_MEMORY)
1601 zone_last = ZONE_MOVABLE;
1604 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1605 * If the memory to be offline is in a zone of 0...zone_last,
1606 * and it is the last present memory, 0...zone_last will
1607 * become empty after offline , thus we can determind we will
1608 * need to clear the node from node_states[N_NORMAL_MEMORY].
1610 for (zt = 0; zt <= zone_last; zt++)
1611 present_pages += pgdat->node_zones[zt].present_pages;
1612 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1613 arg->status_change_nid_normal = zone_to_nid(zone);
1615 arg->status_change_nid_normal = -1;
1617 #ifdef CONFIG_HIGHMEM
1619 * If we have movable node, node_states[N_HIGH_MEMORY]
1620 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1621 * set zone_last to ZONE_HIGHMEM.
1623 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1624 * contains nodes which have zones of 0...ZONE_MOVABLE,
1625 * set zone_last to ZONE_MOVABLE.
1627 zone_last = ZONE_HIGHMEM;
1628 if (N_MEMORY == N_HIGH_MEMORY)
1629 zone_last = ZONE_MOVABLE;
1631 for (; zt <= zone_last; zt++)
1632 present_pages += pgdat->node_zones[zt].present_pages;
1633 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1634 arg->status_change_nid_high = zone_to_nid(zone);
1636 arg->status_change_nid_high = -1;
1638 arg->status_change_nid_high = arg->status_change_nid_normal;
1642 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1644 zone_last = ZONE_MOVABLE;
1647 * check whether node_states[N_HIGH_MEMORY] will be changed
1648 * If we try to offline the last present @nr_pages from the node,
1649 * we can determind we will need to clear the node from
1650 * node_states[N_HIGH_MEMORY].
1652 for (; zt <= zone_last; zt++)
1653 present_pages += pgdat->node_zones[zt].present_pages;
1654 if (nr_pages >= present_pages)
1655 arg->status_change_nid = zone_to_nid(zone);
1657 arg->status_change_nid = -1;
1660 static void node_states_clear_node(int node, struct memory_notify *arg)
1662 if (arg->status_change_nid_normal >= 0)
1663 node_clear_state(node, N_NORMAL_MEMORY);
1665 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1666 (arg->status_change_nid_high >= 0))
1667 node_clear_state(node, N_HIGH_MEMORY);
1669 if ((N_MEMORY != N_HIGH_MEMORY) &&
1670 (arg->status_change_nid >= 0))
1671 node_clear_state(node, N_MEMORY);
1674 static int __ref __offline_pages(unsigned long start_pfn,
1675 unsigned long end_pfn, unsigned long timeout)
1677 unsigned long pfn, nr_pages, expire;
1678 long offlined_pages;
1679 int ret, drain, retry_max, node;
1680 unsigned long flags;
1681 unsigned long valid_start, valid_end;
1683 struct memory_notify arg;
1685 /* at least, alignment against pageblock is necessary */
1686 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1688 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1690 /* This makes hotplug much easier...and readable.
1691 we assume this for now. .*/
1692 if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, &valid_end))
1695 zone = page_zone(pfn_to_page(valid_start));
1696 node = zone_to_nid(zone);
1697 nr_pages = end_pfn - start_pfn;
1699 /* set above range as isolated */
1700 ret = start_isolate_page_range(start_pfn, end_pfn,
1701 MIGRATE_MOVABLE, true);
1705 arg.start_pfn = start_pfn;
1706 arg.nr_pages = nr_pages;
1707 node_states_check_changes_offline(nr_pages, zone, &arg);
1709 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1710 ret = notifier_to_errno(ret);
1712 goto failed_removal;
1715 expire = jiffies + timeout;
1719 /* start memory hot removal */
1721 if (time_after(jiffies, expire))
1722 goto failed_removal;
1724 if (signal_pending(current))
1725 goto failed_removal;
1728 lru_add_drain_all();
1730 drain_all_pages(zone);
1733 pfn = scan_movable_pages(start_pfn, end_pfn);
1734 if (pfn) { /* We have movable pages */
1735 ret = do_migrate_range(pfn, end_pfn);
1741 if (--retry_max == 0)
1742 goto failed_removal;
1748 /* drain all zone's lru pagevec, this is asynchronous... */
1749 lru_add_drain_all();
1751 /* drain pcp pages, this is synchronous. */
1752 drain_all_pages(zone);
1754 * dissolve free hugepages in the memory block before doing offlining
1755 * actually in order to make hugetlbfs's object counting consistent.
1757 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1759 goto failed_removal;
1761 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1762 if (offlined_pages < 0) {
1764 goto failed_removal;
1766 pr_info("Offlined Pages %ld\n", offlined_pages);
1767 /* Ok, all of our target is isolated.
1768 We cannot do rollback at this point. */
1769 offline_isolated_pages(start_pfn, end_pfn);
1770 /* reset pagetype flags and makes migrate type to be MOVABLE */
1771 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1772 /* removal success */
1773 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1774 zone->present_pages -= offlined_pages;
1776 pgdat_resize_lock(zone->zone_pgdat, &flags);
1777 zone->zone_pgdat->node_present_pages -= offlined_pages;
1778 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1780 init_per_zone_wmark_min();
1782 if (!populated_zone(zone)) {
1783 zone_pcp_reset(zone);
1784 mutex_lock(&zonelists_mutex);
1785 build_all_zonelists(NULL, NULL);
1786 mutex_unlock(&zonelists_mutex);
1788 zone_pcp_update(zone);
1790 node_states_clear_node(node, &arg);
1791 if (arg.status_change_nid >= 0) {
1793 kcompactd_stop(node);
1796 vm_total_pages = nr_free_pagecache_pages();
1797 writeback_set_ratelimit();
1799 memory_notify(MEM_OFFLINE, &arg);
1803 pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n",
1804 (unsigned long long) start_pfn << PAGE_SHIFT,
1805 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1806 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1807 /* pushback to free area */
1808 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1812 /* Must be protected by mem_hotplug_begin() */
1813 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1815 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1817 #endif /* CONFIG_MEMORY_HOTREMOVE */
1820 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1821 * @start_pfn: start pfn of the memory range
1822 * @end_pfn: end pfn of the memory range
1823 * @arg: argument passed to func
1824 * @func: callback for each memory section walked
1826 * This function walks through all present mem sections in range
1827 * [start_pfn, end_pfn) and call func on each mem section.
1829 * Returns the return value of func.
1831 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1832 void *arg, int (*func)(struct memory_block *, void *))
1834 struct memory_block *mem = NULL;
1835 struct mem_section *section;
1836 unsigned long pfn, section_nr;
1839 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1840 section_nr = pfn_to_section_nr(pfn);
1841 if (!present_section_nr(section_nr))
1844 section = __nr_to_section(section_nr);
1845 /* same memblock? */
1847 if ((section_nr >= mem->start_section_nr) &&
1848 (section_nr <= mem->end_section_nr))
1851 mem = find_memory_block_hinted(section, mem);
1855 ret = func(mem, arg);
1857 kobject_put(&mem->dev.kobj);
1863 kobject_put(&mem->dev.kobj);
1868 #ifdef CONFIG_MEMORY_HOTREMOVE
1869 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1871 int ret = !is_memblock_offlined(mem);
1873 if (unlikely(ret)) {
1874 phys_addr_t beginpa, endpa;
1876 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1877 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1878 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1885 static int check_cpu_on_node(pg_data_t *pgdat)
1889 for_each_present_cpu(cpu) {
1890 if (cpu_to_node(cpu) == pgdat->node_id)
1892 * the cpu on this node isn't removed, and we can't
1893 * offline this node.
1901 static void unmap_cpu_on_node(pg_data_t *pgdat)
1903 #ifdef CONFIG_ACPI_NUMA
1906 for_each_possible_cpu(cpu)
1907 if (cpu_to_node(cpu) == pgdat->node_id)
1908 numa_clear_node(cpu);
1912 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
1916 ret = check_cpu_on_node(pgdat);
1921 * the node will be offlined when we come here, so we can clear
1922 * the cpu_to_node() now.
1925 unmap_cpu_on_node(pgdat);
1932 * Offline a node if all memory sections and cpus of the node are removed.
1934 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1935 * and online/offline operations before this call.
1937 void try_offline_node(int nid)
1939 pg_data_t *pgdat = NODE_DATA(nid);
1940 unsigned long start_pfn = pgdat->node_start_pfn;
1941 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1944 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1945 unsigned long section_nr = pfn_to_section_nr(pfn);
1947 if (!present_section_nr(section_nr))
1950 if (pfn_to_nid(pfn) != nid)
1954 * some memory sections of this node are not removed, and we
1955 * can't offline node now.
1960 if (check_and_unmap_cpu_on_node(pgdat))
1964 * all memory/cpu of this node are removed, we can offline this
1967 node_set_offline(nid);
1968 unregister_one_node(nid);
1970 EXPORT_SYMBOL(try_offline_node);
1975 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1976 * and online/offline operations before this call, as required by
1977 * try_offline_node().
1979 void __ref remove_memory(int nid, u64 start, u64 size)
1983 BUG_ON(check_hotplug_memory_range(start, size));
1985 mem_hotplug_begin();
1988 * All memory blocks must be offlined before removing memory. Check
1989 * whether all memory blocks in question are offline and trigger a BUG()
1990 * if this is not the case.
1992 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
1993 check_memblock_offlined_cb);
1997 /* remove memmap entry */
1998 firmware_map_remove(start, start + size, "System RAM");
1999 memblock_free(start, size);
2000 memblock_remove(start, size);
2002 arch_remove_memory(start, size);
2004 try_offline_node(nid);
2008 EXPORT_SYMBOL_GPL(remove_memory);
2009 #endif /* CONFIG_MEMORY_HOTREMOVE */