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>
37 #include <asm/tlbflush.h>
42 * online_page_callback contains pointer to current page onlining function.
43 * Initially it is generic_online_page(). If it is required it could be
44 * changed by calling set_online_page_callback() for callback registration
45 * and restore_online_page_callback() for generic callback restore.
48 static void generic_online_page(struct page *page);
50 static online_page_callback_t online_page_callback = generic_online_page;
51 static DEFINE_MUTEX(online_page_callback_lock);
53 /* The same as the cpu_hotplug lock, but for memory hotplug. */
55 struct task_struct *active_writer;
56 struct mutex lock; /* Synchronizes accesses to refcount, */
58 * Also blocks the new readers during
59 * an ongoing mem hotplug operation.
63 #ifdef CONFIG_DEBUG_LOCK_ALLOC
64 struct lockdep_map dep_map;
67 .active_writer = NULL,
68 .lock = __MUTEX_INITIALIZER(mem_hotplug.lock),
70 #ifdef CONFIG_DEBUG_LOCK_ALLOC
71 .dep_map = {.name = "mem_hotplug.lock" },
75 /* Lockdep annotations for get/put_online_mems() and mem_hotplug_begin/end() */
76 #define memhp_lock_acquire_read() lock_map_acquire_read(&mem_hotplug.dep_map)
77 #define memhp_lock_acquire() lock_map_acquire(&mem_hotplug.dep_map)
78 #define memhp_lock_release() lock_map_release(&mem_hotplug.dep_map)
80 void get_online_mems(void)
83 if (mem_hotplug.active_writer == current)
85 memhp_lock_acquire_read();
86 mutex_lock(&mem_hotplug.lock);
87 mem_hotplug.refcount++;
88 mutex_unlock(&mem_hotplug.lock);
92 void put_online_mems(void)
94 if (mem_hotplug.active_writer == current)
96 mutex_lock(&mem_hotplug.lock);
98 if (WARN_ON(!mem_hotplug.refcount))
99 mem_hotplug.refcount++; /* try to fix things up */
101 if (!--mem_hotplug.refcount && unlikely(mem_hotplug.active_writer))
102 wake_up_process(mem_hotplug.active_writer);
103 mutex_unlock(&mem_hotplug.lock);
104 memhp_lock_release();
108 void mem_hotplug_begin(void)
110 mem_hotplug.active_writer = current;
112 memhp_lock_acquire();
114 mutex_lock(&mem_hotplug.lock);
115 if (likely(!mem_hotplug.refcount))
117 __set_current_state(TASK_UNINTERRUPTIBLE);
118 mutex_unlock(&mem_hotplug.lock);
123 void mem_hotplug_done(void)
125 mem_hotplug.active_writer = NULL;
126 mutex_unlock(&mem_hotplug.lock);
127 memhp_lock_release();
130 /* add this memory to iomem resource */
131 static struct resource *register_memory_resource(u64 start, u64 size)
133 struct resource *res;
134 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
136 return ERR_PTR(-ENOMEM);
138 res->name = "System RAM";
140 res->end = start + size - 1;
141 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
142 if (request_resource(&iomem_resource, res) < 0) {
143 pr_debug("System RAM resource %pR cannot be added\n", res);
145 return ERR_PTR(-EEXIST);
150 static void release_memory_resource(struct resource *res)
154 release_resource(res);
159 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
160 void get_page_bootmem(unsigned long info, struct page *page,
163 page->lru.next = (struct list_head *) type;
164 SetPagePrivate(page);
165 set_page_private(page, info);
166 atomic_inc(&page->_count);
169 void put_page_bootmem(struct page *page)
173 type = (unsigned long) page->lru.next;
174 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
175 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
177 if (atomic_dec_return(&page->_count) == 1) {
178 ClearPagePrivate(page);
179 set_page_private(page, 0);
180 INIT_LIST_HEAD(&page->lru);
181 free_reserved_page(page);
185 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
186 #ifndef CONFIG_SPARSEMEM_VMEMMAP
187 static void register_page_bootmem_info_section(unsigned long start_pfn)
189 unsigned long *usemap, mapsize, section_nr, i;
190 struct mem_section *ms;
191 struct page *page, *memmap;
193 section_nr = pfn_to_section_nr(start_pfn);
194 ms = __nr_to_section(section_nr);
196 /* Get section's memmap address */
197 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
200 * Get page for the memmap's phys address
201 * XXX: need more consideration for sparse_vmemmap...
203 page = virt_to_page(memmap);
204 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
205 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
207 /* remember memmap's page */
208 for (i = 0; i < mapsize; i++, page++)
209 get_page_bootmem(section_nr, page, SECTION_INFO);
211 usemap = __nr_to_section(section_nr)->pageblock_flags;
212 page = virt_to_page(usemap);
214 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
216 for (i = 0; i < mapsize; i++, page++)
217 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
220 #else /* CONFIG_SPARSEMEM_VMEMMAP */
221 static void register_page_bootmem_info_section(unsigned long start_pfn)
223 unsigned long *usemap, mapsize, section_nr, i;
224 struct mem_section *ms;
225 struct page *page, *memmap;
227 if (!pfn_valid(start_pfn))
230 section_nr = pfn_to_section_nr(start_pfn);
231 ms = __nr_to_section(section_nr);
233 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
235 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
237 usemap = __nr_to_section(section_nr)->pageblock_flags;
238 page = virt_to_page(usemap);
240 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
242 for (i = 0; i < mapsize; i++, page++)
243 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
245 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
247 void register_page_bootmem_info_node(struct pglist_data *pgdat)
249 unsigned long i, pfn, end_pfn, nr_pages;
250 int node = pgdat->node_id;
254 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
255 page = virt_to_page(pgdat);
257 for (i = 0; i < nr_pages; i++, page++)
258 get_page_bootmem(node, page, NODE_INFO);
260 zone = &pgdat->node_zones[0];
261 for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
262 if (zone_is_initialized(zone)) {
263 nr_pages = zone->wait_table_hash_nr_entries
264 * sizeof(wait_queue_head_t);
265 nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
266 page = virt_to_page(zone->wait_table);
268 for (i = 0; i < nr_pages; i++, page++)
269 get_page_bootmem(node, page, NODE_INFO);
273 pfn = pgdat->node_start_pfn;
274 end_pfn = pgdat_end_pfn(pgdat);
276 /* register section info */
277 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
279 * Some platforms can assign the same pfn to multiple nodes - on
280 * node0 as well as nodeN. To avoid registering a pfn against
281 * multiple nodes we check that this pfn does not already
282 * reside in some other nodes.
284 if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node))
285 register_page_bootmem_info_section(pfn);
288 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
290 static void __meminit grow_zone_span(struct zone *zone, unsigned long start_pfn,
291 unsigned long end_pfn)
293 unsigned long old_zone_end_pfn;
295 zone_span_writelock(zone);
297 old_zone_end_pfn = zone_end_pfn(zone);
298 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
299 zone->zone_start_pfn = start_pfn;
301 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
302 zone->zone_start_pfn;
304 zone_span_writeunlock(zone);
307 static void resize_zone(struct zone *zone, unsigned long start_pfn,
308 unsigned long end_pfn)
310 zone_span_writelock(zone);
312 if (end_pfn - start_pfn) {
313 zone->zone_start_pfn = start_pfn;
314 zone->spanned_pages = end_pfn - start_pfn;
317 * make it consist as free_area_init_core(),
318 * if spanned_pages = 0, then keep start_pfn = 0
320 zone->zone_start_pfn = 0;
321 zone->spanned_pages = 0;
324 zone_span_writeunlock(zone);
327 static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
328 unsigned long end_pfn)
330 enum zone_type zid = zone_idx(zone);
331 int nid = zone->zone_pgdat->node_id;
334 for (pfn = start_pfn; pfn < end_pfn; pfn++)
335 set_page_links(pfn_to_page(pfn), zid, nid, pfn);
338 /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
339 * alloc_bootmem_node_nopanic()/memblock_virt_alloc_node_nopanic() */
340 static int __ref ensure_zone_is_initialized(struct zone *zone,
341 unsigned long start_pfn, unsigned long num_pages)
343 if (!zone_is_initialized(zone))
344 return init_currently_empty_zone(zone, start_pfn, num_pages);
349 static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
350 unsigned long start_pfn, unsigned long end_pfn)
354 unsigned long z1_start_pfn;
356 ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn);
360 pgdat_resize_lock(z1->zone_pgdat, &flags);
362 /* can't move pfns which are higher than @z2 */
363 if (end_pfn > zone_end_pfn(z2))
365 /* the move out part must be at the left most of @z2 */
366 if (start_pfn > z2->zone_start_pfn)
368 /* must included/overlap */
369 if (end_pfn <= z2->zone_start_pfn)
372 /* use start_pfn for z1's start_pfn if z1 is empty */
373 if (!zone_is_empty(z1))
374 z1_start_pfn = z1->zone_start_pfn;
376 z1_start_pfn = start_pfn;
378 resize_zone(z1, z1_start_pfn, end_pfn);
379 resize_zone(z2, end_pfn, zone_end_pfn(z2));
381 pgdat_resize_unlock(z1->zone_pgdat, &flags);
383 fix_zone_id(z1, start_pfn, end_pfn);
387 pgdat_resize_unlock(z1->zone_pgdat, &flags);
391 static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
392 unsigned long start_pfn, unsigned long end_pfn)
396 unsigned long z2_end_pfn;
398 ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn);
402 pgdat_resize_lock(z1->zone_pgdat, &flags);
404 /* can't move pfns which are lower than @z1 */
405 if (z1->zone_start_pfn > start_pfn)
407 /* the move out part mast at the right most of @z1 */
408 if (zone_end_pfn(z1) > end_pfn)
410 /* must included/overlap */
411 if (start_pfn >= zone_end_pfn(z1))
414 /* use end_pfn for z2's end_pfn if z2 is empty */
415 if (!zone_is_empty(z2))
416 z2_end_pfn = zone_end_pfn(z2);
418 z2_end_pfn = end_pfn;
420 resize_zone(z1, z1->zone_start_pfn, start_pfn);
421 resize_zone(z2, start_pfn, z2_end_pfn);
423 pgdat_resize_unlock(z1->zone_pgdat, &flags);
425 fix_zone_id(z2, start_pfn, end_pfn);
429 pgdat_resize_unlock(z1->zone_pgdat, &flags);
433 static void __meminit grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
434 unsigned long end_pfn)
436 unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat);
438 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
439 pgdat->node_start_pfn = start_pfn;
441 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
442 pgdat->node_start_pfn;
445 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
447 struct pglist_data *pgdat = zone->zone_pgdat;
448 int nr_pages = PAGES_PER_SECTION;
449 int nid = pgdat->node_id;
451 unsigned long flags, pfn;
454 zone_type = zone - pgdat->node_zones;
455 ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
459 pgdat_resize_lock(zone->zone_pgdat, &flags);
460 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
461 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
462 phys_start_pfn + nr_pages);
463 pgdat_resize_unlock(zone->zone_pgdat, &flags);
464 memmap_init_zone(nr_pages, nid, zone_type,
465 phys_start_pfn, MEMMAP_HOTPLUG);
467 /* online_page_range is called later and expects pages reserved */
468 for (pfn = phys_start_pfn; pfn < phys_start_pfn + nr_pages; pfn++) {
472 SetPageReserved(pfn_to_page(pfn));
477 static int __meminit __add_section(int nid, struct zone *zone,
478 unsigned long phys_start_pfn)
482 if (pfn_valid(phys_start_pfn))
485 ret = sparse_add_one_section(zone, phys_start_pfn);
490 ret = __add_zone(zone, phys_start_pfn);
495 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
499 * Reasonably generic function for adding memory. It is
500 * expected that archs that support memory hotplug will
501 * call this function after deciding the zone to which to
504 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
505 unsigned long nr_pages)
509 int start_sec, end_sec;
510 struct vmem_altmap *altmap;
512 /* during initialize mem_map, align hot-added range to section */
513 start_sec = pfn_to_section_nr(phys_start_pfn);
514 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
516 altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn));
519 * Validate altmap is within bounds of the total request
521 if (altmap->base_pfn != phys_start_pfn
522 || vmem_altmap_offset(altmap) > nr_pages) {
523 pr_warn_once("memory add fail, invalid altmap\n");
529 for (i = start_sec; i <= end_sec; i++) {
530 err = __add_section(nid, zone, section_nr_to_pfn(i));
533 * EEXIST is finally dealt with by ioresource collision
534 * check. see add_memory() => register_memory_resource()
535 * Warning will be printed if there is collision.
537 if (err && (err != -EEXIST))
541 vmemmap_populate_print_last();
545 EXPORT_SYMBOL_GPL(__add_pages);
547 #ifdef CONFIG_MEMORY_HOTREMOVE
548 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
549 static int find_smallest_section_pfn(int nid, struct zone *zone,
550 unsigned long start_pfn,
551 unsigned long end_pfn)
553 struct mem_section *ms;
555 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
556 ms = __pfn_to_section(start_pfn);
558 if (unlikely(!valid_section(ms)))
561 if (unlikely(pfn_to_nid(start_pfn) != nid))
564 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
573 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
574 static int find_biggest_section_pfn(int nid, struct zone *zone,
575 unsigned long start_pfn,
576 unsigned long end_pfn)
578 struct mem_section *ms;
581 /* pfn is the end pfn of a memory section. */
583 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
584 ms = __pfn_to_section(pfn);
586 if (unlikely(!valid_section(ms)))
589 if (unlikely(pfn_to_nid(pfn) != nid))
592 if (zone && zone != page_zone(pfn_to_page(pfn)))
601 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
602 unsigned long end_pfn)
604 unsigned long zone_start_pfn = zone->zone_start_pfn;
605 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
606 unsigned long zone_end_pfn = z;
608 struct mem_section *ms;
609 int nid = zone_to_nid(zone);
611 zone_span_writelock(zone);
612 if (zone_start_pfn == start_pfn) {
614 * If the section is smallest section in the zone, it need
615 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
616 * In this case, we find second smallest valid mem_section
617 * for shrinking zone.
619 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
622 zone->zone_start_pfn = pfn;
623 zone->spanned_pages = zone_end_pfn - pfn;
625 } else if (zone_end_pfn == end_pfn) {
627 * If the section is biggest section in the zone, it need
628 * shrink zone->spanned_pages.
629 * In this case, we find second biggest valid mem_section for
632 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
635 zone->spanned_pages = pfn - zone_start_pfn + 1;
639 * The section is not biggest or smallest mem_section in the zone, it
640 * only creates a hole in the zone. So in this case, we need not
641 * change the zone. But perhaps, the zone has only hole data. Thus
642 * it check the zone has only hole or not.
644 pfn = zone_start_pfn;
645 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
646 ms = __pfn_to_section(pfn);
648 if (unlikely(!valid_section(ms)))
651 if (page_zone(pfn_to_page(pfn)) != zone)
654 /* If the section is current section, it continues the loop */
655 if (start_pfn == pfn)
658 /* If we find valid section, we have nothing to do */
659 zone_span_writeunlock(zone);
663 /* The zone has no valid section */
664 zone->zone_start_pfn = 0;
665 zone->spanned_pages = 0;
666 zone_span_writeunlock(zone);
669 static void shrink_pgdat_span(struct pglist_data *pgdat,
670 unsigned long start_pfn, unsigned long end_pfn)
672 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
673 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
674 unsigned long pgdat_end_pfn = p;
676 struct mem_section *ms;
677 int nid = pgdat->node_id;
679 if (pgdat_start_pfn == start_pfn) {
681 * If the section is smallest section in the pgdat, it need
682 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
683 * In this case, we find second smallest valid mem_section
684 * for shrinking zone.
686 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
689 pgdat->node_start_pfn = pfn;
690 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
692 } else if (pgdat_end_pfn == end_pfn) {
694 * If the section is biggest section in the pgdat, it need
695 * shrink pgdat->node_spanned_pages.
696 * In this case, we find second biggest valid mem_section for
699 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
702 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
706 * If the section is not biggest or smallest mem_section in the pgdat,
707 * it only creates a hole in the pgdat. So in this case, we need not
709 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
710 * has only hole or not.
712 pfn = pgdat_start_pfn;
713 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
714 ms = __pfn_to_section(pfn);
716 if (unlikely(!valid_section(ms)))
719 if (pfn_to_nid(pfn) != nid)
722 /* If the section is current section, it continues the loop */
723 if (start_pfn == pfn)
726 /* If we find valid section, we have nothing to do */
730 /* The pgdat has no valid section */
731 pgdat->node_start_pfn = 0;
732 pgdat->node_spanned_pages = 0;
735 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
737 struct pglist_data *pgdat = zone->zone_pgdat;
738 int nr_pages = PAGES_PER_SECTION;
742 zone_type = zone - pgdat->node_zones;
744 pgdat_resize_lock(zone->zone_pgdat, &flags);
745 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
746 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
747 pgdat_resize_unlock(zone->zone_pgdat, &flags);
750 static int __remove_section(struct zone *zone, struct mem_section *ms,
751 unsigned long map_offset)
753 unsigned long start_pfn;
757 if (!valid_section(ms))
760 ret = unregister_memory_section(ms);
764 scn_nr = __section_nr(ms);
765 start_pfn = section_nr_to_pfn(scn_nr);
766 __remove_zone(zone, start_pfn);
768 sparse_remove_one_section(zone, ms, map_offset);
773 * __remove_pages() - remove sections of pages from a zone
774 * @zone: zone from which pages need to be removed
775 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
776 * @nr_pages: number of pages to remove (must be multiple of section size)
778 * Generic helper function to remove section mappings and sysfs entries
779 * for the section of the memory we are removing. Caller needs to make
780 * sure that pages are marked reserved and zones are adjust properly by
781 * calling offline_pages().
783 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
784 unsigned long nr_pages)
787 unsigned long map_offset = 0;
788 int sections_to_remove, ret = 0;
790 /* In the ZONE_DEVICE case device driver owns the memory region */
791 if (is_dev_zone(zone)) {
792 struct page *page = pfn_to_page(phys_start_pfn);
793 struct vmem_altmap *altmap;
795 altmap = to_vmem_altmap((unsigned long) page);
797 map_offset = vmem_altmap_offset(altmap);
799 resource_size_t start, size;
801 start = phys_start_pfn << PAGE_SHIFT;
802 size = nr_pages * PAGE_SIZE;
804 ret = release_mem_region_adjustable(&iomem_resource, start,
807 resource_size_t endres = start + size - 1;
809 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
810 &start, &endres, ret);
815 * We can only remove entire sections
817 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
818 BUG_ON(nr_pages % PAGES_PER_SECTION);
820 sections_to_remove = nr_pages / PAGES_PER_SECTION;
821 for (i = 0; i < sections_to_remove; i++) {
822 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
824 ret = __remove_section(zone, __pfn_to_section(pfn), map_offset);
831 EXPORT_SYMBOL_GPL(__remove_pages);
832 #endif /* CONFIG_MEMORY_HOTREMOVE */
834 int set_online_page_callback(online_page_callback_t callback)
839 mutex_lock(&online_page_callback_lock);
841 if (online_page_callback == generic_online_page) {
842 online_page_callback = callback;
846 mutex_unlock(&online_page_callback_lock);
851 EXPORT_SYMBOL_GPL(set_online_page_callback);
853 int restore_online_page_callback(online_page_callback_t callback)
858 mutex_lock(&online_page_callback_lock);
860 if (online_page_callback == callback) {
861 online_page_callback = generic_online_page;
865 mutex_unlock(&online_page_callback_lock);
870 EXPORT_SYMBOL_GPL(restore_online_page_callback);
872 void __online_page_set_limits(struct page *page)
875 EXPORT_SYMBOL_GPL(__online_page_set_limits);
877 void __online_page_increment_counters(struct page *page)
879 adjust_managed_page_count(page, 1);
881 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
883 void __online_page_free(struct page *page)
885 __free_reserved_page(page);
887 EXPORT_SYMBOL_GPL(__online_page_free);
889 static void generic_online_page(struct page *page)
891 __online_page_set_limits(page);
892 __online_page_increment_counters(page);
893 __online_page_free(page);
896 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
900 unsigned long onlined_pages = *(unsigned long *)arg;
902 if (PageReserved(pfn_to_page(start_pfn)))
903 for (i = 0; i < nr_pages; i++) {
904 page = pfn_to_page(start_pfn + i);
905 (*online_page_callback)(page);
908 *(unsigned long *)arg = onlined_pages;
912 #ifdef CONFIG_MOVABLE_NODE
914 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
917 static bool can_online_high_movable(struct zone *zone)
921 #else /* CONFIG_MOVABLE_NODE */
922 /* ensure every online node has NORMAL memory */
923 static bool can_online_high_movable(struct zone *zone)
925 return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
927 #endif /* CONFIG_MOVABLE_NODE */
929 /* check which state of node_states will be changed when online memory */
930 static void node_states_check_changes_online(unsigned long nr_pages,
931 struct zone *zone, struct memory_notify *arg)
933 int nid = zone_to_nid(zone);
934 enum zone_type zone_last = ZONE_NORMAL;
937 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
938 * contains nodes which have zones of 0...ZONE_NORMAL,
939 * set zone_last to ZONE_NORMAL.
941 * If we don't have HIGHMEM nor movable node,
942 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
943 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
945 if (N_MEMORY == N_NORMAL_MEMORY)
946 zone_last = ZONE_MOVABLE;
949 * if the memory to be online is in a zone of 0...zone_last, and
950 * the zones of 0...zone_last don't have memory before online, we will
951 * need to set the node to node_states[N_NORMAL_MEMORY] after
952 * the memory is online.
954 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
955 arg->status_change_nid_normal = nid;
957 arg->status_change_nid_normal = -1;
959 #ifdef CONFIG_HIGHMEM
961 * If we have movable node, node_states[N_HIGH_MEMORY]
962 * contains nodes which have zones of 0...ZONE_HIGHMEM,
963 * set zone_last to ZONE_HIGHMEM.
965 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
966 * contains nodes which have zones of 0...ZONE_MOVABLE,
967 * set zone_last to ZONE_MOVABLE.
969 zone_last = ZONE_HIGHMEM;
970 if (N_MEMORY == N_HIGH_MEMORY)
971 zone_last = ZONE_MOVABLE;
973 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
974 arg->status_change_nid_high = nid;
976 arg->status_change_nid_high = -1;
978 arg->status_change_nid_high = arg->status_change_nid_normal;
982 * if the node don't have memory befor online, we will need to
983 * set the node to node_states[N_MEMORY] after the memory
986 if (!node_state(nid, N_MEMORY))
987 arg->status_change_nid = nid;
989 arg->status_change_nid = -1;
992 static void node_states_set_node(int node, struct memory_notify *arg)
994 if (arg->status_change_nid_normal >= 0)
995 node_set_state(node, N_NORMAL_MEMORY);
997 if (arg->status_change_nid_high >= 0)
998 node_set_state(node, N_HIGH_MEMORY);
1000 node_set_state(node, N_MEMORY);
1004 /* Must be protected by mem_hotplug_begin() */
1005 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
1007 unsigned long flags;
1008 unsigned long onlined_pages = 0;
1010 int need_zonelists_rebuild = 0;
1013 struct memory_notify arg;
1016 * This doesn't need a lock to do pfn_to_page().
1017 * The section can't be removed here because of the
1018 * memory_block->state_mutex.
1020 zone = page_zone(pfn_to_page(pfn));
1022 if ((zone_idx(zone) > ZONE_NORMAL ||
1023 online_type == MMOP_ONLINE_MOVABLE) &&
1024 !can_online_high_movable(zone))
1027 if (online_type == MMOP_ONLINE_KERNEL &&
1028 zone_idx(zone) == ZONE_MOVABLE) {
1029 if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages))
1032 if (online_type == MMOP_ONLINE_MOVABLE &&
1033 zone_idx(zone) == ZONE_MOVABLE - 1) {
1034 if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages))
1038 /* Previous code may changed the zone of the pfn range */
1039 zone = page_zone(pfn_to_page(pfn));
1041 arg.start_pfn = pfn;
1042 arg.nr_pages = nr_pages;
1043 node_states_check_changes_online(nr_pages, zone, &arg);
1045 nid = pfn_to_nid(pfn);
1047 ret = memory_notify(MEM_GOING_ONLINE, &arg);
1048 ret = notifier_to_errno(ret);
1050 memory_notify(MEM_CANCEL_ONLINE, &arg);
1054 * If this zone is not populated, then it is not in zonelist.
1055 * This means the page allocator ignores this zone.
1056 * So, zonelist must be updated after online.
1058 mutex_lock(&zonelists_mutex);
1059 if (!populated_zone(zone)) {
1060 need_zonelists_rebuild = 1;
1061 build_all_zonelists(NULL, zone);
1064 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
1065 online_pages_range);
1067 if (need_zonelists_rebuild)
1068 zone_pcp_reset(zone);
1069 mutex_unlock(&zonelists_mutex);
1070 printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n",
1071 (unsigned long long) pfn << PAGE_SHIFT,
1072 (((unsigned long long) pfn + nr_pages)
1073 << PAGE_SHIFT) - 1);
1074 memory_notify(MEM_CANCEL_ONLINE, &arg);
1078 zone->present_pages += onlined_pages;
1080 pgdat_resize_lock(zone->zone_pgdat, &flags);
1081 zone->zone_pgdat->node_present_pages += onlined_pages;
1082 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1084 if (onlined_pages) {
1085 node_states_set_node(zone_to_nid(zone), &arg);
1086 if (need_zonelists_rebuild)
1087 build_all_zonelists(NULL, NULL);
1089 zone_pcp_update(zone);
1092 mutex_unlock(&zonelists_mutex);
1094 init_per_zone_wmark_min();
1097 kswapd_run(zone_to_nid(zone));
1099 vm_total_pages = nr_free_pagecache_pages();
1101 writeback_set_ratelimit();
1104 memory_notify(MEM_ONLINE, &arg);
1107 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1109 static void reset_node_present_pages(pg_data_t *pgdat)
1113 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
1114 z->present_pages = 0;
1116 pgdat->node_present_pages = 0;
1119 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1120 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1122 struct pglist_data *pgdat;
1123 unsigned long zones_size[MAX_NR_ZONES] = {0};
1124 unsigned long zholes_size[MAX_NR_ZONES] = {0};
1125 unsigned long start_pfn = PFN_DOWN(start);
1127 pgdat = NODE_DATA(nid);
1129 pgdat = arch_alloc_nodedata(nid);
1133 arch_refresh_nodedata(nid, pgdat);
1135 /* Reset the nr_zones and classzone_idx to 0 before reuse */
1136 pgdat->nr_zones = 0;
1137 pgdat->classzone_idx = 0;
1140 /* we can use NODE_DATA(nid) from here */
1142 /* init node's zones as empty zones, we don't have any present pages.*/
1143 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1146 * The node we allocated has no zone fallback lists. For avoiding
1147 * to access not-initialized zonelist, build here.
1149 mutex_lock(&zonelists_mutex);
1150 build_all_zonelists(pgdat, NULL);
1151 mutex_unlock(&zonelists_mutex);
1154 * zone->managed_pages is set to an approximate value in
1155 * free_area_init_core(), which will cause
1156 * /sys/device/system/node/nodeX/meminfo has wrong data.
1157 * So reset it to 0 before any memory is onlined.
1159 reset_node_managed_pages(pgdat);
1162 * When memory is hot-added, all the memory is in offline state. So
1163 * clear all zones' present_pages because they will be updated in
1164 * online_pages() and offline_pages().
1166 reset_node_present_pages(pgdat);
1171 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1173 arch_refresh_nodedata(nid, NULL);
1174 arch_free_nodedata(pgdat);
1180 * try_online_node - online a node if offlined
1182 * called by cpu_up() to online a node without onlined memory.
1184 int try_online_node(int nid)
1189 if (node_online(nid))
1192 mem_hotplug_begin();
1193 pgdat = hotadd_new_pgdat(nid, 0);
1195 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1199 node_set_online(nid);
1200 ret = register_one_node(nid);
1203 if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
1204 mutex_lock(&zonelists_mutex);
1205 build_all_zonelists(NULL, NULL);
1206 mutex_unlock(&zonelists_mutex);
1214 static int check_hotplug_memory_range(u64 start, u64 size)
1216 u64 start_pfn = PFN_DOWN(start);
1217 u64 nr_pages = size >> PAGE_SHIFT;
1219 /* Memory range must be aligned with section */
1220 if ((start_pfn & ~PAGE_SECTION_MASK) ||
1221 (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1222 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1223 (unsigned long long)start,
1224 (unsigned long long)size);
1232 * If movable zone has already been setup, newly added memory should be check.
1233 * If its address is higher than movable zone, it should be added as movable.
1234 * Without this check, movable zone may overlap with other zone.
1236 static int should_add_memory_movable(int nid, u64 start, u64 size)
1238 unsigned long start_pfn = start >> PAGE_SHIFT;
1239 pg_data_t *pgdat = NODE_DATA(nid);
1240 struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE;
1242 if (zone_is_empty(movable_zone))
1245 if (movable_zone->zone_start_pfn <= start_pfn)
1251 int zone_for_memory(int nid, u64 start, u64 size, int zone_default,
1254 #ifdef CONFIG_ZONE_DEVICE
1258 if (should_add_memory_movable(nid, start, size))
1259 return ZONE_MOVABLE;
1261 return zone_default;
1264 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1265 int __ref add_memory_resource(int nid, struct resource *res)
1268 pg_data_t *pgdat = NULL;
1274 size = resource_size(res);
1276 ret = check_hotplug_memory_range(start, size);
1280 { /* Stupid hack to suppress address-never-null warning */
1281 void *p = NODE_DATA(nid);
1285 mem_hotplug_begin();
1288 * Add new range to memblock so that when hotadd_new_pgdat() is called
1289 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1290 * this new range and calculate total pages correctly. The range will
1291 * be removed at hot-remove time.
1293 memblock_add_node(start, size, nid);
1295 new_node = !node_online(nid);
1297 pgdat = hotadd_new_pgdat(nid, start);
1303 /* call arch's memory hotadd */
1304 ret = arch_add_memory(nid, start, size, false);
1309 /* we online node here. we can't roll back from here. */
1310 node_set_online(nid);
1313 ret = register_one_node(nid);
1315 * If sysfs file of new node can't create, cpu on the node
1316 * can't be hot-added. There is no rollback way now.
1317 * So, check by BUG_ON() to catch it reluctantly..
1322 /* create new memmap entry */
1323 firmware_map_add_hotplug(start, start + size, "System RAM");
1328 /* rollback pgdat allocation and others */
1330 rollback_node_hotadd(nid, pgdat);
1331 memblock_remove(start, size);
1337 EXPORT_SYMBOL_GPL(add_memory_resource);
1339 int __ref add_memory(int nid, u64 start, u64 size)
1341 struct resource *res;
1344 res = register_memory_resource(start, size);
1346 return PTR_ERR(res);
1348 ret = add_memory_resource(nid, res);
1350 release_memory_resource(res);
1353 EXPORT_SYMBOL_GPL(add_memory);
1355 #ifdef CONFIG_MEMORY_HOTREMOVE
1357 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1358 * set and the size of the free page is given by page_order(). Using this,
1359 * the function determines if the pageblock contains only free pages.
1360 * Due to buddy contraints, a free page at least the size of a pageblock will
1361 * be located at the start of the pageblock
1363 static inline int pageblock_free(struct page *page)
1365 return PageBuddy(page) && page_order(page) >= pageblock_order;
1368 /* Return the start of the next active pageblock after a given page */
1369 static struct page *next_active_pageblock(struct page *page)
1371 /* Ensure the starting page is pageblock-aligned */
1372 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1374 /* If the entire pageblock is free, move to the end of free page */
1375 if (pageblock_free(page)) {
1377 /* be careful. we don't have locks, page_order can be changed.*/
1378 order = page_order(page);
1379 if ((order < MAX_ORDER) && (order >= pageblock_order))
1380 return page + (1 << order);
1383 return page + pageblock_nr_pages;
1386 /* Checks if this range of memory is likely to be hot-removable. */
1387 int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1389 struct page *page = pfn_to_page(start_pfn);
1390 struct page *end_page = page + nr_pages;
1392 /* Check the starting page of each pageblock within the range */
1393 for (; page < end_page; page = next_active_pageblock(page)) {
1394 if (!is_pageblock_removable_nolock(page))
1399 /* All pageblocks in the memory block are likely to be hot-removable */
1404 * Confirm all pages in a range [start, end) is belongs to the same zone.
1406 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
1408 unsigned long pfn, sec_end_pfn;
1409 struct zone *zone = NULL;
1412 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn);
1414 pfn = sec_end_pfn + 1, sec_end_pfn += PAGES_PER_SECTION) {
1415 /* Make sure the memory section is present first */
1416 if (!present_section_nr(pfn_to_section_nr(pfn)))
1418 for (; pfn < sec_end_pfn && pfn < end_pfn;
1419 pfn += MAX_ORDER_NR_PAGES) {
1421 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1422 while ((i < MAX_ORDER_NR_PAGES) &&
1423 !pfn_valid_within(pfn + i))
1425 if (i == MAX_ORDER_NR_PAGES)
1427 page = pfn_to_page(pfn + i);
1428 if (zone && page_zone(page) != zone)
1430 zone = page_zone(page);
1437 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages
1438 * and hugepages). We scan pfn because it's much easier than scanning over
1439 * linked list. This function returns the pfn of the first found movable
1440 * page if it's found, otherwise 0.
1442 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1446 for (pfn = start; pfn < end; pfn++) {
1447 if (pfn_valid(pfn)) {
1448 page = pfn_to_page(pfn);
1451 if (PageHuge(page)) {
1452 if (page_huge_active(page))
1455 pfn = round_up(pfn + 1,
1456 1 << compound_order(page)) - 1;
1463 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1465 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1469 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1470 int not_managed = 0;
1474 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1475 if (!pfn_valid(pfn))
1477 page = pfn_to_page(pfn);
1479 if (PageHuge(page)) {
1480 struct page *head = compound_head(page);
1481 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1482 if (compound_order(head) > PFN_SECTION_SHIFT) {
1486 if (isolate_huge_page(page, &source))
1487 move_pages -= 1 << compound_order(head);
1491 if (!get_page_unless_zero(page))
1494 * We can skip free pages. And we can only deal with pages on
1497 ret = isolate_lru_page(page);
1498 if (!ret) { /* Success */
1500 list_add_tail(&page->lru, &source);
1502 inc_zone_page_state(page, NR_ISOLATED_ANON +
1503 page_is_file_cache(page));
1506 #ifdef CONFIG_DEBUG_VM
1507 printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
1509 dump_page(page, "failed to remove from LRU");
1512 /* Because we don't have big zone->lock. we should
1513 check this again here. */
1514 if (page_count(page)) {
1521 if (!list_empty(&source)) {
1523 putback_movable_pages(&source);
1528 * alloc_migrate_target should be improooooved!!
1529 * migrate_pages returns # of failed pages.
1531 ret = migrate_pages(&source, alloc_migrate_target, NULL, 0,
1532 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1534 putback_movable_pages(&source);
1541 * remove from free_area[] and mark all as Reserved.
1544 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1547 __offline_isolated_pages(start, start + nr_pages);
1552 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1554 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1555 offline_isolated_pages_cb);
1559 * Check all pages in range, recoreded as memory resource, are isolated.
1562 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1566 long offlined = *(long *)data;
1567 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1568 offlined = nr_pages;
1570 *(long *)data += offlined;
1575 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1580 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1581 check_pages_isolated_cb);
1583 offlined = (long)ret;
1587 #ifdef CONFIG_MOVABLE_NODE
1589 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1592 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1596 #else /* CONFIG_MOVABLE_NODE */
1597 /* ensure the node has NORMAL memory if it is still online */
1598 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1600 struct pglist_data *pgdat = zone->zone_pgdat;
1601 unsigned long present_pages = 0;
1604 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1605 present_pages += pgdat->node_zones[zt].present_pages;
1607 if (present_pages > nr_pages)
1611 for (; zt <= ZONE_MOVABLE; zt++)
1612 present_pages += pgdat->node_zones[zt].present_pages;
1615 * we can't offline the last normal memory until all
1616 * higher memory is offlined.
1618 return present_pages == 0;
1620 #endif /* CONFIG_MOVABLE_NODE */
1622 static int __init cmdline_parse_movable_node(char *p)
1624 #ifdef CONFIG_MOVABLE_NODE
1626 * Memory used by the kernel cannot be hot-removed because Linux
1627 * cannot migrate the kernel pages. When memory hotplug is
1628 * enabled, we should prevent memblock from allocating memory
1631 * ACPI SRAT records all hotpluggable memory ranges. But before
1632 * SRAT is parsed, we don't know about it.
1634 * The kernel image is loaded into memory at very early time. We
1635 * cannot prevent this anyway. So on NUMA system, we set any
1636 * node the kernel resides in as un-hotpluggable.
1638 * Since on modern servers, one node could have double-digit
1639 * gigabytes memory, we can assume the memory around the kernel
1640 * image is also un-hotpluggable. So before SRAT is parsed, just
1641 * allocate memory near the kernel image to try the best to keep
1642 * the kernel away from hotpluggable memory.
1644 memblock_set_bottom_up(true);
1645 movable_node_enabled = true;
1647 pr_warn("movable_node option not supported\n");
1651 early_param("movable_node", cmdline_parse_movable_node);
1653 /* check which state of node_states will be changed when offline memory */
1654 static void node_states_check_changes_offline(unsigned long nr_pages,
1655 struct zone *zone, struct memory_notify *arg)
1657 struct pglist_data *pgdat = zone->zone_pgdat;
1658 unsigned long present_pages = 0;
1659 enum zone_type zt, zone_last = ZONE_NORMAL;
1662 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1663 * contains nodes which have zones of 0...ZONE_NORMAL,
1664 * set zone_last to ZONE_NORMAL.
1666 * If we don't have HIGHMEM nor movable node,
1667 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1668 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1670 if (N_MEMORY == N_NORMAL_MEMORY)
1671 zone_last = ZONE_MOVABLE;
1674 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1675 * If the memory to be offline is in a zone of 0...zone_last,
1676 * and it is the last present memory, 0...zone_last will
1677 * become empty after offline , thus we can determind we will
1678 * need to clear the node from node_states[N_NORMAL_MEMORY].
1680 for (zt = 0; zt <= zone_last; zt++)
1681 present_pages += pgdat->node_zones[zt].present_pages;
1682 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1683 arg->status_change_nid_normal = zone_to_nid(zone);
1685 arg->status_change_nid_normal = -1;
1687 #ifdef CONFIG_HIGHMEM
1689 * If we have movable node, node_states[N_HIGH_MEMORY]
1690 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1691 * set zone_last to ZONE_HIGHMEM.
1693 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1694 * contains nodes which have zones of 0...ZONE_MOVABLE,
1695 * set zone_last to ZONE_MOVABLE.
1697 zone_last = ZONE_HIGHMEM;
1698 if (N_MEMORY == N_HIGH_MEMORY)
1699 zone_last = ZONE_MOVABLE;
1701 for (; zt <= zone_last; zt++)
1702 present_pages += pgdat->node_zones[zt].present_pages;
1703 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1704 arg->status_change_nid_high = zone_to_nid(zone);
1706 arg->status_change_nid_high = -1;
1708 arg->status_change_nid_high = arg->status_change_nid_normal;
1712 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1714 zone_last = ZONE_MOVABLE;
1717 * check whether node_states[N_HIGH_MEMORY] will be changed
1718 * If we try to offline the last present @nr_pages from the node,
1719 * we can determind we will need to clear the node from
1720 * node_states[N_HIGH_MEMORY].
1722 for (; zt <= zone_last; zt++)
1723 present_pages += pgdat->node_zones[zt].present_pages;
1724 if (nr_pages >= present_pages)
1725 arg->status_change_nid = zone_to_nid(zone);
1727 arg->status_change_nid = -1;
1730 static void node_states_clear_node(int node, struct memory_notify *arg)
1732 if (arg->status_change_nid_normal >= 0)
1733 node_clear_state(node, N_NORMAL_MEMORY);
1735 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1736 (arg->status_change_nid_high >= 0))
1737 node_clear_state(node, N_HIGH_MEMORY);
1739 if ((N_MEMORY != N_HIGH_MEMORY) &&
1740 (arg->status_change_nid >= 0))
1741 node_clear_state(node, N_MEMORY);
1744 static int __ref __offline_pages(unsigned long start_pfn,
1745 unsigned long end_pfn, unsigned long timeout)
1747 unsigned long pfn, nr_pages, expire;
1748 long offlined_pages;
1749 int ret, drain, retry_max, node;
1750 unsigned long flags;
1752 struct memory_notify arg;
1754 /* at least, alignment against pageblock is necessary */
1755 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1757 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1759 /* This makes hotplug much easier...and readable.
1760 we assume this for now. .*/
1761 if (!test_pages_in_a_zone(start_pfn, end_pfn))
1764 zone = page_zone(pfn_to_page(start_pfn));
1765 node = zone_to_nid(zone);
1766 nr_pages = end_pfn - start_pfn;
1768 if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1771 /* set above range as isolated */
1772 ret = start_isolate_page_range(start_pfn, end_pfn,
1773 MIGRATE_MOVABLE, true);
1777 arg.start_pfn = start_pfn;
1778 arg.nr_pages = nr_pages;
1779 node_states_check_changes_offline(nr_pages, zone, &arg);
1781 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1782 ret = notifier_to_errno(ret);
1784 goto failed_removal;
1787 expire = jiffies + timeout;
1791 /* start memory hot removal */
1793 if (time_after(jiffies, expire))
1794 goto failed_removal;
1796 if (signal_pending(current))
1797 goto failed_removal;
1800 lru_add_drain_all();
1802 drain_all_pages(zone);
1805 pfn = scan_movable_pages(start_pfn, end_pfn);
1806 if (pfn) { /* We have movable pages */
1807 ret = do_migrate_range(pfn, end_pfn);
1813 if (--retry_max == 0)
1814 goto failed_removal;
1820 /* drain all zone's lru pagevec, this is asynchronous... */
1821 lru_add_drain_all();
1823 /* drain pcp pages, this is synchronous. */
1824 drain_all_pages(zone);
1826 * dissolve free hugepages in the memory block before doing offlining
1827 * actually in order to make hugetlbfs's object counting consistent.
1829 dissolve_free_huge_pages(start_pfn, end_pfn);
1831 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1832 if (offlined_pages < 0) {
1834 goto failed_removal;
1836 printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
1837 /* Ok, all of our target is isolated.
1838 We cannot do rollback at this point. */
1839 offline_isolated_pages(start_pfn, end_pfn);
1840 /* reset pagetype flags and makes migrate type to be MOVABLE */
1841 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1842 /* removal success */
1843 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1844 zone->present_pages -= offlined_pages;
1846 pgdat_resize_lock(zone->zone_pgdat, &flags);
1847 zone->zone_pgdat->node_present_pages -= offlined_pages;
1848 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1850 init_per_zone_wmark_min();
1852 if (!populated_zone(zone)) {
1853 zone_pcp_reset(zone);
1854 mutex_lock(&zonelists_mutex);
1855 build_all_zonelists(NULL, NULL);
1856 mutex_unlock(&zonelists_mutex);
1858 zone_pcp_update(zone);
1860 node_states_clear_node(node, &arg);
1861 if (arg.status_change_nid >= 0)
1864 vm_total_pages = nr_free_pagecache_pages();
1865 writeback_set_ratelimit();
1867 memory_notify(MEM_OFFLINE, &arg);
1871 printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n",
1872 (unsigned long long) start_pfn << PAGE_SHIFT,
1873 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1874 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1875 /* pushback to free area */
1876 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1880 /* Must be protected by mem_hotplug_begin() */
1881 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1883 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1885 #endif /* CONFIG_MEMORY_HOTREMOVE */
1888 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1889 * @start_pfn: start pfn of the memory range
1890 * @end_pfn: end pfn of the memory range
1891 * @arg: argument passed to func
1892 * @func: callback for each memory section walked
1894 * This function walks through all present mem sections in range
1895 * [start_pfn, end_pfn) and call func on each mem section.
1897 * Returns the return value of func.
1899 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1900 void *arg, int (*func)(struct memory_block *, void *))
1902 struct memory_block *mem = NULL;
1903 struct mem_section *section;
1904 unsigned long pfn, section_nr;
1907 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1908 section_nr = pfn_to_section_nr(pfn);
1909 if (!present_section_nr(section_nr))
1912 section = __nr_to_section(section_nr);
1913 /* same memblock? */
1915 if ((section_nr >= mem->start_section_nr) &&
1916 (section_nr <= mem->end_section_nr))
1919 mem = find_memory_block_hinted(section, mem);
1923 ret = func(mem, arg);
1925 kobject_put(&mem->dev.kobj);
1931 kobject_put(&mem->dev.kobj);
1936 #ifdef CONFIG_MEMORY_HOTREMOVE
1937 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1939 int ret = !is_memblock_offlined(mem);
1941 if (unlikely(ret)) {
1942 phys_addr_t beginpa, endpa;
1944 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1945 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1946 pr_warn("removing memory fails, because memory "
1947 "[%pa-%pa] is onlined\n",
1954 static int check_cpu_on_node(pg_data_t *pgdat)
1958 for_each_present_cpu(cpu) {
1959 if (cpu_to_node(cpu) == pgdat->node_id)
1961 * the cpu on this node isn't removed, and we can't
1962 * offline this node.
1970 static void unmap_cpu_on_node(pg_data_t *pgdat)
1972 #ifdef CONFIG_ACPI_NUMA
1975 for_each_possible_cpu(cpu)
1976 if (cpu_to_node(cpu) == pgdat->node_id)
1977 numa_clear_node(cpu);
1981 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
1985 ret = check_cpu_on_node(pgdat);
1990 * the node will be offlined when we come here, so we can clear
1991 * the cpu_to_node() now.
1994 unmap_cpu_on_node(pgdat);
2001 * Offline a node if all memory sections and cpus of the node are removed.
2003 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2004 * and online/offline operations before this call.
2006 void try_offline_node(int nid)
2008 pg_data_t *pgdat = NODE_DATA(nid);
2009 unsigned long start_pfn = pgdat->node_start_pfn;
2010 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
2014 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2015 unsigned long section_nr = pfn_to_section_nr(pfn);
2017 if (!present_section_nr(section_nr))
2020 if (pfn_to_nid(pfn) != nid)
2024 * some memory sections of this node are not removed, and we
2025 * can't offline node now.
2030 if (check_and_unmap_cpu_on_node(pgdat))
2034 * all memory/cpu of this node are removed, we can offline this
2037 node_set_offline(nid);
2038 unregister_one_node(nid);
2040 /* free waittable in each zone */
2041 for (i = 0; i < MAX_NR_ZONES; i++) {
2042 struct zone *zone = pgdat->node_zones + i;
2045 * wait_table may be allocated from boot memory,
2046 * here only free if it's allocated by vmalloc.
2048 if (is_vmalloc_addr(zone->wait_table)) {
2049 vfree(zone->wait_table);
2050 zone->wait_table = NULL;
2054 EXPORT_SYMBOL(try_offline_node);
2059 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2060 * and online/offline operations before this call, as required by
2061 * try_offline_node().
2063 void __ref remove_memory(int nid, u64 start, u64 size)
2067 BUG_ON(check_hotplug_memory_range(start, size));
2069 mem_hotplug_begin();
2072 * All memory blocks must be offlined before removing memory. Check
2073 * whether all memory blocks in question are offline and trigger a BUG()
2074 * if this is not the case.
2076 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
2077 check_memblock_offlined_cb);
2081 /* remove memmap entry */
2082 firmware_map_remove(start, start + size, "System RAM");
2083 memblock_free(start, size);
2084 memblock_remove(start, size);
2086 arch_remove_memory(start, size);
2088 try_offline_node(nid);
2092 EXPORT_SYMBOL_GPL(remove_memory);
2093 #endif /* CONFIG_MEMORY_HOTREMOVE */