2 * linux/mm/memory_hotplug.c
7 #include <linux/stddef.h>
9 #include <linux/swap.h>
10 #include <linux/interrupt.h>
11 #include <linux/pagemap.h>
12 #include <linux/compiler.h>
13 #include <linux/export.h>
14 #include <linux/pagevec.h>
15 #include <linux/writeback.h>
16 #include <linux/slab.h>
17 #include <linux/sysctl.h>
18 #include <linux/cpu.h>
19 #include <linux/memory.h>
20 #include <linux/memremap.h>
21 #include <linux/memory_hotplug.h>
22 #include <linux/highmem.h>
23 #include <linux/vmalloc.h>
24 #include <linux/ioport.h>
25 #include <linux/delay.h>
26 #include <linux/migrate.h>
27 #include <linux/page-isolation.h>
28 #include <linux/pfn.h>
29 #include <linux/suspend.h>
30 #include <linux/mm_inline.h>
31 #include <linux/firmware-map.h>
32 #include <linux/stop_machine.h>
33 #include <linux/hugetlb.h>
34 #include <linux/memblock.h>
35 #include <linux/bootmem.h>
36 #include <linux/compaction.h>
38 #include <asm/tlbflush.h>
43 * online_page_callback contains pointer to current page onlining function.
44 * Initially it is generic_online_page(). If it is required it could be
45 * changed by calling set_online_page_callback() for callback registration
46 * and restore_online_page_callback() for generic callback restore.
49 static void generic_online_page(struct page *page);
51 static online_page_callback_t online_page_callback = generic_online_page;
52 static DEFINE_MUTEX(online_page_callback_lock);
54 /* The same as the cpu_hotplug lock, but for memory hotplug. */
56 struct task_struct *active_writer;
57 struct mutex lock; /* Synchronizes accesses to refcount, */
59 * Also blocks the new readers during
60 * an ongoing mem hotplug operation.
64 #ifdef CONFIG_DEBUG_LOCK_ALLOC
65 struct lockdep_map dep_map;
68 .active_writer = NULL,
69 .lock = __MUTEX_INITIALIZER(mem_hotplug.lock),
71 #ifdef CONFIG_DEBUG_LOCK_ALLOC
72 .dep_map = {.name = "mem_hotplug.lock" },
76 /* Lockdep annotations for get/put_online_mems() and mem_hotplug_begin/end() */
77 #define memhp_lock_acquire_read() lock_map_acquire_read(&mem_hotplug.dep_map)
78 #define memhp_lock_acquire() lock_map_acquire(&mem_hotplug.dep_map)
79 #define memhp_lock_release() lock_map_release(&mem_hotplug.dep_map)
81 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
82 bool memhp_auto_online;
84 bool memhp_auto_online = true;
86 EXPORT_SYMBOL_GPL(memhp_auto_online);
88 static int __init setup_memhp_default_state(char *str)
90 if (!strcmp(str, "online"))
91 memhp_auto_online = true;
92 else if (!strcmp(str, "offline"))
93 memhp_auto_online = false;
97 __setup("memhp_default_state=", setup_memhp_default_state);
99 void get_online_mems(void)
102 if (mem_hotplug.active_writer == current)
104 memhp_lock_acquire_read();
105 mutex_lock(&mem_hotplug.lock);
106 mem_hotplug.refcount++;
107 mutex_unlock(&mem_hotplug.lock);
111 void put_online_mems(void)
113 if (mem_hotplug.active_writer == current)
115 mutex_lock(&mem_hotplug.lock);
117 if (WARN_ON(!mem_hotplug.refcount))
118 mem_hotplug.refcount++; /* try to fix things up */
120 if (!--mem_hotplug.refcount && unlikely(mem_hotplug.active_writer))
121 wake_up_process(mem_hotplug.active_writer);
122 mutex_unlock(&mem_hotplug.lock);
123 memhp_lock_release();
127 void mem_hotplug_begin(void)
129 mem_hotplug.active_writer = current;
131 memhp_lock_acquire();
133 mutex_lock(&mem_hotplug.lock);
134 if (likely(!mem_hotplug.refcount))
136 __set_current_state(TASK_UNINTERRUPTIBLE);
137 mutex_unlock(&mem_hotplug.lock);
142 void mem_hotplug_done(void)
144 mem_hotplug.active_writer = NULL;
145 mutex_unlock(&mem_hotplug.lock);
146 memhp_lock_release();
149 /* add this memory to iomem resource */
150 static struct resource *register_memory_resource(u64 start, u64 size)
152 struct resource *res;
153 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
155 return ERR_PTR(-ENOMEM);
157 res->name = "System RAM";
159 res->end = start + size - 1;
160 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
161 if (request_resource(&iomem_resource, res) < 0) {
162 pr_debug("System RAM resource %pR cannot be added\n", res);
164 return ERR_PTR(-EEXIST);
169 static void release_memory_resource(struct resource *res)
173 release_resource(res);
178 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
179 void get_page_bootmem(unsigned long info, struct page *page,
182 page->freelist = (void *)type;
183 SetPagePrivate(page);
184 set_page_private(page, info);
188 void put_page_bootmem(struct page *page)
192 type = (unsigned long) page->freelist;
193 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
194 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
196 if (page_ref_dec_return(page) == 1) {
197 page->freelist = NULL;
198 ClearPagePrivate(page);
199 set_page_private(page, 0);
200 INIT_LIST_HEAD(&page->lru);
201 free_reserved_page(page);
205 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
206 #ifndef CONFIG_SPARSEMEM_VMEMMAP
207 static void register_page_bootmem_info_section(unsigned long start_pfn)
209 unsigned long *usemap, mapsize, section_nr, i;
210 struct mem_section *ms;
211 struct page *page, *memmap;
213 section_nr = pfn_to_section_nr(start_pfn);
214 ms = __nr_to_section(section_nr);
216 /* Get section's memmap address */
217 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
220 * Get page for the memmap's phys address
221 * XXX: need more consideration for sparse_vmemmap...
223 page = virt_to_page(memmap);
224 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
225 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
227 /* remember memmap's page */
228 for (i = 0; i < mapsize; i++, page++)
229 get_page_bootmem(section_nr, page, SECTION_INFO);
231 usemap = __nr_to_section(section_nr)->pageblock_flags;
232 page = virt_to_page(usemap);
234 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
236 for (i = 0; i < mapsize; i++, page++)
237 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
240 #else /* CONFIG_SPARSEMEM_VMEMMAP */
241 static void register_page_bootmem_info_section(unsigned long start_pfn)
243 unsigned long *usemap, mapsize, section_nr, i;
244 struct mem_section *ms;
245 struct page *page, *memmap;
247 if (!pfn_valid(start_pfn))
250 section_nr = pfn_to_section_nr(start_pfn);
251 ms = __nr_to_section(section_nr);
253 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
255 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
257 usemap = __nr_to_section(section_nr)->pageblock_flags;
258 page = virt_to_page(usemap);
260 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
262 for (i = 0; i < mapsize; i++, page++)
263 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
265 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
267 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
269 unsigned long i, pfn, end_pfn, nr_pages;
270 int node = pgdat->node_id;
273 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
274 page = virt_to_page(pgdat);
276 for (i = 0; i < nr_pages; i++, page++)
277 get_page_bootmem(node, page, NODE_INFO);
279 pfn = pgdat->node_start_pfn;
280 end_pfn = pgdat_end_pfn(pgdat);
282 /* register section info */
283 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
285 * Some platforms can assign the same pfn to multiple nodes - on
286 * node0 as well as nodeN. To avoid registering a pfn against
287 * multiple nodes we check that this pfn does not already
288 * reside in some other nodes.
290 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
291 register_page_bootmem_info_section(pfn);
294 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
296 static void __meminit grow_zone_span(struct zone *zone, unsigned long start_pfn,
297 unsigned long end_pfn)
299 unsigned long old_zone_end_pfn;
301 zone_span_writelock(zone);
303 old_zone_end_pfn = zone_end_pfn(zone);
304 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
305 zone->zone_start_pfn = start_pfn;
307 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
308 zone->zone_start_pfn;
310 zone_span_writeunlock(zone);
313 static void resize_zone(struct zone *zone, unsigned long start_pfn,
314 unsigned long end_pfn)
316 zone_span_writelock(zone);
318 if (end_pfn - start_pfn) {
319 zone->zone_start_pfn = start_pfn;
320 zone->spanned_pages = end_pfn - start_pfn;
323 * make it consist as free_area_init_core(),
324 * if spanned_pages = 0, then keep start_pfn = 0
326 zone->zone_start_pfn = 0;
327 zone->spanned_pages = 0;
330 zone_span_writeunlock(zone);
333 static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
334 unsigned long end_pfn)
336 enum zone_type zid = zone_idx(zone);
337 int nid = zone->zone_pgdat->node_id;
340 for (pfn = start_pfn; pfn < end_pfn; pfn++)
341 set_page_links(pfn_to_page(pfn), zid, nid, pfn);
344 /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
345 * alloc_bootmem_node_nopanic()/memblock_virt_alloc_node_nopanic() */
346 static int __ref ensure_zone_is_initialized(struct zone *zone,
347 unsigned long start_pfn, unsigned long num_pages)
349 if (!zone_is_initialized(zone))
350 return init_currently_empty_zone(zone, start_pfn, num_pages);
355 static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
356 unsigned long start_pfn, unsigned long end_pfn)
360 unsigned long z1_start_pfn;
362 ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn);
366 pgdat_resize_lock(z1->zone_pgdat, &flags);
368 /* can't move pfns which are higher than @z2 */
369 if (end_pfn > zone_end_pfn(z2))
371 /* the move out part must be at the left most of @z2 */
372 if (start_pfn > z2->zone_start_pfn)
374 /* must included/overlap */
375 if (end_pfn <= z2->zone_start_pfn)
378 /* use start_pfn for z1's start_pfn if z1 is empty */
379 if (!zone_is_empty(z1))
380 z1_start_pfn = z1->zone_start_pfn;
382 z1_start_pfn = start_pfn;
384 resize_zone(z1, z1_start_pfn, end_pfn);
385 resize_zone(z2, end_pfn, zone_end_pfn(z2));
387 pgdat_resize_unlock(z1->zone_pgdat, &flags);
389 fix_zone_id(z1, start_pfn, end_pfn);
393 pgdat_resize_unlock(z1->zone_pgdat, &flags);
397 static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
398 unsigned long start_pfn, unsigned long end_pfn)
402 unsigned long z2_end_pfn;
404 ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn);
408 pgdat_resize_lock(z1->zone_pgdat, &flags);
410 /* can't move pfns which are lower than @z1 */
411 if (z1->zone_start_pfn > start_pfn)
413 /* the move out part mast at the right most of @z1 */
414 if (zone_end_pfn(z1) > end_pfn)
416 /* must included/overlap */
417 if (start_pfn >= zone_end_pfn(z1))
420 /* use end_pfn for z2's end_pfn if z2 is empty */
421 if (!zone_is_empty(z2))
422 z2_end_pfn = zone_end_pfn(z2);
424 z2_end_pfn = end_pfn;
426 resize_zone(z1, z1->zone_start_pfn, start_pfn);
427 resize_zone(z2, start_pfn, z2_end_pfn);
429 pgdat_resize_unlock(z1->zone_pgdat, &flags);
431 fix_zone_id(z2, start_pfn, end_pfn);
435 pgdat_resize_unlock(z1->zone_pgdat, &flags);
439 static struct zone * __meminit move_pfn_range(int zone_shift,
440 unsigned long start_pfn, unsigned long end_pfn)
442 struct zone *zone = page_zone(pfn_to_page(start_pfn));
446 ret = move_pfn_range_left(zone + zone_shift, zone,
449 ret = move_pfn_range_right(zone, zone + zone_shift,
455 return zone + zone_shift;
458 static void __meminit grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
459 unsigned long end_pfn)
461 unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat);
463 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
464 pgdat->node_start_pfn = start_pfn;
466 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
467 pgdat->node_start_pfn;
470 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
472 struct pglist_data *pgdat = zone->zone_pgdat;
473 int nr_pages = PAGES_PER_SECTION;
474 int nid = pgdat->node_id;
476 unsigned long flags, pfn;
479 zone_type = zone - pgdat->node_zones;
480 ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
484 pgdat_resize_lock(zone->zone_pgdat, &flags);
485 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
486 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
487 phys_start_pfn + nr_pages);
488 pgdat_resize_unlock(zone->zone_pgdat, &flags);
489 memmap_init_zone(nr_pages, nid, zone_type,
490 phys_start_pfn, MEMMAP_HOTPLUG);
492 /* online_page_range is called later and expects pages reserved */
493 for (pfn = phys_start_pfn; pfn < phys_start_pfn + nr_pages; pfn++) {
497 SetPageReserved(pfn_to_page(pfn));
502 static int __meminit __add_section(int nid, struct zone *zone,
503 unsigned long phys_start_pfn)
507 if (pfn_valid(phys_start_pfn))
510 ret = sparse_add_one_section(zone, phys_start_pfn);
515 ret = __add_zone(zone, phys_start_pfn);
520 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
524 * Reasonably generic function for adding memory. It is
525 * expected that archs that support memory hotplug will
526 * call this function after deciding the zone to which to
529 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
530 unsigned long nr_pages)
534 int start_sec, end_sec;
535 struct vmem_altmap *altmap;
537 clear_zone_contiguous(zone);
539 /* during initialize mem_map, align hot-added range to section */
540 start_sec = pfn_to_section_nr(phys_start_pfn);
541 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
543 altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn));
546 * Validate altmap is within bounds of the total request
548 if (altmap->base_pfn != phys_start_pfn
549 || vmem_altmap_offset(altmap) > nr_pages) {
550 pr_warn_once("memory add fail, invalid altmap\n");
557 for (i = start_sec; i <= end_sec; i++) {
558 err = __add_section(nid, zone, section_nr_to_pfn(i));
561 * EEXIST is finally dealt with by ioresource collision
562 * check. see add_memory() => register_memory_resource()
563 * Warning will be printed if there is collision.
565 if (err && (err != -EEXIST))
569 vmemmap_populate_print_last();
571 set_zone_contiguous(zone);
574 EXPORT_SYMBOL_GPL(__add_pages);
576 #ifdef CONFIG_MEMORY_HOTREMOVE
577 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
578 static int find_smallest_section_pfn(int nid, struct zone *zone,
579 unsigned long start_pfn,
580 unsigned long end_pfn)
582 struct mem_section *ms;
584 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
585 ms = __pfn_to_section(start_pfn);
587 if (unlikely(!valid_section(ms)))
590 if (unlikely(pfn_to_nid(start_pfn) != nid))
593 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
602 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
603 static int find_biggest_section_pfn(int nid, struct zone *zone,
604 unsigned long start_pfn,
605 unsigned long end_pfn)
607 struct mem_section *ms;
610 /* pfn is the end pfn of a memory section. */
612 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
613 ms = __pfn_to_section(pfn);
615 if (unlikely(!valid_section(ms)))
618 if (unlikely(pfn_to_nid(pfn) != nid))
621 if (zone && zone != page_zone(pfn_to_page(pfn)))
630 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
631 unsigned long end_pfn)
633 unsigned long zone_start_pfn = zone->zone_start_pfn;
634 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
635 unsigned long zone_end_pfn = z;
637 struct mem_section *ms;
638 int nid = zone_to_nid(zone);
640 zone_span_writelock(zone);
641 if (zone_start_pfn == start_pfn) {
643 * If the section is smallest section in the zone, it need
644 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
645 * In this case, we find second smallest valid mem_section
646 * for shrinking zone.
648 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
651 zone->zone_start_pfn = pfn;
652 zone->spanned_pages = zone_end_pfn - pfn;
654 } else if (zone_end_pfn == end_pfn) {
656 * If the section is biggest section in the zone, it need
657 * shrink zone->spanned_pages.
658 * In this case, we find second biggest valid mem_section for
661 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
664 zone->spanned_pages = pfn - zone_start_pfn + 1;
668 * The section is not biggest or smallest mem_section in the zone, it
669 * only creates a hole in the zone. So in this case, we need not
670 * change the zone. But perhaps, the zone has only hole data. Thus
671 * it check the zone has only hole or not.
673 pfn = zone_start_pfn;
674 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
675 ms = __pfn_to_section(pfn);
677 if (unlikely(!valid_section(ms)))
680 if (page_zone(pfn_to_page(pfn)) != zone)
683 /* If the section is current section, it continues the loop */
684 if (start_pfn == pfn)
687 /* If we find valid section, we have nothing to do */
688 zone_span_writeunlock(zone);
692 /* The zone has no valid section */
693 zone->zone_start_pfn = 0;
694 zone->spanned_pages = 0;
695 zone_span_writeunlock(zone);
698 static void shrink_pgdat_span(struct pglist_data *pgdat,
699 unsigned long start_pfn, unsigned long end_pfn)
701 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
702 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
703 unsigned long pgdat_end_pfn = p;
705 struct mem_section *ms;
706 int nid = pgdat->node_id;
708 if (pgdat_start_pfn == start_pfn) {
710 * If the section is smallest section in the pgdat, it need
711 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
712 * In this case, we find second smallest valid mem_section
713 * for shrinking zone.
715 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
718 pgdat->node_start_pfn = pfn;
719 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
721 } else if (pgdat_end_pfn == end_pfn) {
723 * If the section is biggest section in the pgdat, it need
724 * shrink pgdat->node_spanned_pages.
725 * In this case, we find second biggest valid mem_section for
728 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
731 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
735 * If the section is not biggest or smallest mem_section in the pgdat,
736 * it only creates a hole in the pgdat. So in this case, we need not
738 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
739 * has only hole or not.
741 pfn = pgdat_start_pfn;
742 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
743 ms = __pfn_to_section(pfn);
745 if (unlikely(!valid_section(ms)))
748 if (pfn_to_nid(pfn) != nid)
751 /* If the section is current section, it continues the loop */
752 if (start_pfn == pfn)
755 /* If we find valid section, we have nothing to do */
759 /* The pgdat has no valid section */
760 pgdat->node_start_pfn = 0;
761 pgdat->node_spanned_pages = 0;
764 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
766 struct pglist_data *pgdat = zone->zone_pgdat;
767 int nr_pages = PAGES_PER_SECTION;
771 zone_type = zone - pgdat->node_zones;
773 pgdat_resize_lock(zone->zone_pgdat, &flags);
774 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
775 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
776 pgdat_resize_unlock(zone->zone_pgdat, &flags);
779 static int __remove_section(struct zone *zone, struct mem_section *ms,
780 unsigned long map_offset)
782 unsigned long start_pfn;
786 if (!valid_section(ms))
789 ret = unregister_memory_section(ms);
793 scn_nr = __section_nr(ms);
794 start_pfn = section_nr_to_pfn(scn_nr);
795 __remove_zone(zone, start_pfn);
797 sparse_remove_one_section(zone, ms, map_offset);
802 * __remove_pages() - remove sections of pages from a zone
803 * @zone: zone from which pages need to be removed
804 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
805 * @nr_pages: number of pages to remove (must be multiple of section size)
807 * Generic helper function to remove section mappings and sysfs entries
808 * for the section of the memory we are removing. Caller needs to make
809 * sure that pages are marked reserved and zones are adjust properly by
810 * calling offline_pages().
812 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
813 unsigned long nr_pages)
816 unsigned long map_offset = 0;
817 int sections_to_remove, ret = 0;
819 /* In the ZONE_DEVICE case device driver owns the memory region */
820 if (is_dev_zone(zone)) {
821 struct page *page = pfn_to_page(phys_start_pfn);
822 struct vmem_altmap *altmap;
824 altmap = to_vmem_altmap((unsigned long) page);
826 map_offset = vmem_altmap_offset(altmap);
828 resource_size_t start, size;
830 start = phys_start_pfn << PAGE_SHIFT;
831 size = nr_pages * PAGE_SIZE;
833 ret = release_mem_region_adjustable(&iomem_resource, start,
836 resource_size_t endres = start + size - 1;
838 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
839 &start, &endres, ret);
843 clear_zone_contiguous(zone);
846 * We can only remove entire sections
848 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
849 BUG_ON(nr_pages % PAGES_PER_SECTION);
851 sections_to_remove = nr_pages / PAGES_PER_SECTION;
852 for (i = 0; i < sections_to_remove; i++) {
853 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
855 ret = __remove_section(zone, __pfn_to_section(pfn), map_offset);
861 set_zone_contiguous(zone);
865 EXPORT_SYMBOL_GPL(__remove_pages);
866 #endif /* CONFIG_MEMORY_HOTREMOVE */
868 int set_online_page_callback(online_page_callback_t callback)
873 mutex_lock(&online_page_callback_lock);
875 if (online_page_callback == generic_online_page) {
876 online_page_callback = callback;
880 mutex_unlock(&online_page_callback_lock);
885 EXPORT_SYMBOL_GPL(set_online_page_callback);
887 int restore_online_page_callback(online_page_callback_t callback)
892 mutex_lock(&online_page_callback_lock);
894 if (online_page_callback == callback) {
895 online_page_callback = generic_online_page;
899 mutex_unlock(&online_page_callback_lock);
904 EXPORT_SYMBOL_GPL(restore_online_page_callback);
906 void __online_page_set_limits(struct page *page)
909 EXPORT_SYMBOL_GPL(__online_page_set_limits);
911 void __online_page_increment_counters(struct page *page)
913 adjust_managed_page_count(page, 1);
915 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
917 void __online_page_free(struct page *page)
919 __free_reserved_page(page);
921 EXPORT_SYMBOL_GPL(__online_page_free);
923 static void generic_online_page(struct page *page)
925 __online_page_set_limits(page);
926 __online_page_increment_counters(page);
927 __online_page_free(page);
930 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
934 unsigned long onlined_pages = *(unsigned long *)arg;
936 if (PageReserved(pfn_to_page(start_pfn)))
937 for (i = 0; i < nr_pages; i++) {
938 page = pfn_to_page(start_pfn + i);
939 (*online_page_callback)(page);
942 *(unsigned long *)arg = onlined_pages;
946 #ifdef CONFIG_MOVABLE_NODE
948 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
951 static bool can_online_high_movable(struct zone *zone)
955 #else /* CONFIG_MOVABLE_NODE */
956 /* ensure every online node has NORMAL memory */
957 static bool can_online_high_movable(struct zone *zone)
959 return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
961 #endif /* CONFIG_MOVABLE_NODE */
963 /* check which state of node_states will be changed when online memory */
964 static void node_states_check_changes_online(unsigned long nr_pages,
965 struct zone *zone, struct memory_notify *arg)
967 int nid = zone_to_nid(zone);
968 enum zone_type zone_last = ZONE_NORMAL;
971 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
972 * contains nodes which have zones of 0...ZONE_NORMAL,
973 * set zone_last to ZONE_NORMAL.
975 * If we don't have HIGHMEM nor movable node,
976 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
977 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
979 if (N_MEMORY == N_NORMAL_MEMORY)
980 zone_last = ZONE_MOVABLE;
983 * if the memory to be online is in a zone of 0...zone_last, and
984 * the zones of 0...zone_last don't have memory before online, we will
985 * need to set the node to node_states[N_NORMAL_MEMORY] after
986 * the memory is online.
988 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
989 arg->status_change_nid_normal = nid;
991 arg->status_change_nid_normal = -1;
993 #ifdef CONFIG_HIGHMEM
995 * If we have movable node, node_states[N_HIGH_MEMORY]
996 * contains nodes which have zones of 0...ZONE_HIGHMEM,
997 * set zone_last to ZONE_HIGHMEM.
999 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1000 * contains nodes which have zones of 0...ZONE_MOVABLE,
1001 * set zone_last to ZONE_MOVABLE.
1003 zone_last = ZONE_HIGHMEM;
1004 if (N_MEMORY == N_HIGH_MEMORY)
1005 zone_last = ZONE_MOVABLE;
1007 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
1008 arg->status_change_nid_high = nid;
1010 arg->status_change_nid_high = -1;
1012 arg->status_change_nid_high = arg->status_change_nid_normal;
1016 * if the node don't have memory befor online, we will need to
1017 * set the node to node_states[N_MEMORY] after the memory
1020 if (!node_state(nid, N_MEMORY))
1021 arg->status_change_nid = nid;
1023 arg->status_change_nid = -1;
1026 static void node_states_set_node(int node, struct memory_notify *arg)
1028 if (arg->status_change_nid_normal >= 0)
1029 node_set_state(node, N_NORMAL_MEMORY);
1031 if (arg->status_change_nid_high >= 0)
1032 node_set_state(node, N_HIGH_MEMORY);
1034 node_set_state(node, N_MEMORY);
1037 bool zone_can_shift(unsigned long pfn, unsigned long nr_pages,
1038 enum zone_type target, int *zone_shift)
1040 struct zone *zone = page_zone(pfn_to_page(pfn));
1041 enum zone_type idx = zone_idx(zone);
1047 /* pages must be at end of current zone */
1048 if (pfn + nr_pages != zone_end_pfn(zone))
1051 /* no zones in use between current zone and target */
1052 for (i = idx + 1; i < target; i++)
1053 if (zone_is_initialized(zone - idx + i))
1058 /* pages must be at beginning of current zone */
1059 if (pfn != zone->zone_start_pfn)
1062 /* no zones in use between current zone and target */
1063 for (i = target + 1; i < idx; i++)
1064 if (zone_is_initialized(zone - idx + i))
1068 *zone_shift = target - idx;
1072 /* Must be protected by mem_hotplug_begin() */
1073 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
1075 unsigned long flags;
1076 unsigned long onlined_pages = 0;
1078 int need_zonelists_rebuild = 0;
1081 struct memory_notify arg;
1085 * This doesn't need a lock to do pfn_to_page().
1086 * The section can't be removed here because of the
1087 * memory_block->state_mutex.
1089 zone = page_zone(pfn_to_page(pfn));
1091 if ((zone_idx(zone) > ZONE_NORMAL ||
1092 online_type == MMOP_ONLINE_MOVABLE) &&
1093 !can_online_high_movable(zone))
1096 if (online_type == MMOP_ONLINE_KERNEL) {
1097 if (!zone_can_shift(pfn, nr_pages, ZONE_NORMAL, &zone_shift))
1099 } else if (online_type == MMOP_ONLINE_MOVABLE) {
1100 if (!zone_can_shift(pfn, nr_pages, ZONE_MOVABLE, &zone_shift))
1104 zone = move_pfn_range(zone_shift, pfn, pfn + nr_pages);
1108 arg.start_pfn = pfn;
1109 arg.nr_pages = nr_pages;
1110 node_states_check_changes_online(nr_pages, zone, &arg);
1112 nid = zone_to_nid(zone);
1114 ret = memory_notify(MEM_GOING_ONLINE, &arg);
1115 ret = notifier_to_errno(ret);
1117 goto failed_addition;
1120 * If this zone is not populated, then it is not in zonelist.
1121 * This means the page allocator ignores this zone.
1122 * So, zonelist must be updated after online.
1124 mutex_lock(&zonelists_mutex);
1125 if (!populated_zone(zone)) {
1126 need_zonelists_rebuild = 1;
1127 build_all_zonelists(NULL, zone);
1130 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
1131 online_pages_range);
1133 if (need_zonelists_rebuild)
1134 zone_pcp_reset(zone);
1135 mutex_unlock(&zonelists_mutex);
1136 goto failed_addition;
1139 zone->present_pages += onlined_pages;
1141 pgdat_resize_lock(zone->zone_pgdat, &flags);
1142 zone->zone_pgdat->node_present_pages += onlined_pages;
1143 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1145 if (onlined_pages) {
1146 node_states_set_node(nid, &arg);
1147 if (need_zonelists_rebuild)
1148 build_all_zonelists(NULL, NULL);
1150 zone_pcp_update(zone);
1153 mutex_unlock(&zonelists_mutex);
1155 init_per_zone_wmark_min();
1157 if (onlined_pages) {
1162 vm_total_pages = nr_free_pagecache_pages();
1164 writeback_set_ratelimit();
1167 memory_notify(MEM_ONLINE, &arg);
1171 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
1172 (unsigned long long) pfn << PAGE_SHIFT,
1173 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
1174 memory_notify(MEM_CANCEL_ONLINE, &arg);
1177 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1179 static void reset_node_present_pages(pg_data_t *pgdat)
1183 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
1184 z->present_pages = 0;
1186 pgdat->node_present_pages = 0;
1189 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1190 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1192 struct pglist_data *pgdat;
1193 unsigned long zones_size[MAX_NR_ZONES] = {0};
1194 unsigned long zholes_size[MAX_NR_ZONES] = {0};
1195 unsigned long start_pfn = PFN_DOWN(start);
1197 pgdat = NODE_DATA(nid);
1199 pgdat = arch_alloc_nodedata(nid);
1203 arch_refresh_nodedata(nid, pgdat);
1205 /* Reset the nr_zones, order and classzone_idx before reuse */
1206 pgdat->nr_zones = 0;
1207 pgdat->kswapd_order = 0;
1208 pgdat->kswapd_classzone_idx = 0;
1211 /* we can use NODE_DATA(nid) from here */
1213 /* init node's zones as empty zones, we don't have any present pages.*/
1214 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1215 pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
1218 * The node we allocated has no zone fallback lists. For avoiding
1219 * to access not-initialized zonelist, build here.
1221 mutex_lock(&zonelists_mutex);
1222 build_all_zonelists(pgdat, NULL);
1223 mutex_unlock(&zonelists_mutex);
1226 * zone->managed_pages is set to an approximate value in
1227 * free_area_init_core(), which will cause
1228 * /sys/device/system/node/nodeX/meminfo has wrong data.
1229 * So reset it to 0 before any memory is onlined.
1231 reset_node_managed_pages(pgdat);
1234 * When memory is hot-added, all the memory is in offline state. So
1235 * clear all zones' present_pages because they will be updated in
1236 * online_pages() and offline_pages().
1238 reset_node_present_pages(pgdat);
1243 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1245 arch_refresh_nodedata(nid, NULL);
1246 free_percpu(pgdat->per_cpu_nodestats);
1247 arch_free_nodedata(pgdat);
1253 * try_online_node - online a node if offlined
1255 * called by cpu_up() to online a node without onlined memory.
1257 int try_online_node(int nid)
1262 if (node_online(nid))
1265 mem_hotplug_begin();
1266 pgdat = hotadd_new_pgdat(nid, 0);
1268 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1272 node_set_online(nid);
1273 ret = register_one_node(nid);
1276 if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
1277 mutex_lock(&zonelists_mutex);
1278 build_all_zonelists(NULL, NULL);
1279 mutex_unlock(&zonelists_mutex);
1287 static int check_hotplug_memory_range(u64 start, u64 size)
1289 u64 start_pfn = PFN_DOWN(start);
1290 u64 nr_pages = size >> PAGE_SHIFT;
1292 /* Memory range must be aligned with section */
1293 if ((start_pfn & ~PAGE_SECTION_MASK) ||
1294 (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1295 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1296 (unsigned long long)start,
1297 (unsigned long long)size);
1305 * If movable zone has already been setup, newly added memory should be check.
1306 * If its address is higher than movable zone, it should be added as movable.
1307 * Without this check, movable zone may overlap with other zone.
1309 static int should_add_memory_movable(int nid, u64 start, u64 size)
1311 unsigned long start_pfn = start >> PAGE_SHIFT;
1312 pg_data_t *pgdat = NODE_DATA(nid);
1313 struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE;
1315 if (zone_is_empty(movable_zone))
1318 if (movable_zone->zone_start_pfn <= start_pfn)
1324 int zone_for_memory(int nid, u64 start, u64 size, int zone_default,
1327 #ifdef CONFIG_ZONE_DEVICE
1331 if (should_add_memory_movable(nid, start, size))
1332 return ZONE_MOVABLE;
1334 return zone_default;
1337 static int online_memory_block(struct memory_block *mem, void *arg)
1339 return memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
1342 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1343 int __ref add_memory_resource(int nid, struct resource *res, bool online)
1346 pg_data_t *pgdat = NULL;
1352 size = resource_size(res);
1354 ret = check_hotplug_memory_range(start, size);
1358 { /* Stupid hack to suppress address-never-null warning */
1359 void *p = NODE_DATA(nid);
1363 mem_hotplug_begin();
1366 * Add new range to memblock so that when hotadd_new_pgdat() is called
1367 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1368 * this new range and calculate total pages correctly. The range will
1369 * be removed at hot-remove time.
1371 memblock_add_node(start, size, nid);
1373 new_node = !node_online(nid);
1375 pgdat = hotadd_new_pgdat(nid, start);
1381 /* call arch's memory hotadd */
1382 ret = arch_add_memory(nid, start, size, false);
1387 /* we online node here. we can't roll back from here. */
1388 node_set_online(nid);
1391 ret = register_one_node(nid);
1393 * If sysfs file of new node can't create, cpu on the node
1394 * can't be hot-added. There is no rollback way now.
1395 * So, check by BUG_ON() to catch it reluctantly..
1400 /* create new memmap entry */
1401 firmware_map_add_hotplug(start, start + size, "System RAM");
1403 /* online pages if requested */
1405 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1406 NULL, online_memory_block);
1411 /* rollback pgdat allocation and others */
1413 rollback_node_hotadd(nid, pgdat);
1414 memblock_remove(start, size);
1420 EXPORT_SYMBOL_GPL(add_memory_resource);
1422 int __ref add_memory(int nid, u64 start, u64 size)
1424 struct resource *res;
1427 res = register_memory_resource(start, size);
1429 return PTR_ERR(res);
1431 ret = add_memory_resource(nid, res, memhp_auto_online);
1433 release_memory_resource(res);
1436 EXPORT_SYMBOL_GPL(add_memory);
1438 #ifdef CONFIG_MEMORY_HOTREMOVE
1440 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1441 * set and the size of the free page is given by page_order(). Using this,
1442 * the function determines if the pageblock contains only free pages.
1443 * Due to buddy contraints, a free page at least the size of a pageblock will
1444 * be located at the start of the pageblock
1446 static inline int pageblock_free(struct page *page)
1448 return PageBuddy(page) && page_order(page) >= pageblock_order;
1451 /* Return the start of the next active pageblock after a given page */
1452 static struct page *next_active_pageblock(struct page *page)
1454 /* Ensure the starting page is pageblock-aligned */
1455 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1457 /* If the entire pageblock is free, move to the end of free page */
1458 if (pageblock_free(page)) {
1460 /* be careful. we don't have locks, page_order can be changed.*/
1461 order = page_order(page);
1462 if ((order < MAX_ORDER) && (order >= pageblock_order))
1463 return page + (1 << order);
1466 return page + pageblock_nr_pages;
1469 /* Checks if this range of memory is likely to be hot-removable. */
1470 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1472 struct page *page = pfn_to_page(start_pfn);
1473 struct page *end_page = page + nr_pages;
1475 /* Check the starting page of each pageblock within the range */
1476 for (; page < end_page; page = next_active_pageblock(page)) {
1477 if (!is_pageblock_removable_nolock(page))
1482 /* All pageblocks in the memory block are likely to be hot-removable */
1487 * Confirm all pages in a range [start, end) belong to the same zone.
1488 * When true, return its valid [start, end).
1490 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1491 unsigned long *valid_start, unsigned long *valid_end)
1493 unsigned long pfn, sec_end_pfn;
1494 unsigned long start, end;
1495 struct zone *zone = NULL;
1498 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1500 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1501 /* Make sure the memory section is present first */
1502 if (!present_section_nr(pfn_to_section_nr(pfn)))
1504 for (; pfn < sec_end_pfn && pfn < end_pfn;
1505 pfn += MAX_ORDER_NR_PAGES) {
1507 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1508 while ((i < MAX_ORDER_NR_PAGES) &&
1509 !pfn_valid_within(pfn + i))
1511 if (i == MAX_ORDER_NR_PAGES)
1513 page = pfn_to_page(pfn + i);
1514 if (zone && page_zone(page) != zone)
1518 zone = page_zone(page);
1519 end = pfn + MAX_ORDER_NR_PAGES;
1524 *valid_start = start;
1533 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages
1534 * and hugepages). We scan pfn because it's much easier than scanning over
1535 * linked list. This function returns the pfn of the first found movable
1536 * page if it's found, otherwise 0.
1538 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1542 for (pfn = start; pfn < end; pfn++) {
1543 if (pfn_valid(pfn)) {
1544 page = pfn_to_page(pfn);
1547 if (PageHuge(page)) {
1548 if (page_huge_active(page))
1551 pfn = round_up(pfn + 1,
1552 1 << compound_order(page)) - 1;
1559 static struct page *new_node_page(struct page *page, unsigned long private,
1562 gfp_t gfp_mask = GFP_USER | __GFP_MOVABLE;
1563 int nid = page_to_nid(page);
1564 nodemask_t nmask = node_states[N_MEMORY];
1565 struct page *new_page = NULL;
1568 * TODO: allocate a destination hugepage from a nearest neighbor node,
1569 * accordance with memory policy of the user process if possible. For
1570 * now as a simple work-around, we use the next node for destination.
1573 return alloc_huge_page_node(page_hstate(compound_head(page)),
1574 next_node_in(nid, nmask));
1576 node_clear(nid, nmask);
1578 if (PageHighMem(page)
1579 || (zone_idx(page_zone(page)) == ZONE_MOVABLE))
1580 gfp_mask |= __GFP_HIGHMEM;
1582 if (!nodes_empty(nmask))
1583 new_page = __alloc_pages_nodemask(gfp_mask, 0,
1584 node_zonelist(nid, gfp_mask), &nmask);
1586 new_page = __alloc_pages(gfp_mask, 0,
1587 node_zonelist(nid, gfp_mask));
1592 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1594 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1598 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1599 int not_managed = 0;
1603 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1604 if (!pfn_valid(pfn))
1606 page = pfn_to_page(pfn);
1608 if (PageHuge(page)) {
1609 struct page *head = compound_head(page);
1610 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1611 if (compound_order(head) > PFN_SECTION_SHIFT) {
1615 if (isolate_huge_page(page, &source))
1616 move_pages -= 1 << compound_order(head);
1620 if (!get_page_unless_zero(page))
1623 * We can skip free pages. And we can only deal with pages on
1626 ret = isolate_lru_page(page);
1627 if (!ret) { /* Success */
1629 list_add_tail(&page->lru, &source);
1631 inc_node_page_state(page, NR_ISOLATED_ANON +
1632 page_is_file_cache(page));
1635 #ifdef CONFIG_DEBUG_VM
1636 pr_alert("removing pfn %lx from LRU failed\n", pfn);
1637 dump_page(page, "failed to remove from LRU");
1640 /* Because we don't have big zone->lock. we should
1641 check this again here. */
1642 if (page_count(page)) {
1649 if (!list_empty(&source)) {
1651 putback_movable_pages(&source);
1655 /* Allocate a new page from the nearest neighbor node */
1656 ret = migrate_pages(&source, new_node_page, NULL, 0,
1657 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1659 putback_movable_pages(&source);
1666 * remove from free_area[] and mark all as Reserved.
1669 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1672 __offline_isolated_pages(start, start + nr_pages);
1677 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1679 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1680 offline_isolated_pages_cb);
1684 * Check all pages in range, recoreded as memory resource, are isolated.
1687 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1691 long offlined = *(long *)data;
1692 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1693 offlined = nr_pages;
1695 *(long *)data += offlined;
1700 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1705 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1706 check_pages_isolated_cb);
1708 offlined = (long)ret;
1712 #ifdef CONFIG_MOVABLE_NODE
1714 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1717 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1721 #else /* CONFIG_MOVABLE_NODE */
1722 /* ensure the node has NORMAL memory if it is still online */
1723 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1725 struct pglist_data *pgdat = zone->zone_pgdat;
1726 unsigned long present_pages = 0;
1729 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1730 present_pages += pgdat->node_zones[zt].present_pages;
1732 if (present_pages > nr_pages)
1736 for (; zt <= ZONE_MOVABLE; zt++)
1737 present_pages += pgdat->node_zones[zt].present_pages;
1740 * we can't offline the last normal memory until all
1741 * higher memory is offlined.
1743 return present_pages == 0;
1745 #endif /* CONFIG_MOVABLE_NODE */
1747 static int __init cmdline_parse_movable_node(char *p)
1749 #ifdef CONFIG_MOVABLE_NODE
1750 movable_node_enabled = true;
1752 pr_warn("movable_node option not supported\n");
1756 early_param("movable_node", cmdline_parse_movable_node);
1758 /* check which state of node_states will be changed when offline memory */
1759 static void node_states_check_changes_offline(unsigned long nr_pages,
1760 struct zone *zone, struct memory_notify *arg)
1762 struct pglist_data *pgdat = zone->zone_pgdat;
1763 unsigned long present_pages = 0;
1764 enum zone_type zt, zone_last = ZONE_NORMAL;
1767 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1768 * contains nodes which have zones of 0...ZONE_NORMAL,
1769 * set zone_last to ZONE_NORMAL.
1771 * If we don't have HIGHMEM nor movable node,
1772 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1773 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1775 if (N_MEMORY == N_NORMAL_MEMORY)
1776 zone_last = ZONE_MOVABLE;
1779 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1780 * If the memory to be offline is in a zone of 0...zone_last,
1781 * and it is the last present memory, 0...zone_last will
1782 * become empty after offline , thus we can determind we will
1783 * need to clear the node from node_states[N_NORMAL_MEMORY].
1785 for (zt = 0; zt <= zone_last; zt++)
1786 present_pages += pgdat->node_zones[zt].present_pages;
1787 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1788 arg->status_change_nid_normal = zone_to_nid(zone);
1790 arg->status_change_nid_normal = -1;
1792 #ifdef CONFIG_HIGHMEM
1794 * If we have movable node, node_states[N_HIGH_MEMORY]
1795 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1796 * set zone_last to ZONE_HIGHMEM.
1798 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1799 * contains nodes which have zones of 0...ZONE_MOVABLE,
1800 * set zone_last to ZONE_MOVABLE.
1802 zone_last = ZONE_HIGHMEM;
1803 if (N_MEMORY == N_HIGH_MEMORY)
1804 zone_last = ZONE_MOVABLE;
1806 for (; zt <= zone_last; zt++)
1807 present_pages += pgdat->node_zones[zt].present_pages;
1808 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1809 arg->status_change_nid_high = zone_to_nid(zone);
1811 arg->status_change_nid_high = -1;
1813 arg->status_change_nid_high = arg->status_change_nid_normal;
1817 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1819 zone_last = ZONE_MOVABLE;
1822 * check whether node_states[N_HIGH_MEMORY] will be changed
1823 * If we try to offline the last present @nr_pages from the node,
1824 * we can determind we will need to clear the node from
1825 * node_states[N_HIGH_MEMORY].
1827 for (; zt <= zone_last; zt++)
1828 present_pages += pgdat->node_zones[zt].present_pages;
1829 if (nr_pages >= present_pages)
1830 arg->status_change_nid = zone_to_nid(zone);
1832 arg->status_change_nid = -1;
1835 static void node_states_clear_node(int node, struct memory_notify *arg)
1837 if (arg->status_change_nid_normal >= 0)
1838 node_clear_state(node, N_NORMAL_MEMORY);
1840 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1841 (arg->status_change_nid_high >= 0))
1842 node_clear_state(node, N_HIGH_MEMORY);
1844 if ((N_MEMORY != N_HIGH_MEMORY) &&
1845 (arg->status_change_nid >= 0))
1846 node_clear_state(node, N_MEMORY);
1849 static int __ref __offline_pages(unsigned long start_pfn,
1850 unsigned long end_pfn, unsigned long timeout)
1852 unsigned long pfn, nr_pages, expire;
1853 long offlined_pages;
1854 int ret, drain, retry_max, node;
1855 unsigned long flags;
1856 unsigned long valid_start, valid_end;
1858 struct memory_notify arg;
1860 /* at least, alignment against pageblock is necessary */
1861 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1863 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1865 /* This makes hotplug much easier...and readable.
1866 we assume this for now. .*/
1867 if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, &valid_end))
1870 zone = page_zone(pfn_to_page(valid_start));
1871 node = zone_to_nid(zone);
1872 nr_pages = end_pfn - start_pfn;
1874 if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1877 /* set above range as isolated */
1878 ret = start_isolate_page_range(start_pfn, end_pfn,
1879 MIGRATE_MOVABLE, true);
1883 arg.start_pfn = start_pfn;
1884 arg.nr_pages = nr_pages;
1885 node_states_check_changes_offline(nr_pages, zone, &arg);
1887 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1888 ret = notifier_to_errno(ret);
1890 goto failed_removal;
1893 expire = jiffies + timeout;
1897 /* start memory hot removal */
1899 if (time_after(jiffies, expire))
1900 goto failed_removal;
1902 if (signal_pending(current))
1903 goto failed_removal;
1906 lru_add_drain_all();
1908 drain_all_pages(zone);
1911 pfn = scan_movable_pages(start_pfn, end_pfn);
1912 if (pfn) { /* We have movable pages */
1913 ret = do_migrate_range(pfn, end_pfn);
1919 if (--retry_max == 0)
1920 goto failed_removal;
1926 /* drain all zone's lru pagevec, this is asynchronous... */
1927 lru_add_drain_all();
1929 /* drain pcp pages, this is synchronous. */
1930 drain_all_pages(zone);
1932 * dissolve free hugepages in the memory block before doing offlining
1933 * actually in order to make hugetlbfs's object counting consistent.
1935 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1937 goto failed_removal;
1939 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1940 if (offlined_pages < 0) {
1942 goto failed_removal;
1944 pr_info("Offlined Pages %ld\n", offlined_pages);
1945 /* Ok, all of our target is isolated.
1946 We cannot do rollback at this point. */
1947 offline_isolated_pages(start_pfn, end_pfn);
1948 /* reset pagetype flags and makes migrate type to be MOVABLE */
1949 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1950 /* removal success */
1951 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1952 zone->present_pages -= offlined_pages;
1954 pgdat_resize_lock(zone->zone_pgdat, &flags);
1955 zone->zone_pgdat->node_present_pages -= offlined_pages;
1956 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1958 init_per_zone_wmark_min();
1960 if (!populated_zone(zone)) {
1961 zone_pcp_reset(zone);
1962 mutex_lock(&zonelists_mutex);
1963 build_all_zonelists(NULL, NULL);
1964 mutex_unlock(&zonelists_mutex);
1966 zone_pcp_update(zone);
1968 node_states_clear_node(node, &arg);
1969 if (arg.status_change_nid >= 0) {
1971 kcompactd_stop(node);
1974 vm_total_pages = nr_free_pagecache_pages();
1975 writeback_set_ratelimit();
1977 memory_notify(MEM_OFFLINE, &arg);
1981 pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n",
1982 (unsigned long long) start_pfn << PAGE_SHIFT,
1983 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1984 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1985 /* pushback to free area */
1986 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1990 /* Must be protected by mem_hotplug_begin() */
1991 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1993 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1995 #endif /* CONFIG_MEMORY_HOTREMOVE */
1998 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1999 * @start_pfn: start pfn of the memory range
2000 * @end_pfn: end pfn of the memory range
2001 * @arg: argument passed to func
2002 * @func: callback for each memory section walked
2004 * This function walks through all present mem sections in range
2005 * [start_pfn, end_pfn) and call func on each mem section.
2007 * Returns the return value of func.
2009 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
2010 void *arg, int (*func)(struct memory_block *, void *))
2012 struct memory_block *mem = NULL;
2013 struct mem_section *section;
2014 unsigned long pfn, section_nr;
2017 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2018 section_nr = pfn_to_section_nr(pfn);
2019 if (!present_section_nr(section_nr))
2022 section = __nr_to_section(section_nr);
2023 /* same memblock? */
2025 if ((section_nr >= mem->start_section_nr) &&
2026 (section_nr <= mem->end_section_nr))
2029 mem = find_memory_block_hinted(section, mem);
2033 ret = func(mem, arg);
2035 kobject_put(&mem->dev.kobj);
2041 kobject_put(&mem->dev.kobj);
2046 #ifdef CONFIG_MEMORY_HOTREMOVE
2047 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
2049 int ret = !is_memblock_offlined(mem);
2051 if (unlikely(ret)) {
2052 phys_addr_t beginpa, endpa;
2054 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
2055 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
2056 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
2063 static int check_cpu_on_node(pg_data_t *pgdat)
2067 for_each_present_cpu(cpu) {
2068 if (cpu_to_node(cpu) == pgdat->node_id)
2070 * the cpu on this node isn't removed, and we can't
2071 * offline this node.
2079 static void unmap_cpu_on_node(pg_data_t *pgdat)
2081 #ifdef CONFIG_ACPI_NUMA
2084 for_each_possible_cpu(cpu)
2085 if (cpu_to_node(cpu) == pgdat->node_id)
2086 numa_clear_node(cpu);
2090 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
2094 ret = check_cpu_on_node(pgdat);
2099 * the node will be offlined when we come here, so we can clear
2100 * the cpu_to_node() now.
2103 unmap_cpu_on_node(pgdat);
2110 * Offline a node if all memory sections and cpus of the node are removed.
2112 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2113 * and online/offline operations before this call.
2115 void try_offline_node(int nid)
2117 pg_data_t *pgdat = NODE_DATA(nid);
2118 unsigned long start_pfn = pgdat->node_start_pfn;
2119 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
2122 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2123 unsigned long section_nr = pfn_to_section_nr(pfn);
2125 if (!present_section_nr(section_nr))
2128 if (pfn_to_nid(pfn) != nid)
2132 * some memory sections of this node are not removed, and we
2133 * can't offline node now.
2138 if (check_and_unmap_cpu_on_node(pgdat))
2142 * all memory/cpu of this node are removed, we can offline this
2145 node_set_offline(nid);
2146 unregister_one_node(nid);
2148 EXPORT_SYMBOL(try_offline_node);
2153 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2154 * and online/offline operations before this call, as required by
2155 * try_offline_node().
2157 void __ref remove_memory(int nid, u64 start, u64 size)
2161 BUG_ON(check_hotplug_memory_range(start, size));
2163 mem_hotplug_begin();
2166 * All memory blocks must be offlined before removing memory. Check
2167 * whether all memory blocks in question are offline and trigger a BUG()
2168 * if this is not the case.
2170 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
2171 check_memblock_offlined_cb);
2175 /* remove memmap entry */
2176 firmware_map_remove(start, start + size, "System RAM");
2177 memblock_free(start, size);
2178 memblock_remove(start, size);
2180 arch_remove_memory(start, size);
2182 try_offline_node(nid);
2186 EXPORT_SYMBOL_GPL(remove_memory);
2187 #endif /* CONFIG_MEMORY_HOTREMOVE */