1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/mm/memory_hotplug.c
8 #include <linux/stddef.h>
10 #include <linux/sched/signal.h>
11 #include <linux/swap.h>
12 #include <linux/interrupt.h>
13 #include <linux/pagemap.h>
14 #include <linux/compiler.h>
15 #include <linux/export.h>
16 #include <linux/pagevec.h>
17 #include <linux/writeback.h>
18 #include <linux/slab.h>
19 #include <linux/sysctl.h>
20 #include <linux/cpu.h>
21 #include <linux/memory.h>
22 #include <linux/memremap.h>
23 #include <linux/memory_hotplug.h>
24 #include <linux/highmem.h>
25 #include <linux/vmalloc.h>
26 #include <linux/ioport.h>
27 #include <linux/delay.h>
28 #include <linux/migrate.h>
29 #include <linux/page-isolation.h>
30 #include <linux/pfn.h>
31 #include <linux/suspend.h>
32 #include <linux/mm_inline.h>
33 #include <linux/firmware-map.h>
34 #include <linux/stop_machine.h>
35 #include <linux/hugetlb.h>
36 #include <linux/memblock.h>
37 #include <linux/compaction.h>
38 #include <linux/rmap.h>
40 #include <asm/tlbflush.h>
46 * online_page_callback contains pointer to current page onlining function.
47 * Initially it is generic_online_page(). If it is required it could be
48 * changed by calling set_online_page_callback() for callback registration
49 * and restore_online_page_callback() for generic callback restore.
52 static void generic_online_page(struct page *page, unsigned int order);
54 static online_page_callback_t online_page_callback = generic_online_page;
55 static DEFINE_MUTEX(online_page_callback_lock);
57 DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock);
59 void get_online_mems(void)
61 percpu_down_read(&mem_hotplug_lock);
64 void put_online_mems(void)
66 percpu_up_read(&mem_hotplug_lock);
69 bool movable_node_enabled = false;
71 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
72 bool memhp_auto_online;
74 bool memhp_auto_online = true;
76 EXPORT_SYMBOL_GPL(memhp_auto_online);
78 static int __init setup_memhp_default_state(char *str)
80 if (!strcmp(str, "online"))
81 memhp_auto_online = true;
82 else if (!strcmp(str, "offline"))
83 memhp_auto_online = false;
87 __setup("memhp_default_state=", setup_memhp_default_state);
89 void mem_hotplug_begin(void)
92 percpu_down_write(&mem_hotplug_lock);
95 void mem_hotplug_done(void)
97 percpu_up_write(&mem_hotplug_lock);
101 u64 max_mem_size = U64_MAX;
103 /* add this memory to iomem resource */
104 static struct resource *register_memory_resource(u64 start, u64 size)
106 struct resource *res;
107 unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
108 char *resource_name = "System RAM";
110 if (start + size > max_mem_size)
111 return ERR_PTR(-E2BIG);
114 * Request ownership of the new memory range. This might be
115 * a child of an existing resource that was present but
116 * not marked as busy.
118 res = __request_region(&iomem_resource, start, size,
119 resource_name, flags);
122 pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n",
123 start, start + size);
124 return ERR_PTR(-EEXIST);
129 static void release_memory_resource(struct resource *res)
133 release_resource(res);
138 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
139 void get_page_bootmem(unsigned long info, struct page *page,
142 page->freelist = (void *)type;
143 SetPagePrivate(page);
144 set_page_private(page, info);
148 void put_page_bootmem(struct page *page)
152 type = (unsigned long) page->freelist;
153 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
154 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
156 if (page_ref_dec_return(page) == 1) {
157 page->freelist = NULL;
158 ClearPagePrivate(page);
159 set_page_private(page, 0);
160 INIT_LIST_HEAD(&page->lru);
161 free_reserved_page(page);
165 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
166 #ifndef CONFIG_SPARSEMEM_VMEMMAP
167 static void register_page_bootmem_info_section(unsigned long start_pfn)
169 unsigned long *usemap, mapsize, section_nr, i;
170 struct mem_section *ms;
171 struct page *page, *memmap;
173 section_nr = pfn_to_section_nr(start_pfn);
174 ms = __nr_to_section(section_nr);
176 /* Get section's memmap address */
177 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
180 * Get page for the memmap's phys address
181 * XXX: need more consideration for sparse_vmemmap...
183 page = virt_to_page(memmap);
184 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
185 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
187 /* remember memmap's page */
188 for (i = 0; i < mapsize; i++, page++)
189 get_page_bootmem(section_nr, page, SECTION_INFO);
191 usemap = ms->pageblock_flags;
192 page = virt_to_page(usemap);
194 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
196 for (i = 0; i < mapsize; i++, page++)
197 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
200 #else /* CONFIG_SPARSEMEM_VMEMMAP */
201 static void register_page_bootmem_info_section(unsigned long start_pfn)
203 unsigned long *usemap, mapsize, section_nr, i;
204 struct mem_section *ms;
205 struct page *page, *memmap;
207 section_nr = pfn_to_section_nr(start_pfn);
208 ms = __nr_to_section(section_nr);
210 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
212 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
214 usemap = ms->pageblock_flags;
215 page = virt_to_page(usemap);
217 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
219 for (i = 0; i < mapsize; i++, page++)
220 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
222 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
224 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
226 unsigned long i, pfn, end_pfn, nr_pages;
227 int node = pgdat->node_id;
230 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
231 page = virt_to_page(pgdat);
233 for (i = 0; i < nr_pages; i++, page++)
234 get_page_bootmem(node, page, NODE_INFO);
236 pfn = pgdat->node_start_pfn;
237 end_pfn = pgdat_end_pfn(pgdat);
239 /* register section info */
240 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
242 * Some platforms can assign the same pfn to multiple nodes - on
243 * node0 as well as nodeN. To avoid registering a pfn against
244 * multiple nodes we check that this pfn does not already
245 * reside in some other nodes.
247 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
248 register_page_bootmem_info_section(pfn);
251 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
253 static int __meminit __add_section(int nid, unsigned long phys_start_pfn,
254 struct vmem_altmap *altmap, bool want_memblock)
258 if (pfn_valid(phys_start_pfn))
261 ret = sparse_add_one_section(nid, phys_start_pfn, altmap);
262 return ret < 0 ? ret : 0;
266 * Reasonably generic function for adding memory. It is
267 * expected that archs that support memory hotplug will
268 * call this function after deciding the zone to which to
271 int __ref __add_pages(int nid, unsigned long phys_start_pfn,
272 unsigned long nr_pages, struct mhp_restrictions *restrictions)
276 int start_sec, end_sec;
277 struct vmem_altmap *altmap = restrictions->altmap;
279 /* during initialize mem_map, align hot-added range to section */
280 start_sec = pfn_to_section_nr(phys_start_pfn);
281 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
285 * Validate altmap is within bounds of the total request
287 if (altmap->base_pfn != phys_start_pfn
288 || vmem_altmap_offset(altmap) > nr_pages) {
289 pr_warn_once("memory add fail, invalid altmap\n");
296 for (i = start_sec; i <= end_sec; i++) {
297 err = __add_section(nid, section_nr_to_pfn(i), altmap,
298 restrictions->flags & MHP_MEMBLOCK_API);
301 * EEXIST is finally dealt with by ioresource collision
302 * check. see add_memory() => register_memory_resource()
303 * Warning will be printed if there is collision.
305 if (err && (err != -EEXIST))
310 vmemmap_populate_print_last();
315 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
316 static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
317 unsigned long start_pfn,
318 unsigned long end_pfn)
320 struct mem_section *ms;
322 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
323 ms = __pfn_to_section(start_pfn);
325 if (unlikely(!valid_section(ms)))
328 if (unlikely(pfn_to_nid(start_pfn) != nid))
331 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
340 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
341 static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
342 unsigned long start_pfn,
343 unsigned long end_pfn)
345 struct mem_section *ms;
348 /* pfn is the end pfn of a memory section. */
350 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
351 ms = __pfn_to_section(pfn);
353 if (unlikely(!valid_section(ms)))
356 if (unlikely(pfn_to_nid(pfn) != nid))
359 if (zone && zone != page_zone(pfn_to_page(pfn)))
368 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
369 unsigned long end_pfn)
371 unsigned long zone_start_pfn = zone->zone_start_pfn;
372 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
373 unsigned long zone_end_pfn = z;
375 struct mem_section *ms;
376 int nid = zone_to_nid(zone);
378 zone_span_writelock(zone);
379 if (zone_start_pfn == start_pfn) {
381 * If the section is smallest section in the zone, it need
382 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
383 * In this case, we find second smallest valid mem_section
384 * for shrinking zone.
386 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
389 zone->zone_start_pfn = pfn;
390 zone->spanned_pages = zone_end_pfn - pfn;
392 } else if (zone_end_pfn == end_pfn) {
394 * If the section is biggest section in the zone, it need
395 * shrink zone->spanned_pages.
396 * In this case, we find second biggest valid mem_section for
399 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
402 zone->spanned_pages = pfn - zone_start_pfn + 1;
406 * The section is not biggest or smallest mem_section in the zone, it
407 * only creates a hole in the zone. So in this case, we need not
408 * change the zone. But perhaps, the zone has only hole data. Thus
409 * it check the zone has only hole or not.
411 pfn = zone_start_pfn;
412 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
413 ms = __pfn_to_section(pfn);
415 if (unlikely(!valid_section(ms)))
418 if (page_zone(pfn_to_page(pfn)) != zone)
421 /* If the section is current section, it continues the loop */
422 if (start_pfn == pfn)
425 /* If we find valid section, we have nothing to do */
426 zone_span_writeunlock(zone);
430 /* The zone has no valid section */
431 zone->zone_start_pfn = 0;
432 zone->spanned_pages = 0;
433 zone_span_writeunlock(zone);
436 static void shrink_pgdat_span(struct pglist_data *pgdat,
437 unsigned long start_pfn, unsigned long end_pfn)
439 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
440 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
441 unsigned long pgdat_end_pfn = p;
443 struct mem_section *ms;
444 int nid = pgdat->node_id;
446 if (pgdat_start_pfn == start_pfn) {
448 * If the section is smallest section in the pgdat, it need
449 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
450 * In this case, we find second smallest valid mem_section
451 * for shrinking zone.
453 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
456 pgdat->node_start_pfn = pfn;
457 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
459 } else if (pgdat_end_pfn == end_pfn) {
461 * If the section is biggest section in the pgdat, it need
462 * shrink pgdat->node_spanned_pages.
463 * In this case, we find second biggest valid mem_section for
466 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
469 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
473 * If the section is not biggest or smallest mem_section in the pgdat,
474 * it only creates a hole in the pgdat. So in this case, we need not
476 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
477 * has only hole or not.
479 pfn = pgdat_start_pfn;
480 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
481 ms = __pfn_to_section(pfn);
483 if (unlikely(!valid_section(ms)))
486 if (pfn_to_nid(pfn) != nid)
489 /* If the section is current section, it continues the loop */
490 if (start_pfn == pfn)
493 /* If we find valid section, we have nothing to do */
497 /* The pgdat has no valid section */
498 pgdat->node_start_pfn = 0;
499 pgdat->node_spanned_pages = 0;
502 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
504 struct pglist_data *pgdat = zone->zone_pgdat;
505 int nr_pages = PAGES_PER_SECTION;
508 pgdat_resize_lock(zone->zone_pgdat, &flags);
509 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
510 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
511 pgdat_resize_unlock(zone->zone_pgdat, &flags);
514 static void __remove_section(struct zone *zone, struct mem_section *ms,
515 unsigned long map_offset,
516 struct vmem_altmap *altmap)
518 unsigned long start_pfn;
521 if (WARN_ON_ONCE(!valid_section(ms)))
524 unregister_memory_section(ms);
526 scn_nr = __section_nr(ms);
527 start_pfn = section_nr_to_pfn((unsigned long)scn_nr);
528 __remove_zone(zone, start_pfn);
530 sparse_remove_one_section(zone, ms, map_offset, altmap);
534 * __remove_pages() - remove sections of pages from a zone
535 * @zone: zone from which pages need to be removed
536 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
537 * @nr_pages: number of pages to remove (must be multiple of section size)
538 * @altmap: alternative device page map or %NULL if default memmap is used
540 * Generic helper function to remove section mappings and sysfs entries
541 * for the section of the memory we are removing. Caller needs to make
542 * sure that pages are marked reserved and zones are adjust properly by
543 * calling offline_pages().
545 void __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
546 unsigned long nr_pages, struct vmem_altmap *altmap)
549 unsigned long map_offset = 0;
550 int sections_to_remove;
552 /* In the ZONE_DEVICE case device driver owns the memory region */
553 if (is_dev_zone(zone))
554 map_offset = vmem_altmap_offset(altmap);
556 clear_zone_contiguous(zone);
559 * We can only remove entire sections
561 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
562 BUG_ON(nr_pages % PAGES_PER_SECTION);
564 sections_to_remove = nr_pages / PAGES_PER_SECTION;
565 for (i = 0; i < sections_to_remove; i++) {
566 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
569 __remove_section(zone, __pfn_to_section(pfn), map_offset,
574 set_zone_contiguous(zone);
577 int set_online_page_callback(online_page_callback_t callback)
582 mutex_lock(&online_page_callback_lock);
584 if (online_page_callback == generic_online_page) {
585 online_page_callback = callback;
589 mutex_unlock(&online_page_callback_lock);
594 EXPORT_SYMBOL_GPL(set_online_page_callback);
596 int restore_online_page_callback(online_page_callback_t callback)
601 mutex_lock(&online_page_callback_lock);
603 if (online_page_callback == callback) {
604 online_page_callback = generic_online_page;
608 mutex_unlock(&online_page_callback_lock);
613 EXPORT_SYMBOL_GPL(restore_online_page_callback);
615 void __online_page_set_limits(struct page *page)
618 EXPORT_SYMBOL_GPL(__online_page_set_limits);
620 void __online_page_increment_counters(struct page *page)
622 adjust_managed_page_count(page, 1);
624 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
626 void __online_page_free(struct page *page)
628 __free_reserved_page(page);
630 EXPORT_SYMBOL_GPL(__online_page_free);
632 static void generic_online_page(struct page *page, unsigned int order)
634 kernel_map_pages(page, 1 << order, 1);
635 __free_pages_core(page, order);
636 totalram_pages_add(1UL << order);
637 #ifdef CONFIG_HIGHMEM
638 if (PageHighMem(page))
639 totalhigh_pages_add(1UL << order);
643 static int online_pages_blocks(unsigned long start, unsigned long nr_pages)
645 unsigned long end = start + nr_pages;
646 int order, onlined_pages = 0;
648 while (start < end) {
649 order = min(MAX_ORDER - 1,
650 get_order(PFN_PHYS(end) - PFN_PHYS(start)));
651 (*online_page_callback)(pfn_to_page(start), order);
653 onlined_pages += (1UL << order);
654 start += (1UL << order);
656 return onlined_pages;
659 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
662 unsigned long onlined_pages = *(unsigned long *)arg;
664 if (PageReserved(pfn_to_page(start_pfn)))
665 onlined_pages += online_pages_blocks(start_pfn, nr_pages);
667 online_mem_sections(start_pfn, start_pfn + nr_pages);
669 *(unsigned long *)arg = onlined_pages;
673 /* check which state of node_states will be changed when online memory */
674 static void node_states_check_changes_online(unsigned long nr_pages,
675 struct zone *zone, struct memory_notify *arg)
677 int nid = zone_to_nid(zone);
679 arg->status_change_nid = NUMA_NO_NODE;
680 arg->status_change_nid_normal = NUMA_NO_NODE;
681 arg->status_change_nid_high = NUMA_NO_NODE;
683 if (!node_state(nid, N_MEMORY))
684 arg->status_change_nid = nid;
685 if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
686 arg->status_change_nid_normal = nid;
687 #ifdef CONFIG_HIGHMEM
688 if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY))
689 arg->status_change_nid_high = nid;
693 static void node_states_set_node(int node, struct memory_notify *arg)
695 if (arg->status_change_nid_normal >= 0)
696 node_set_state(node, N_NORMAL_MEMORY);
698 if (arg->status_change_nid_high >= 0)
699 node_set_state(node, N_HIGH_MEMORY);
701 if (arg->status_change_nid >= 0)
702 node_set_state(node, N_MEMORY);
705 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
706 unsigned long nr_pages)
708 unsigned long old_end_pfn = zone_end_pfn(zone);
710 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
711 zone->zone_start_pfn = start_pfn;
713 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
716 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
717 unsigned long nr_pages)
719 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
721 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
722 pgdat->node_start_pfn = start_pfn;
724 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
727 void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
728 unsigned long nr_pages, struct vmem_altmap *altmap)
730 struct pglist_data *pgdat = zone->zone_pgdat;
731 int nid = pgdat->node_id;
734 clear_zone_contiguous(zone);
736 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
737 pgdat_resize_lock(pgdat, &flags);
738 zone_span_writelock(zone);
739 if (zone_is_empty(zone))
740 init_currently_empty_zone(zone, start_pfn, nr_pages);
741 resize_zone_range(zone, start_pfn, nr_pages);
742 zone_span_writeunlock(zone);
743 resize_pgdat_range(pgdat, start_pfn, nr_pages);
744 pgdat_resize_unlock(pgdat, &flags);
747 * TODO now we have a visible range of pages which are not associated
748 * with their zone properly. Not nice but set_pfnblock_flags_mask
749 * expects the zone spans the pfn range. All the pages in the range
750 * are reserved so nobody should be touching them so we should be safe
752 memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
753 MEMMAP_HOTPLUG, altmap);
755 set_zone_contiguous(zone);
759 * Returns a default kernel memory zone for the given pfn range.
760 * If no kernel zone covers this pfn range it will automatically go
761 * to the ZONE_NORMAL.
763 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
764 unsigned long nr_pages)
766 struct pglist_data *pgdat = NODE_DATA(nid);
769 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
770 struct zone *zone = &pgdat->node_zones[zid];
772 if (zone_intersects(zone, start_pfn, nr_pages))
776 return &pgdat->node_zones[ZONE_NORMAL];
779 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
780 unsigned long nr_pages)
782 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
784 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
785 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
786 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
789 * We inherit the existing zone in a simple case where zones do not
790 * overlap in the given range
792 if (in_kernel ^ in_movable)
793 return (in_kernel) ? kernel_zone : movable_zone;
796 * If the range doesn't belong to any zone or two zones overlap in the
797 * given range then we use movable zone only if movable_node is
798 * enabled because we always online to a kernel zone by default.
800 return movable_node_enabled ? movable_zone : kernel_zone;
803 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
804 unsigned long nr_pages)
806 if (online_type == MMOP_ONLINE_KERNEL)
807 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
809 if (online_type == MMOP_ONLINE_MOVABLE)
810 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
812 return default_zone_for_pfn(nid, start_pfn, nr_pages);
816 * Associates the given pfn range with the given node and the zone appropriate
817 * for the given online type.
819 static struct zone * __meminit move_pfn_range(int online_type, int nid,
820 unsigned long start_pfn, unsigned long nr_pages)
824 zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages);
825 move_pfn_range_to_zone(zone, start_pfn, nr_pages, NULL);
829 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
832 unsigned long onlined_pages = 0;
834 int need_zonelists_rebuild = 0;
837 struct memory_notify arg;
838 struct memory_block *mem;
843 * We can't use pfn_to_nid() because nid might be stored in struct page
844 * which is not yet initialized. Instead, we find nid from memory block.
846 mem = find_memory_block(__pfn_to_section(pfn));
848 put_device(&mem->dev);
850 /* associate pfn range with the zone */
851 zone = move_pfn_range(online_type, nid, pfn, nr_pages);
854 arg.nr_pages = nr_pages;
855 node_states_check_changes_online(nr_pages, zone, &arg);
857 ret = memory_notify(MEM_GOING_ONLINE, &arg);
858 ret = notifier_to_errno(ret);
860 goto failed_addition;
863 * If this zone is not populated, then it is not in zonelist.
864 * This means the page allocator ignores this zone.
865 * So, zonelist must be updated after online.
867 if (!populated_zone(zone)) {
868 need_zonelists_rebuild = 1;
869 setup_zone_pageset(zone);
872 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
875 if (need_zonelists_rebuild)
876 zone_pcp_reset(zone);
877 goto failed_addition;
880 zone->present_pages += onlined_pages;
882 pgdat_resize_lock(zone->zone_pgdat, &flags);
883 zone->zone_pgdat->node_present_pages += onlined_pages;
884 pgdat_resize_unlock(zone->zone_pgdat, &flags);
889 node_states_set_node(nid, &arg);
890 if (need_zonelists_rebuild)
891 build_all_zonelists(NULL);
893 zone_pcp_update(zone);
896 init_per_zone_wmark_min();
903 vm_total_pages = nr_free_pagecache_pages();
905 writeback_set_ratelimit();
908 memory_notify(MEM_ONLINE, &arg);
913 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
914 (unsigned long long) pfn << PAGE_SHIFT,
915 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
916 memory_notify(MEM_CANCEL_ONLINE, &arg);
920 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
922 static void reset_node_present_pages(pg_data_t *pgdat)
926 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
927 z->present_pages = 0;
929 pgdat->node_present_pages = 0;
932 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
933 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
935 struct pglist_data *pgdat;
936 unsigned long start_pfn = PFN_DOWN(start);
938 pgdat = NODE_DATA(nid);
940 pgdat = arch_alloc_nodedata(nid);
944 arch_refresh_nodedata(nid, pgdat);
947 * Reset the nr_zones, order and classzone_idx before reuse.
948 * Note that kswapd will init kswapd_classzone_idx properly
949 * when it starts in the near future.
952 pgdat->kswapd_order = 0;
953 pgdat->kswapd_classzone_idx = 0;
956 /* we can use NODE_DATA(nid) from here */
958 pgdat->node_id = nid;
959 pgdat->node_start_pfn = start_pfn;
961 /* init node's zones as empty zones, we don't have any present pages.*/
962 free_area_init_core_hotplug(nid);
963 pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
966 * The node we allocated has no zone fallback lists. For avoiding
967 * to access not-initialized zonelist, build here.
969 build_all_zonelists(pgdat);
972 * When memory is hot-added, all the memory is in offline state. So
973 * clear all zones' present_pages because they will be updated in
974 * online_pages() and offline_pages().
976 reset_node_managed_pages(pgdat);
977 reset_node_present_pages(pgdat);
982 static void rollback_node_hotadd(int nid)
984 pg_data_t *pgdat = NODE_DATA(nid);
986 arch_refresh_nodedata(nid, NULL);
987 free_percpu(pgdat->per_cpu_nodestats);
988 arch_free_nodedata(pgdat);
994 * try_online_node - online a node if offlined
996 * @start: start addr of the node
997 * @set_node_online: Whether we want to online the node
998 * called by cpu_up() to online a node without onlined memory.
1001 * 1 -> a new node has been allocated
1002 * 0 -> the node is already online
1003 * -ENOMEM -> the node could not be allocated
1005 static int __try_online_node(int nid, u64 start, bool set_node_online)
1010 if (node_online(nid))
1013 pgdat = hotadd_new_pgdat(nid, start);
1015 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1020 if (set_node_online) {
1021 node_set_online(nid);
1022 ret = register_one_node(nid);
1030 * Users of this function always want to online/register the node
1032 int try_online_node(int nid)
1036 mem_hotplug_begin();
1037 ret = __try_online_node(nid, 0, true);
1042 static int check_hotplug_memory_range(u64 start, u64 size)
1044 /* memory range must be block size aligned */
1045 if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) ||
1046 !IS_ALIGNED(size, memory_block_size_bytes())) {
1047 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1048 memory_block_size_bytes(), start, size);
1055 static int online_memory_block(struct memory_block *mem, void *arg)
1057 return device_online(&mem->dev);
1061 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1062 * and online/offline operations (triggered e.g. by sysfs).
1064 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1066 int __ref add_memory_resource(int nid, struct resource *res)
1068 struct mhp_restrictions restrictions = {
1069 .flags = MHP_MEMBLOCK_API,
1072 bool new_node = false;
1076 size = resource_size(res);
1078 ret = check_hotplug_memory_range(start, size);
1082 mem_hotplug_begin();
1085 * Add new range to memblock so that when hotadd_new_pgdat() is called
1086 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1087 * this new range and calculate total pages correctly. The range will
1088 * be removed at hot-remove time.
1090 memblock_add_node(start, size, nid);
1092 ret = __try_online_node(nid, start, false);
1097 /* call arch's memory hotadd */
1098 ret = arch_add_memory(nid, start, size, &restrictions);
1102 /* create memory block devices after memory was added */
1103 ret = create_memory_block_devices(start, size);
1105 arch_remove_memory(nid, start, size, NULL);
1110 /* If sysfs file of new node can't be created, cpu on the node
1111 * can't be hot-added. There is no rollback way now.
1112 * So, check by BUG_ON() to catch it reluctantly..
1113 * We online node here. We can't roll back from here.
1115 node_set_online(nid);
1116 ret = __register_one_node(nid);
1120 /* link memory sections under this node.*/
1121 ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1));
1124 /* create new memmap entry */
1125 firmware_map_add_hotplug(start, start + size, "System RAM");
1127 /* device_online() will take the lock when calling online_pages() */
1130 /* online pages if requested */
1131 if (memhp_auto_online)
1132 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1133 NULL, online_memory_block);
1137 /* rollback pgdat allocation and others */
1139 rollback_node_hotadd(nid);
1140 memblock_remove(start, size);
1145 /* requires device_hotplug_lock, see add_memory_resource() */
1146 int __ref __add_memory(int nid, u64 start, u64 size)
1148 struct resource *res;
1151 res = register_memory_resource(start, size);
1153 return PTR_ERR(res);
1155 ret = add_memory_resource(nid, res);
1157 release_memory_resource(res);
1161 int add_memory(int nid, u64 start, u64 size)
1165 lock_device_hotplug();
1166 rc = __add_memory(nid, start, size);
1167 unlock_device_hotplug();
1171 EXPORT_SYMBOL_GPL(add_memory);
1173 #ifdef CONFIG_MEMORY_HOTREMOVE
1175 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1176 * set and the size of the free page is given by page_order(). Using this,
1177 * the function determines if the pageblock contains only free pages.
1178 * Due to buddy contraints, a free page at least the size of a pageblock will
1179 * be located at the start of the pageblock
1181 static inline int pageblock_free(struct page *page)
1183 return PageBuddy(page) && page_order(page) >= pageblock_order;
1186 /* Return the pfn of the start of the next active pageblock after a given pfn */
1187 static unsigned long next_active_pageblock(unsigned long pfn)
1189 struct page *page = pfn_to_page(pfn);
1191 /* Ensure the starting page is pageblock-aligned */
1192 BUG_ON(pfn & (pageblock_nr_pages - 1));
1194 /* If the entire pageblock is free, move to the end of free page */
1195 if (pageblock_free(page)) {
1197 /* be careful. we don't have locks, page_order can be changed.*/
1198 order = page_order(page);
1199 if ((order < MAX_ORDER) && (order >= pageblock_order))
1200 return pfn + (1 << order);
1203 return pfn + pageblock_nr_pages;
1206 static bool is_pageblock_removable_nolock(unsigned long pfn)
1208 struct page *page = pfn_to_page(pfn);
1212 * We have to be careful here because we are iterating over memory
1213 * sections which are not zone aware so we might end up outside of
1214 * the zone but still within the section.
1215 * We have to take care about the node as well. If the node is offline
1216 * its NODE_DATA will be NULL - see page_zone.
1218 if (!node_online(page_to_nid(page)))
1221 zone = page_zone(page);
1222 pfn = page_to_pfn(page);
1223 if (!zone_spans_pfn(zone, pfn))
1226 return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE, SKIP_HWPOISON);
1229 /* Checks if this range of memory is likely to be hot-removable. */
1230 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1232 unsigned long end_pfn, pfn;
1234 end_pfn = min(start_pfn + nr_pages,
1235 zone_end_pfn(page_zone(pfn_to_page(start_pfn))));
1237 /* Check the starting page of each pageblock within the range */
1238 for (pfn = start_pfn; pfn < end_pfn; pfn = next_active_pageblock(pfn)) {
1239 if (!is_pageblock_removable_nolock(pfn))
1244 /* All pageblocks in the memory block are likely to be hot-removable */
1249 * Confirm all pages in a range [start, end) belong to the same zone.
1250 * When true, return its valid [start, end).
1252 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1253 unsigned long *valid_start, unsigned long *valid_end)
1255 unsigned long pfn, sec_end_pfn;
1256 unsigned long start, end;
1257 struct zone *zone = NULL;
1260 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1262 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1263 /* Make sure the memory section is present first */
1264 if (!present_section_nr(pfn_to_section_nr(pfn)))
1266 for (; pfn < sec_end_pfn && pfn < end_pfn;
1267 pfn += MAX_ORDER_NR_PAGES) {
1269 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1270 while ((i < MAX_ORDER_NR_PAGES) &&
1271 !pfn_valid_within(pfn + i))
1273 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1275 /* Check if we got outside of the zone */
1276 if (zone && !zone_spans_pfn(zone, pfn + i))
1278 page = pfn_to_page(pfn + i);
1279 if (zone && page_zone(page) != zone)
1283 zone = page_zone(page);
1284 end = pfn + MAX_ORDER_NR_PAGES;
1289 *valid_start = start;
1290 *valid_end = min(end, end_pfn);
1298 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1299 * non-lru movable pages and hugepages). We scan pfn because it's much
1300 * easier than scanning over linked list. This function returns the pfn
1301 * of the first found movable page if it's found, otherwise 0.
1303 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1307 for (pfn = start; pfn < end; pfn++) {
1308 struct page *page, *head;
1311 if (!pfn_valid(pfn))
1313 page = pfn_to_page(pfn);
1316 if (__PageMovable(page))
1319 if (!PageHuge(page))
1321 head = compound_head(page);
1322 if (page_huge_active(head))
1324 skip = (1 << compound_order(head)) - (page - head);
1330 static struct page *new_node_page(struct page *page, unsigned long private)
1332 int nid = page_to_nid(page);
1333 nodemask_t nmask = node_states[N_MEMORY];
1336 * try to allocate from a different node but reuse this node if there
1337 * are no other online nodes to be used (e.g. we are offlining a part
1338 * of the only existing node)
1340 node_clear(nid, nmask);
1341 if (nodes_empty(nmask))
1342 node_set(nid, nmask);
1344 return new_page_nodemask(page, nid, &nmask);
1348 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1355 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1356 if (!pfn_valid(pfn))
1358 page = pfn_to_page(pfn);
1360 if (PageHuge(page)) {
1361 struct page *head = compound_head(page);
1362 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1363 isolate_huge_page(head, &source);
1365 } else if (PageTransHuge(page))
1366 pfn = page_to_pfn(compound_head(page))
1367 + hpage_nr_pages(page) - 1;
1370 * HWPoison pages have elevated reference counts so the migration would
1371 * fail on them. It also doesn't make any sense to migrate them in the
1372 * first place. Still try to unmap such a page in case it is still mapped
1373 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1374 * the unmap as the catch all safety net).
1376 if (PageHWPoison(page)) {
1377 if (WARN_ON(PageLRU(page)))
1378 isolate_lru_page(page);
1379 if (page_mapped(page))
1380 try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1384 if (!get_page_unless_zero(page))
1387 * We can skip free pages. And we can deal with pages on
1388 * LRU and non-lru movable pages.
1391 ret = isolate_lru_page(page);
1393 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1394 if (!ret) { /* Success */
1395 list_add_tail(&page->lru, &source);
1396 if (!__PageMovable(page))
1397 inc_node_page_state(page, NR_ISOLATED_ANON +
1398 page_is_file_cache(page));
1401 pr_warn("failed to isolate pfn %lx\n", pfn);
1402 dump_page(page, "isolation failed");
1406 if (!list_empty(&source)) {
1407 /* Allocate a new page from the nearest neighbor node */
1408 ret = migrate_pages(&source, new_node_page, NULL, 0,
1409 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1411 list_for_each_entry(page, &source, lru) {
1412 pr_warn("migrating pfn %lx failed ret:%d ",
1413 page_to_pfn(page), ret);
1414 dump_page(page, "migration failure");
1416 putback_movable_pages(&source);
1424 * remove from free_area[] and mark all as Reserved.
1427 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1430 unsigned long *offlined_pages = (unsigned long *)data;
1432 *offlined_pages += __offline_isolated_pages(start, start + nr_pages);
1437 * Check all pages in range, recoreded as memory resource, are isolated.
1440 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1443 return test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1446 static int __init cmdline_parse_movable_node(char *p)
1448 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1449 movable_node_enabled = true;
1451 pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1455 early_param("movable_node", cmdline_parse_movable_node);
1457 /* check which state of node_states will be changed when offline memory */
1458 static void node_states_check_changes_offline(unsigned long nr_pages,
1459 struct zone *zone, struct memory_notify *arg)
1461 struct pglist_data *pgdat = zone->zone_pgdat;
1462 unsigned long present_pages = 0;
1465 arg->status_change_nid = NUMA_NO_NODE;
1466 arg->status_change_nid_normal = NUMA_NO_NODE;
1467 arg->status_change_nid_high = NUMA_NO_NODE;
1470 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1471 * If the memory to be offline is within the range
1472 * [0..ZONE_NORMAL], and it is the last present memory there,
1473 * the zones in that range will become empty after the offlining,
1474 * thus we can determine that we need to clear the node from
1475 * node_states[N_NORMAL_MEMORY].
1477 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1478 present_pages += pgdat->node_zones[zt].present_pages;
1479 if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
1480 arg->status_change_nid_normal = zone_to_nid(zone);
1482 #ifdef CONFIG_HIGHMEM
1484 * node_states[N_HIGH_MEMORY] contains nodes which
1485 * have normal memory or high memory.
1486 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1487 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1488 * we determine that the zones in that range become empty,
1489 * we need to clear the node for N_HIGH_MEMORY.
1491 present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1492 if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
1493 arg->status_change_nid_high = zone_to_nid(zone);
1497 * We have accounted the pages from [0..ZONE_NORMAL), and
1498 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1500 * Here we count the possible pages from ZONE_MOVABLE.
1501 * If after having accounted all the pages, we see that the nr_pages
1502 * to be offlined is over or equal to the accounted pages,
1503 * we know that the node will become empty, and so, we can clear
1504 * it for N_MEMORY as well.
1506 present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
1508 if (nr_pages >= present_pages)
1509 arg->status_change_nid = zone_to_nid(zone);
1512 static void node_states_clear_node(int node, struct memory_notify *arg)
1514 if (arg->status_change_nid_normal >= 0)
1515 node_clear_state(node, N_NORMAL_MEMORY);
1517 if (arg->status_change_nid_high >= 0)
1518 node_clear_state(node, N_HIGH_MEMORY);
1520 if (arg->status_change_nid >= 0)
1521 node_clear_state(node, N_MEMORY);
1524 static int __ref __offline_pages(unsigned long start_pfn,
1525 unsigned long end_pfn)
1527 unsigned long pfn, nr_pages;
1528 unsigned long offlined_pages = 0;
1529 int ret, node, nr_isolate_pageblock;
1530 unsigned long flags;
1531 unsigned long valid_start, valid_end;
1533 struct memory_notify arg;
1536 mem_hotplug_begin();
1538 /* This makes hotplug much easier...and readable.
1539 we assume this for now. .*/
1540 if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start,
1543 reason = "multizone range";
1544 goto failed_removal;
1547 zone = page_zone(pfn_to_page(valid_start));
1548 node = zone_to_nid(zone);
1549 nr_pages = end_pfn - start_pfn;
1551 /* set above range as isolated */
1552 ret = start_isolate_page_range(start_pfn, end_pfn,
1554 SKIP_HWPOISON | REPORT_FAILURE);
1556 reason = "failure to isolate range";
1557 goto failed_removal;
1559 nr_isolate_pageblock = ret;
1561 arg.start_pfn = start_pfn;
1562 arg.nr_pages = nr_pages;
1563 node_states_check_changes_offline(nr_pages, zone, &arg);
1565 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1566 ret = notifier_to_errno(ret);
1568 reason = "notifier failure";
1569 goto failed_removal_isolated;
1573 for (pfn = start_pfn; pfn;) {
1574 if (signal_pending(current)) {
1576 reason = "signal backoff";
1577 goto failed_removal_isolated;
1581 lru_add_drain_all();
1583 pfn = scan_movable_pages(pfn, end_pfn);
1586 * TODO: fatal migration failures should bail
1589 do_migrate_range(pfn, end_pfn);
1594 * Dissolve free hugepages in the memory block before doing
1595 * offlining actually in order to make hugetlbfs's object
1596 * counting consistent.
1598 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1600 reason = "failure to dissolve huge pages";
1601 goto failed_removal_isolated;
1604 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1605 NULL, check_pages_isolated_cb);
1608 /* Ok, all of our target is isolated.
1609 We cannot do rollback at this point. */
1610 walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1611 &offlined_pages, offline_isolated_pages_cb);
1612 pr_info("Offlined Pages %ld\n", offlined_pages);
1614 * Onlining will reset pagetype flags and makes migrate type
1615 * MOVABLE, so just need to decrease the number of isolated
1616 * pageblocks zone counter here.
1618 spin_lock_irqsave(&zone->lock, flags);
1619 zone->nr_isolate_pageblock -= nr_isolate_pageblock;
1620 spin_unlock_irqrestore(&zone->lock, flags);
1622 /* removal success */
1623 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1624 zone->present_pages -= offlined_pages;
1626 pgdat_resize_lock(zone->zone_pgdat, &flags);
1627 zone->zone_pgdat->node_present_pages -= offlined_pages;
1628 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1630 init_per_zone_wmark_min();
1632 if (!populated_zone(zone)) {
1633 zone_pcp_reset(zone);
1634 build_all_zonelists(NULL);
1636 zone_pcp_update(zone);
1638 node_states_clear_node(node, &arg);
1639 if (arg.status_change_nid >= 0) {
1641 kcompactd_stop(node);
1644 vm_total_pages = nr_free_pagecache_pages();
1645 writeback_set_ratelimit();
1647 memory_notify(MEM_OFFLINE, &arg);
1651 failed_removal_isolated:
1652 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1653 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1655 pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1656 (unsigned long long) start_pfn << PAGE_SHIFT,
1657 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1659 /* pushback to free area */
1664 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1666 return __offline_pages(start_pfn, start_pfn + nr_pages);
1668 #endif /* CONFIG_MEMORY_HOTREMOVE */
1671 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1672 * @start_pfn: start pfn of the memory range
1673 * @end_pfn: end pfn of the memory range
1674 * @arg: argument passed to func
1675 * @func: callback for each memory section walked
1677 * This function walks through all present mem sections in range
1678 * [start_pfn, end_pfn) and call func on each mem section.
1680 * Returns the return value of func.
1682 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1683 void *arg, int (*func)(struct memory_block *, void *))
1685 struct memory_block *mem = NULL;
1686 struct mem_section *section;
1687 unsigned long pfn, section_nr;
1690 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1691 section_nr = pfn_to_section_nr(pfn);
1692 if (!present_section_nr(section_nr))
1695 section = __nr_to_section(section_nr);
1696 /* same memblock? */
1698 if ((section_nr >= mem->start_section_nr) &&
1699 (section_nr <= mem->end_section_nr))
1702 mem = find_memory_block_hinted(section, mem);
1706 ret = func(mem, arg);
1708 kobject_put(&mem->dev.kobj);
1714 kobject_put(&mem->dev.kobj);
1719 #ifdef CONFIG_MEMORY_HOTREMOVE
1720 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1722 int ret = !is_memblock_offlined(mem);
1724 if (unlikely(ret)) {
1725 phys_addr_t beginpa, endpa;
1727 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1728 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1729 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1737 static int check_cpu_on_node(pg_data_t *pgdat)
1741 for_each_present_cpu(cpu) {
1742 if (cpu_to_node(cpu) == pgdat->node_id)
1744 * the cpu on this node isn't removed, and we can't
1745 * offline this node.
1757 * Offline a node if all memory sections and cpus of the node are removed.
1759 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1760 * and online/offline operations before this call.
1762 void try_offline_node(int nid)
1764 pg_data_t *pgdat = NODE_DATA(nid);
1765 unsigned long start_pfn = pgdat->node_start_pfn;
1766 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1769 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1770 unsigned long section_nr = pfn_to_section_nr(pfn);
1772 if (!present_section_nr(section_nr))
1775 if (pfn_to_nid(pfn) != nid)
1779 * some memory sections of this node are not removed, and we
1780 * can't offline node now.
1785 if (check_cpu_on_node(pgdat))
1789 * all memory/cpu of this node are removed, we can offline this
1792 node_set_offline(nid);
1793 unregister_one_node(nid);
1795 EXPORT_SYMBOL(try_offline_node);
1797 static void __release_memory_resource(resource_size_t start,
1798 resource_size_t size)
1803 * When removing memory in the same granularity as it was added,
1804 * this function never fails. It might only fail if resources
1805 * have to be adjusted or split. We'll ignore the error, as
1806 * removing of memory cannot fail.
1808 ret = release_mem_region_adjustable(&iomem_resource, start, size);
1810 resource_size_t endres = start + size - 1;
1812 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
1813 &start, &endres, ret);
1817 static int __ref try_remove_memory(int nid, u64 start, u64 size)
1821 BUG_ON(check_hotplug_memory_range(start, size));
1823 mem_hotplug_begin();
1826 * All memory blocks must be offlined before removing memory. Check
1827 * whether all memory blocks in question are offline and return error
1828 * if this is not the case.
1830 rc = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
1831 check_memblock_offlined_cb);
1835 /* remove memmap entry */
1836 firmware_map_remove(start, start + size, "System RAM");
1837 memblock_free(start, size);
1838 memblock_remove(start, size);
1840 arch_remove_memory(nid, start, size, NULL);
1841 __release_memory_resource(start, size);
1843 try_offline_node(nid);
1853 * @start: physical address of the region to remove
1854 * @size: size of the region to remove
1856 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1857 * and online/offline operations before this call, as required by
1858 * try_offline_node().
1860 void __remove_memory(int nid, u64 start, u64 size)
1864 * trigger BUG() is some memory is not offlined prior to calling this
1867 if (try_remove_memory(nid, start, size))
1872 * Remove memory if every memory block is offline, otherwise return -EBUSY is
1873 * some memory is not offline
1875 int remove_memory(int nid, u64 start, u64 size)
1879 lock_device_hotplug();
1880 rc = try_remove_memory(nid, start, size);
1881 unlock_device_hotplug();
1885 EXPORT_SYMBOL_GPL(remove_memory);
1886 #endif /* CONFIG_MEMORY_HOTREMOVE */