2 * linux/mm/memory_hotplug.c
7 #include <linux/stddef.h>
9 #include <linux/sched/signal.h>
10 #include <linux/swap.h>
11 #include <linux/interrupt.h>
12 #include <linux/pagemap.h>
13 #include <linux/compiler.h>
14 #include <linux/export.h>
15 #include <linux/pagevec.h>
16 #include <linux/writeback.h>
17 #include <linux/slab.h>
18 #include <linux/sysctl.h>
19 #include <linux/cpu.h>
20 #include <linux/memory.h>
21 #include <linux/memremap.h>
22 #include <linux/memory_hotplug.h>
23 #include <linux/highmem.h>
24 #include <linux/vmalloc.h>
25 #include <linux/ioport.h>
26 #include <linux/delay.h>
27 #include <linux/migrate.h>
28 #include <linux/page-isolation.h>
29 #include <linux/pfn.h>
30 #include <linux/suspend.h>
31 #include <linux/mm_inline.h>
32 #include <linux/firmware-map.h>
33 #include <linux/stop_machine.h>
34 #include <linux/hugetlb.h>
35 #include <linux/memblock.h>
36 #include <linux/bootmem.h>
37 #include <linux/compaction.h>
39 #include <asm/tlbflush.h>
44 * online_page_callback contains pointer to current page onlining function.
45 * Initially it is generic_online_page(). If it is required it could be
46 * changed by calling set_online_page_callback() for callback registration
47 * and restore_online_page_callback() for generic callback restore.
50 static void generic_online_page(struct page *page);
52 static online_page_callback_t online_page_callback = generic_online_page;
53 static DEFINE_MUTEX(online_page_callback_lock);
55 /* The same as the cpu_hotplug lock, but for memory hotplug. */
57 struct task_struct *active_writer;
58 struct mutex lock; /* Synchronizes accesses to refcount, */
60 * Also blocks the new readers during
61 * an ongoing mem hotplug operation.
65 #ifdef CONFIG_DEBUG_LOCK_ALLOC
66 struct lockdep_map dep_map;
69 .active_writer = NULL,
70 .lock = __MUTEX_INITIALIZER(mem_hotplug.lock),
72 #ifdef CONFIG_DEBUG_LOCK_ALLOC
73 .dep_map = {.name = "mem_hotplug.lock" },
77 /* Lockdep annotations for get/put_online_mems() and mem_hotplug_begin/end() */
78 #define memhp_lock_acquire_read() lock_map_acquire_read(&mem_hotplug.dep_map)
79 #define memhp_lock_acquire() lock_map_acquire(&mem_hotplug.dep_map)
80 #define memhp_lock_release() lock_map_release(&mem_hotplug.dep_map)
82 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
83 bool memhp_auto_online;
85 bool memhp_auto_online = true;
87 EXPORT_SYMBOL_GPL(memhp_auto_online);
89 static int __init setup_memhp_default_state(char *str)
91 if (!strcmp(str, "online"))
92 memhp_auto_online = true;
93 else if (!strcmp(str, "offline"))
94 memhp_auto_online = false;
98 __setup("memhp_default_state=", setup_memhp_default_state);
100 void get_online_mems(void)
103 if (mem_hotplug.active_writer == current)
105 memhp_lock_acquire_read();
106 mutex_lock(&mem_hotplug.lock);
107 mem_hotplug.refcount++;
108 mutex_unlock(&mem_hotplug.lock);
112 void put_online_mems(void)
114 if (mem_hotplug.active_writer == current)
116 mutex_lock(&mem_hotplug.lock);
118 if (WARN_ON(!mem_hotplug.refcount))
119 mem_hotplug.refcount++; /* try to fix things up */
121 if (!--mem_hotplug.refcount && unlikely(mem_hotplug.active_writer))
122 wake_up_process(mem_hotplug.active_writer);
123 mutex_unlock(&mem_hotplug.lock);
124 memhp_lock_release();
128 /* Serializes write accesses to mem_hotplug.active_writer. */
129 static DEFINE_MUTEX(memory_add_remove_lock);
131 void mem_hotplug_begin(void)
133 mutex_lock(&memory_add_remove_lock);
135 mem_hotplug.active_writer = current;
137 memhp_lock_acquire();
139 mutex_lock(&mem_hotplug.lock);
140 if (likely(!mem_hotplug.refcount))
142 __set_current_state(TASK_UNINTERRUPTIBLE);
143 mutex_unlock(&mem_hotplug.lock);
148 void mem_hotplug_done(void)
150 mem_hotplug.active_writer = NULL;
151 mutex_unlock(&mem_hotplug.lock);
152 memhp_lock_release();
153 mutex_unlock(&memory_add_remove_lock);
156 /* add this memory to iomem resource */
157 static struct resource *register_memory_resource(u64 start, u64 size)
159 struct resource *res;
160 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
162 return ERR_PTR(-ENOMEM);
164 res->name = "System RAM";
166 res->end = start + size - 1;
167 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
168 if (request_resource(&iomem_resource, res) < 0) {
169 pr_debug("System RAM resource %pR cannot be added\n", res);
171 return ERR_PTR(-EEXIST);
176 static void release_memory_resource(struct resource *res)
180 release_resource(res);
185 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
186 void get_page_bootmem(unsigned long info, struct page *page,
189 page->freelist = (void *)type;
190 SetPagePrivate(page);
191 set_page_private(page, info);
195 void put_page_bootmem(struct page *page)
199 type = (unsigned long) page->freelist;
200 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
201 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
203 if (page_ref_dec_return(page) == 1) {
204 page->freelist = NULL;
205 ClearPagePrivate(page);
206 set_page_private(page, 0);
207 INIT_LIST_HEAD(&page->lru);
208 free_reserved_page(page);
212 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
213 #ifndef CONFIG_SPARSEMEM_VMEMMAP
214 static void register_page_bootmem_info_section(unsigned long start_pfn)
216 unsigned long *usemap, mapsize, section_nr, i;
217 struct mem_section *ms;
218 struct page *page, *memmap;
220 section_nr = pfn_to_section_nr(start_pfn);
221 ms = __nr_to_section(section_nr);
223 /* Get section's memmap address */
224 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
227 * Get page for the memmap's phys address
228 * XXX: need more consideration for sparse_vmemmap...
230 page = virt_to_page(memmap);
231 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
232 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
234 /* remember memmap's page */
235 for (i = 0; i < mapsize; i++, page++)
236 get_page_bootmem(section_nr, page, SECTION_INFO);
238 usemap = __nr_to_section(section_nr)->pageblock_flags;
239 page = virt_to_page(usemap);
241 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
243 for (i = 0; i < mapsize; i++, page++)
244 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
247 #else /* CONFIG_SPARSEMEM_VMEMMAP */
248 static void register_page_bootmem_info_section(unsigned long start_pfn)
250 unsigned long *usemap, mapsize, section_nr, i;
251 struct mem_section *ms;
252 struct page *page, *memmap;
254 if (!pfn_valid(start_pfn))
257 section_nr = pfn_to_section_nr(start_pfn);
258 ms = __nr_to_section(section_nr);
260 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
262 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
264 usemap = __nr_to_section(section_nr)->pageblock_flags;
265 page = virt_to_page(usemap);
267 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
269 for (i = 0; i < mapsize; i++, page++)
270 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
272 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
274 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
276 unsigned long i, pfn, end_pfn, nr_pages;
277 int node = pgdat->node_id;
280 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
281 page = virt_to_page(pgdat);
283 for (i = 0; i < nr_pages; i++, page++)
284 get_page_bootmem(node, page, NODE_INFO);
286 pfn = pgdat->node_start_pfn;
287 end_pfn = pgdat_end_pfn(pgdat);
289 /* register section info */
290 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
292 * Some platforms can assign the same pfn to multiple nodes - on
293 * node0 as well as nodeN. To avoid registering a pfn against
294 * multiple nodes we check that this pfn does not already
295 * reside in some other nodes.
297 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
298 register_page_bootmem_info_section(pfn);
301 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
303 static void __meminit grow_zone_span(struct zone *zone, unsigned long start_pfn,
304 unsigned long end_pfn)
306 unsigned long old_zone_end_pfn;
308 zone_span_writelock(zone);
310 old_zone_end_pfn = zone_end_pfn(zone);
311 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
312 zone->zone_start_pfn = start_pfn;
314 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
315 zone->zone_start_pfn;
317 zone_span_writeunlock(zone);
320 static void resize_zone(struct zone *zone, unsigned long start_pfn,
321 unsigned long end_pfn)
323 zone_span_writelock(zone);
325 if (end_pfn - start_pfn) {
326 zone->zone_start_pfn = start_pfn;
327 zone->spanned_pages = end_pfn - start_pfn;
330 * make it consist as free_area_init_core(),
331 * if spanned_pages = 0, then keep start_pfn = 0
333 zone->zone_start_pfn = 0;
334 zone->spanned_pages = 0;
337 zone_span_writeunlock(zone);
340 static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
341 unsigned long end_pfn)
343 enum zone_type zid = zone_idx(zone);
344 int nid = zone->zone_pgdat->node_id;
347 for (pfn = start_pfn; pfn < end_pfn; pfn++)
348 set_page_links(pfn_to_page(pfn), zid, nid, pfn);
351 /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
352 * alloc_bootmem_node_nopanic()/memblock_virt_alloc_node_nopanic() */
353 static int __ref ensure_zone_is_initialized(struct zone *zone,
354 unsigned long start_pfn, unsigned long num_pages)
356 if (!zone_is_initialized(zone))
357 return init_currently_empty_zone(zone, start_pfn, num_pages);
362 static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
363 unsigned long start_pfn, unsigned long end_pfn)
367 unsigned long z1_start_pfn;
369 ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn);
373 pgdat_resize_lock(z1->zone_pgdat, &flags);
375 /* can't move pfns which are higher than @z2 */
376 if (end_pfn > zone_end_pfn(z2))
378 /* the move out part must be at the left most of @z2 */
379 if (start_pfn > z2->zone_start_pfn)
381 /* must included/overlap */
382 if (end_pfn <= z2->zone_start_pfn)
385 /* use start_pfn for z1's start_pfn if z1 is empty */
386 if (!zone_is_empty(z1))
387 z1_start_pfn = z1->zone_start_pfn;
389 z1_start_pfn = start_pfn;
391 resize_zone(z1, z1_start_pfn, end_pfn);
392 resize_zone(z2, end_pfn, zone_end_pfn(z2));
394 pgdat_resize_unlock(z1->zone_pgdat, &flags);
396 fix_zone_id(z1, start_pfn, end_pfn);
400 pgdat_resize_unlock(z1->zone_pgdat, &flags);
404 static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
405 unsigned long start_pfn, unsigned long end_pfn)
409 unsigned long z2_end_pfn;
411 ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn);
415 pgdat_resize_lock(z1->zone_pgdat, &flags);
417 /* can't move pfns which are lower than @z1 */
418 if (z1->zone_start_pfn > start_pfn)
420 /* the move out part mast at the right most of @z1 */
421 if (zone_end_pfn(z1) > end_pfn)
423 /* must included/overlap */
424 if (start_pfn >= zone_end_pfn(z1))
427 /* use end_pfn for z2's end_pfn if z2 is empty */
428 if (!zone_is_empty(z2))
429 z2_end_pfn = zone_end_pfn(z2);
431 z2_end_pfn = end_pfn;
433 resize_zone(z1, z1->zone_start_pfn, start_pfn);
434 resize_zone(z2, start_pfn, z2_end_pfn);
436 pgdat_resize_unlock(z1->zone_pgdat, &flags);
438 fix_zone_id(z2, start_pfn, end_pfn);
442 pgdat_resize_unlock(z1->zone_pgdat, &flags);
446 static struct zone * __meminit move_pfn_range(int zone_shift,
447 unsigned long start_pfn, unsigned long end_pfn)
449 struct zone *zone = page_zone(pfn_to_page(start_pfn));
453 ret = move_pfn_range_left(zone + zone_shift, zone,
456 ret = move_pfn_range_right(zone, zone + zone_shift,
462 return zone + zone_shift;
465 static void __meminit grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
466 unsigned long end_pfn)
468 unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat);
470 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
471 pgdat->node_start_pfn = start_pfn;
473 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
474 pgdat->node_start_pfn;
477 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
479 struct pglist_data *pgdat = zone->zone_pgdat;
480 int nr_pages = PAGES_PER_SECTION;
481 int nid = pgdat->node_id;
483 unsigned long flags, pfn;
486 zone_type = zone - pgdat->node_zones;
487 ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
491 pgdat_resize_lock(zone->zone_pgdat, &flags);
492 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
493 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
494 phys_start_pfn + nr_pages);
495 pgdat_resize_unlock(zone->zone_pgdat, &flags);
496 memmap_init_zone(nr_pages, nid, zone_type,
497 phys_start_pfn, MEMMAP_HOTPLUG);
499 /* online_page_range is called later and expects pages reserved */
500 for (pfn = phys_start_pfn; pfn < phys_start_pfn + nr_pages; pfn++) {
504 SetPageReserved(pfn_to_page(pfn));
509 static int __meminit __add_section(int nid, struct zone *zone,
510 unsigned long phys_start_pfn)
514 if (pfn_valid(phys_start_pfn))
517 ret = sparse_add_one_section(zone, phys_start_pfn);
522 ret = __add_zone(zone, phys_start_pfn);
527 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
531 * Reasonably generic function for adding memory. It is
532 * expected that archs that support memory hotplug will
533 * call this function after deciding the zone to which to
536 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
537 unsigned long nr_pages)
541 int start_sec, end_sec;
542 struct vmem_altmap *altmap;
544 clear_zone_contiguous(zone);
546 /* during initialize mem_map, align hot-added range to section */
547 start_sec = pfn_to_section_nr(phys_start_pfn);
548 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
550 altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn));
553 * Validate altmap is within bounds of the total request
555 if (altmap->base_pfn != phys_start_pfn
556 || vmem_altmap_offset(altmap) > nr_pages) {
557 pr_warn_once("memory add fail, invalid altmap\n");
564 for (i = start_sec; i <= end_sec; i++) {
565 err = __add_section(nid, zone, section_nr_to_pfn(i));
568 * EEXIST is finally dealt with by ioresource collision
569 * check. see add_memory() => register_memory_resource()
570 * Warning will be printed if there is collision.
572 if (err && (err != -EEXIST))
576 vmemmap_populate_print_last();
578 set_zone_contiguous(zone);
581 EXPORT_SYMBOL_GPL(__add_pages);
583 #ifdef CONFIG_MEMORY_HOTREMOVE
584 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
585 static int find_smallest_section_pfn(int nid, struct zone *zone,
586 unsigned long start_pfn,
587 unsigned long end_pfn)
589 struct mem_section *ms;
591 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
592 ms = __pfn_to_section(start_pfn);
594 if (unlikely(!valid_section(ms)))
597 if (unlikely(pfn_to_nid(start_pfn) != nid))
600 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
609 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
610 static int find_biggest_section_pfn(int nid, struct zone *zone,
611 unsigned long start_pfn,
612 unsigned long end_pfn)
614 struct mem_section *ms;
617 /* pfn is the end pfn of a memory section. */
619 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
620 ms = __pfn_to_section(pfn);
622 if (unlikely(!valid_section(ms)))
625 if (unlikely(pfn_to_nid(pfn) != nid))
628 if (zone && zone != page_zone(pfn_to_page(pfn)))
637 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
638 unsigned long end_pfn)
640 unsigned long zone_start_pfn = zone->zone_start_pfn;
641 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
642 unsigned long zone_end_pfn = z;
644 struct mem_section *ms;
645 int nid = zone_to_nid(zone);
647 zone_span_writelock(zone);
648 if (zone_start_pfn == start_pfn) {
650 * If the section is smallest section in the zone, it need
651 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
652 * In this case, we find second smallest valid mem_section
653 * for shrinking zone.
655 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
658 zone->zone_start_pfn = pfn;
659 zone->spanned_pages = zone_end_pfn - pfn;
661 } else if (zone_end_pfn == end_pfn) {
663 * If the section is biggest section in the zone, it need
664 * shrink zone->spanned_pages.
665 * In this case, we find second biggest valid mem_section for
668 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
671 zone->spanned_pages = pfn - zone_start_pfn + 1;
675 * The section is not biggest or smallest mem_section in the zone, it
676 * only creates a hole in the zone. So in this case, we need not
677 * change the zone. But perhaps, the zone has only hole data. Thus
678 * it check the zone has only hole or not.
680 pfn = zone_start_pfn;
681 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
682 ms = __pfn_to_section(pfn);
684 if (unlikely(!valid_section(ms)))
687 if (page_zone(pfn_to_page(pfn)) != zone)
690 /* If the section is current section, it continues the loop */
691 if (start_pfn == pfn)
694 /* If we find valid section, we have nothing to do */
695 zone_span_writeunlock(zone);
699 /* The zone has no valid section */
700 zone->zone_start_pfn = 0;
701 zone->spanned_pages = 0;
702 zone_span_writeunlock(zone);
705 static void shrink_pgdat_span(struct pglist_data *pgdat,
706 unsigned long start_pfn, unsigned long end_pfn)
708 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
709 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
710 unsigned long pgdat_end_pfn = p;
712 struct mem_section *ms;
713 int nid = pgdat->node_id;
715 if (pgdat_start_pfn == start_pfn) {
717 * If the section is smallest section in the pgdat, it need
718 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
719 * In this case, we find second smallest valid mem_section
720 * for shrinking zone.
722 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
725 pgdat->node_start_pfn = pfn;
726 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
728 } else if (pgdat_end_pfn == end_pfn) {
730 * If the section is biggest section in the pgdat, it need
731 * shrink pgdat->node_spanned_pages.
732 * In this case, we find second biggest valid mem_section for
735 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
738 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
742 * If the section is not biggest or smallest mem_section in the pgdat,
743 * it only creates a hole in the pgdat. So in this case, we need not
745 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
746 * has only hole or not.
748 pfn = pgdat_start_pfn;
749 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
750 ms = __pfn_to_section(pfn);
752 if (unlikely(!valid_section(ms)))
755 if (pfn_to_nid(pfn) != nid)
758 /* If the section is current section, it continues the loop */
759 if (start_pfn == pfn)
762 /* If we find valid section, we have nothing to do */
766 /* The pgdat has no valid section */
767 pgdat->node_start_pfn = 0;
768 pgdat->node_spanned_pages = 0;
771 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
773 struct pglist_data *pgdat = zone->zone_pgdat;
774 int nr_pages = PAGES_PER_SECTION;
778 zone_type = zone - pgdat->node_zones;
780 pgdat_resize_lock(zone->zone_pgdat, &flags);
781 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
782 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
783 pgdat_resize_unlock(zone->zone_pgdat, &flags);
786 static int __remove_section(struct zone *zone, struct mem_section *ms,
787 unsigned long map_offset)
789 unsigned long start_pfn;
793 if (!valid_section(ms))
796 ret = unregister_memory_section(ms);
800 scn_nr = __section_nr(ms);
801 start_pfn = section_nr_to_pfn(scn_nr);
802 __remove_zone(zone, start_pfn);
804 sparse_remove_one_section(zone, ms, map_offset);
809 * __remove_pages() - remove sections of pages from a zone
810 * @zone: zone from which pages need to be removed
811 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
812 * @nr_pages: number of pages to remove (must be multiple of section size)
814 * Generic helper function to remove section mappings and sysfs entries
815 * for the section of the memory we are removing. Caller needs to make
816 * sure that pages are marked reserved and zones are adjust properly by
817 * calling offline_pages().
819 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
820 unsigned long nr_pages)
823 unsigned long map_offset = 0;
824 int sections_to_remove, ret = 0;
826 /* In the ZONE_DEVICE case device driver owns the memory region */
827 if (is_dev_zone(zone)) {
828 struct page *page = pfn_to_page(phys_start_pfn);
829 struct vmem_altmap *altmap;
831 altmap = to_vmem_altmap((unsigned long) page);
833 map_offset = vmem_altmap_offset(altmap);
835 resource_size_t start, size;
837 start = phys_start_pfn << PAGE_SHIFT;
838 size = nr_pages * PAGE_SIZE;
840 ret = release_mem_region_adjustable(&iomem_resource, start,
843 resource_size_t endres = start + size - 1;
845 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
846 &start, &endres, ret);
850 clear_zone_contiguous(zone);
853 * We can only remove entire sections
855 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
856 BUG_ON(nr_pages % PAGES_PER_SECTION);
858 sections_to_remove = nr_pages / PAGES_PER_SECTION;
859 for (i = 0; i < sections_to_remove; i++) {
860 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
862 ret = __remove_section(zone, __pfn_to_section(pfn), map_offset);
868 set_zone_contiguous(zone);
872 #endif /* CONFIG_MEMORY_HOTREMOVE */
874 int set_online_page_callback(online_page_callback_t callback)
879 mutex_lock(&online_page_callback_lock);
881 if (online_page_callback == generic_online_page) {
882 online_page_callback = callback;
886 mutex_unlock(&online_page_callback_lock);
891 EXPORT_SYMBOL_GPL(set_online_page_callback);
893 int restore_online_page_callback(online_page_callback_t callback)
898 mutex_lock(&online_page_callback_lock);
900 if (online_page_callback == callback) {
901 online_page_callback = generic_online_page;
905 mutex_unlock(&online_page_callback_lock);
910 EXPORT_SYMBOL_GPL(restore_online_page_callback);
912 void __online_page_set_limits(struct page *page)
915 EXPORT_SYMBOL_GPL(__online_page_set_limits);
917 void __online_page_increment_counters(struct page *page)
919 adjust_managed_page_count(page, 1);
921 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
923 void __online_page_free(struct page *page)
925 __free_reserved_page(page);
927 EXPORT_SYMBOL_GPL(__online_page_free);
929 static void generic_online_page(struct page *page)
931 __online_page_set_limits(page);
932 __online_page_increment_counters(page);
933 __online_page_free(page);
936 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
940 unsigned long onlined_pages = *(unsigned long *)arg;
942 if (PageReserved(pfn_to_page(start_pfn)))
943 for (i = 0; i < nr_pages; i++) {
944 page = pfn_to_page(start_pfn + i);
945 (*online_page_callback)(page);
948 *(unsigned long *)arg = onlined_pages;
952 #ifdef CONFIG_MOVABLE_NODE
954 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
957 static bool can_online_high_movable(struct zone *zone)
961 #else /* CONFIG_MOVABLE_NODE */
962 /* ensure every online node has NORMAL memory */
963 static bool can_online_high_movable(struct zone *zone)
965 return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
967 #endif /* CONFIG_MOVABLE_NODE */
969 /* check which state of node_states will be changed when online memory */
970 static void node_states_check_changes_online(unsigned long nr_pages,
971 struct zone *zone, struct memory_notify *arg)
973 int nid = zone_to_nid(zone);
974 enum zone_type zone_last = ZONE_NORMAL;
977 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
978 * contains nodes which have zones of 0...ZONE_NORMAL,
979 * set zone_last to ZONE_NORMAL.
981 * If we don't have HIGHMEM nor movable node,
982 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
983 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
985 if (N_MEMORY == N_NORMAL_MEMORY)
986 zone_last = ZONE_MOVABLE;
989 * if the memory to be online is in a zone of 0...zone_last, and
990 * the zones of 0...zone_last don't have memory before online, we will
991 * need to set the node to node_states[N_NORMAL_MEMORY] after
992 * the memory is online.
994 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
995 arg->status_change_nid_normal = nid;
997 arg->status_change_nid_normal = -1;
999 #ifdef CONFIG_HIGHMEM
1001 * If we have movable node, node_states[N_HIGH_MEMORY]
1002 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1003 * set zone_last to ZONE_HIGHMEM.
1005 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1006 * contains nodes which have zones of 0...ZONE_MOVABLE,
1007 * set zone_last to ZONE_MOVABLE.
1009 zone_last = ZONE_HIGHMEM;
1010 if (N_MEMORY == N_HIGH_MEMORY)
1011 zone_last = ZONE_MOVABLE;
1013 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
1014 arg->status_change_nid_high = nid;
1016 arg->status_change_nid_high = -1;
1018 arg->status_change_nid_high = arg->status_change_nid_normal;
1022 * if the node don't have memory befor online, we will need to
1023 * set the node to node_states[N_MEMORY] after the memory
1026 if (!node_state(nid, N_MEMORY))
1027 arg->status_change_nid = nid;
1029 arg->status_change_nid = -1;
1032 static void node_states_set_node(int node, struct memory_notify *arg)
1034 if (arg->status_change_nid_normal >= 0)
1035 node_set_state(node, N_NORMAL_MEMORY);
1037 if (arg->status_change_nid_high >= 0)
1038 node_set_state(node, N_HIGH_MEMORY);
1040 node_set_state(node, N_MEMORY);
1043 bool zone_can_shift(unsigned long pfn, unsigned long nr_pages,
1044 enum zone_type target, int *zone_shift)
1046 struct zone *zone = page_zone(pfn_to_page(pfn));
1047 enum zone_type idx = zone_idx(zone);
1053 /* pages must be at end of current zone */
1054 if (pfn + nr_pages != zone_end_pfn(zone))
1057 /* no zones in use between current zone and target */
1058 for (i = idx + 1; i < target; i++)
1059 if (zone_is_initialized(zone - idx + i))
1064 /* pages must be at beginning of current zone */
1065 if (pfn != zone->zone_start_pfn)
1068 /* no zones in use between current zone and target */
1069 for (i = target + 1; i < idx; i++)
1070 if (zone_is_initialized(zone - idx + i))
1074 *zone_shift = target - idx;
1078 /* Must be protected by mem_hotplug_begin() */
1079 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
1081 unsigned long flags;
1082 unsigned long onlined_pages = 0;
1084 int need_zonelists_rebuild = 0;
1087 struct memory_notify arg;
1091 * This doesn't need a lock to do pfn_to_page().
1092 * The section can't be removed here because of the
1093 * memory_block->state_mutex.
1095 zone = page_zone(pfn_to_page(pfn));
1097 if ((zone_idx(zone) > ZONE_NORMAL ||
1098 online_type == MMOP_ONLINE_MOVABLE) &&
1099 !can_online_high_movable(zone))
1102 if (online_type == MMOP_ONLINE_KERNEL) {
1103 if (!zone_can_shift(pfn, nr_pages, ZONE_NORMAL, &zone_shift))
1105 } else if (online_type == MMOP_ONLINE_MOVABLE) {
1106 if (!zone_can_shift(pfn, nr_pages, ZONE_MOVABLE, &zone_shift))
1110 zone = move_pfn_range(zone_shift, pfn, pfn + nr_pages);
1114 arg.start_pfn = pfn;
1115 arg.nr_pages = nr_pages;
1116 node_states_check_changes_online(nr_pages, zone, &arg);
1118 nid = zone_to_nid(zone);
1120 ret = memory_notify(MEM_GOING_ONLINE, &arg);
1121 ret = notifier_to_errno(ret);
1123 goto failed_addition;
1126 * If this zone is not populated, then it is not in zonelist.
1127 * This means the page allocator ignores this zone.
1128 * So, zonelist must be updated after online.
1130 mutex_lock(&zonelists_mutex);
1131 if (!populated_zone(zone)) {
1132 need_zonelists_rebuild = 1;
1133 build_all_zonelists(NULL, zone);
1136 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
1137 online_pages_range);
1139 if (need_zonelists_rebuild)
1140 zone_pcp_reset(zone);
1141 mutex_unlock(&zonelists_mutex);
1142 goto failed_addition;
1145 zone->present_pages += onlined_pages;
1147 pgdat_resize_lock(zone->zone_pgdat, &flags);
1148 zone->zone_pgdat->node_present_pages += onlined_pages;
1149 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1151 if (onlined_pages) {
1152 node_states_set_node(nid, &arg);
1153 if (need_zonelists_rebuild)
1154 build_all_zonelists(NULL, NULL);
1156 zone_pcp_update(zone);
1159 mutex_unlock(&zonelists_mutex);
1161 init_per_zone_wmark_min();
1163 if (onlined_pages) {
1168 vm_total_pages = nr_free_pagecache_pages();
1170 writeback_set_ratelimit();
1173 memory_notify(MEM_ONLINE, &arg);
1177 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
1178 (unsigned long long) pfn << PAGE_SHIFT,
1179 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
1180 memory_notify(MEM_CANCEL_ONLINE, &arg);
1183 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1185 static void reset_node_present_pages(pg_data_t *pgdat)
1189 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
1190 z->present_pages = 0;
1192 pgdat->node_present_pages = 0;
1195 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1196 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1198 struct pglist_data *pgdat;
1199 unsigned long zones_size[MAX_NR_ZONES] = {0};
1200 unsigned long zholes_size[MAX_NR_ZONES] = {0};
1201 unsigned long start_pfn = PFN_DOWN(start);
1203 pgdat = NODE_DATA(nid);
1205 pgdat = arch_alloc_nodedata(nid);
1209 arch_refresh_nodedata(nid, pgdat);
1212 * Reset the nr_zones, order and classzone_idx before reuse.
1213 * Note that kswapd will init kswapd_classzone_idx properly
1214 * when it starts in the near future.
1216 pgdat->nr_zones = 0;
1217 pgdat->kswapd_order = 0;
1218 pgdat->kswapd_classzone_idx = 0;
1221 /* we can use NODE_DATA(nid) from here */
1223 /* init node's zones as empty zones, we don't have any present pages.*/
1224 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1225 pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
1228 * The node we allocated has no zone fallback lists. For avoiding
1229 * to access not-initialized zonelist, build here.
1231 mutex_lock(&zonelists_mutex);
1232 build_all_zonelists(pgdat, NULL);
1233 mutex_unlock(&zonelists_mutex);
1236 * zone->managed_pages is set to an approximate value in
1237 * free_area_init_core(), which will cause
1238 * /sys/device/system/node/nodeX/meminfo has wrong data.
1239 * So reset it to 0 before any memory is onlined.
1241 reset_node_managed_pages(pgdat);
1244 * When memory is hot-added, all the memory is in offline state. So
1245 * clear all zones' present_pages because they will be updated in
1246 * online_pages() and offline_pages().
1248 reset_node_present_pages(pgdat);
1253 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1255 arch_refresh_nodedata(nid, NULL);
1256 free_percpu(pgdat->per_cpu_nodestats);
1257 arch_free_nodedata(pgdat);
1263 * try_online_node - online a node if offlined
1265 * called by cpu_up() to online a node without onlined memory.
1267 int try_online_node(int nid)
1272 if (node_online(nid))
1275 mem_hotplug_begin();
1276 pgdat = hotadd_new_pgdat(nid, 0);
1278 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1282 node_set_online(nid);
1283 ret = register_one_node(nid);
1286 if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
1287 mutex_lock(&zonelists_mutex);
1288 build_all_zonelists(NULL, NULL);
1289 mutex_unlock(&zonelists_mutex);
1297 static int check_hotplug_memory_range(u64 start, u64 size)
1299 u64 start_pfn = PFN_DOWN(start);
1300 u64 nr_pages = size >> PAGE_SHIFT;
1302 /* Memory range must be aligned with section */
1303 if ((start_pfn & ~PAGE_SECTION_MASK) ||
1304 (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1305 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1306 (unsigned long long)start,
1307 (unsigned long long)size);
1315 * If movable zone has already been setup, newly added memory should be check.
1316 * If its address is higher than movable zone, it should be added as movable.
1317 * Without this check, movable zone may overlap with other zone.
1319 static int should_add_memory_movable(int nid, u64 start, u64 size)
1321 unsigned long start_pfn = start >> PAGE_SHIFT;
1322 pg_data_t *pgdat = NODE_DATA(nid);
1323 struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE;
1325 if (zone_is_empty(movable_zone))
1328 if (movable_zone->zone_start_pfn <= start_pfn)
1334 int zone_for_memory(int nid, u64 start, u64 size, int zone_default,
1337 #ifdef CONFIG_ZONE_DEVICE
1341 if (should_add_memory_movable(nid, start, size))
1342 return ZONE_MOVABLE;
1344 return zone_default;
1347 static int online_memory_block(struct memory_block *mem, void *arg)
1349 return device_online(&mem->dev);
1352 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1353 int __ref add_memory_resource(int nid, struct resource *res, bool online)
1356 pg_data_t *pgdat = NULL;
1362 size = resource_size(res);
1364 ret = check_hotplug_memory_range(start, size);
1368 { /* Stupid hack to suppress address-never-null warning */
1369 void *p = NODE_DATA(nid);
1373 mem_hotplug_begin();
1376 * Add new range to memblock so that when hotadd_new_pgdat() is called
1377 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1378 * this new range and calculate total pages correctly. The range will
1379 * be removed at hot-remove time.
1381 memblock_add_node(start, size, nid);
1383 new_node = !node_online(nid);
1385 pgdat = hotadd_new_pgdat(nid, start);
1391 /* call arch's memory hotadd */
1392 ret = arch_add_memory(nid, start, size, false);
1397 /* we online node here. we can't roll back from here. */
1398 node_set_online(nid);
1401 ret = register_one_node(nid);
1403 * If sysfs file of new node can't create, cpu on the node
1404 * can't be hot-added. There is no rollback way now.
1405 * So, check by BUG_ON() to catch it reluctantly..
1410 /* create new memmap entry */
1411 firmware_map_add_hotplug(start, start + size, "System RAM");
1413 /* online pages if requested */
1415 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1416 NULL, online_memory_block);
1421 /* rollback pgdat allocation and others */
1423 rollback_node_hotadd(nid, pgdat);
1424 memblock_remove(start, size);
1430 EXPORT_SYMBOL_GPL(add_memory_resource);
1432 int __ref add_memory(int nid, u64 start, u64 size)
1434 struct resource *res;
1437 res = register_memory_resource(start, size);
1439 return PTR_ERR(res);
1441 ret = add_memory_resource(nid, res, memhp_auto_online);
1443 release_memory_resource(res);
1446 EXPORT_SYMBOL_GPL(add_memory);
1448 #ifdef CONFIG_MEMORY_HOTREMOVE
1450 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1451 * set and the size of the free page is given by page_order(). Using this,
1452 * the function determines if the pageblock contains only free pages.
1453 * Due to buddy contraints, a free page at least the size of a pageblock will
1454 * be located at the start of the pageblock
1456 static inline int pageblock_free(struct page *page)
1458 return PageBuddy(page) && page_order(page) >= pageblock_order;
1461 /* Return the start of the next active pageblock after a given page */
1462 static struct page *next_active_pageblock(struct page *page)
1464 /* Ensure the starting page is pageblock-aligned */
1465 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1467 /* If the entire pageblock is free, move to the end of free page */
1468 if (pageblock_free(page)) {
1470 /* be careful. we don't have locks, page_order can be changed.*/
1471 order = page_order(page);
1472 if ((order < MAX_ORDER) && (order >= pageblock_order))
1473 return page + (1 << order);
1476 return page + pageblock_nr_pages;
1479 /* Checks if this range of memory is likely to be hot-removable. */
1480 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1482 struct page *page = pfn_to_page(start_pfn);
1483 struct page *end_page = page + nr_pages;
1485 /* Check the starting page of each pageblock within the range */
1486 for (; page < end_page; page = next_active_pageblock(page)) {
1487 if (!is_pageblock_removable_nolock(page))
1492 /* All pageblocks in the memory block are likely to be hot-removable */
1497 * Confirm all pages in a range [start, end) belong to the same zone.
1498 * When true, return its valid [start, end).
1500 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1501 unsigned long *valid_start, unsigned long *valid_end)
1503 unsigned long pfn, sec_end_pfn;
1504 unsigned long start, end;
1505 struct zone *zone = NULL;
1508 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1510 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1511 /* Make sure the memory section is present first */
1512 if (!present_section_nr(pfn_to_section_nr(pfn)))
1514 for (; pfn < sec_end_pfn && pfn < end_pfn;
1515 pfn += MAX_ORDER_NR_PAGES) {
1517 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1518 while ((i < MAX_ORDER_NR_PAGES) &&
1519 !pfn_valid_within(pfn + i))
1521 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1523 page = pfn_to_page(pfn + i);
1524 if (zone && page_zone(page) != zone)
1528 zone = page_zone(page);
1529 end = pfn + MAX_ORDER_NR_PAGES;
1534 *valid_start = start;
1535 *valid_end = min(end, end_pfn);
1543 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1544 * non-lru movable pages and hugepages). We scan pfn because it's much
1545 * easier than scanning over linked list. This function returns the pfn
1546 * of the first found movable page if it's found, otherwise 0.
1548 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1552 for (pfn = start; pfn < end; pfn++) {
1553 if (pfn_valid(pfn)) {
1554 page = pfn_to_page(pfn);
1557 if (__PageMovable(page))
1559 if (PageHuge(page)) {
1560 if (page_huge_active(page))
1563 pfn = round_up(pfn + 1,
1564 1 << compound_order(page)) - 1;
1571 static struct page *new_node_page(struct page *page, unsigned long private,
1574 gfp_t gfp_mask = GFP_USER | __GFP_MOVABLE;
1575 int nid = page_to_nid(page);
1576 nodemask_t nmask = node_states[N_MEMORY];
1577 struct page *new_page = NULL;
1580 * TODO: allocate a destination hugepage from a nearest neighbor node,
1581 * accordance with memory policy of the user process if possible. For
1582 * now as a simple work-around, we use the next node for destination.
1585 return alloc_huge_page_node(page_hstate(compound_head(page)),
1586 next_node_in(nid, nmask));
1588 node_clear(nid, nmask);
1590 if (PageHighMem(page)
1591 || (zone_idx(page_zone(page)) == ZONE_MOVABLE))
1592 gfp_mask |= __GFP_HIGHMEM;
1594 if (!nodes_empty(nmask))
1595 new_page = __alloc_pages_nodemask(gfp_mask, 0,
1596 node_zonelist(nid, gfp_mask), &nmask);
1598 new_page = __alloc_pages(gfp_mask, 0,
1599 node_zonelist(nid, gfp_mask));
1604 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1606 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1610 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1611 int not_managed = 0;
1615 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1616 if (!pfn_valid(pfn))
1618 page = pfn_to_page(pfn);
1620 if (PageHuge(page)) {
1621 struct page *head = compound_head(page);
1622 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1623 if (compound_order(head) > PFN_SECTION_SHIFT) {
1627 if (isolate_huge_page(page, &source))
1628 move_pages -= 1 << compound_order(head);
1632 if (!get_page_unless_zero(page))
1635 * We can skip free pages. And we can deal with pages on
1636 * LRU and non-lru movable pages.
1639 ret = isolate_lru_page(page);
1641 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1642 if (!ret) { /* Success */
1644 list_add_tail(&page->lru, &source);
1646 if (!__PageMovable(page))
1647 inc_node_page_state(page, NR_ISOLATED_ANON +
1648 page_is_file_cache(page));
1651 #ifdef CONFIG_DEBUG_VM
1652 pr_alert("failed to isolate pfn %lx\n", pfn);
1653 dump_page(page, "isolation failed");
1656 /* Because we don't have big zone->lock. we should
1657 check this again here. */
1658 if (page_count(page)) {
1665 if (!list_empty(&source)) {
1667 putback_movable_pages(&source);
1671 /* Allocate a new page from the nearest neighbor node */
1672 ret = migrate_pages(&source, new_node_page, NULL, 0,
1673 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1675 putback_movable_pages(&source);
1682 * remove from free_area[] and mark all as Reserved.
1685 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1688 __offline_isolated_pages(start, start + nr_pages);
1693 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1695 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1696 offline_isolated_pages_cb);
1700 * Check all pages in range, recoreded as memory resource, are isolated.
1703 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1707 long offlined = *(long *)data;
1708 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1709 offlined = nr_pages;
1711 *(long *)data += offlined;
1716 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1721 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1722 check_pages_isolated_cb);
1724 offlined = (long)ret;
1728 #ifdef CONFIG_MOVABLE_NODE
1730 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1733 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1737 #else /* CONFIG_MOVABLE_NODE */
1738 /* ensure the node has NORMAL memory if it is still online */
1739 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1741 struct pglist_data *pgdat = zone->zone_pgdat;
1742 unsigned long present_pages = 0;
1745 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1746 present_pages += pgdat->node_zones[zt].present_pages;
1748 if (present_pages > nr_pages)
1752 for (; zt <= ZONE_MOVABLE; zt++)
1753 present_pages += pgdat->node_zones[zt].present_pages;
1756 * we can't offline the last normal memory until all
1757 * higher memory is offlined.
1759 return present_pages == 0;
1761 #endif /* CONFIG_MOVABLE_NODE */
1763 static int __init cmdline_parse_movable_node(char *p)
1765 #ifdef CONFIG_MOVABLE_NODE
1766 movable_node_enabled = true;
1768 pr_warn("movable_node option not supported\n");
1772 early_param("movable_node", cmdline_parse_movable_node);
1774 /* check which state of node_states will be changed when offline memory */
1775 static void node_states_check_changes_offline(unsigned long nr_pages,
1776 struct zone *zone, struct memory_notify *arg)
1778 struct pglist_data *pgdat = zone->zone_pgdat;
1779 unsigned long present_pages = 0;
1780 enum zone_type zt, zone_last = ZONE_NORMAL;
1783 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1784 * contains nodes which have zones of 0...ZONE_NORMAL,
1785 * set zone_last to ZONE_NORMAL.
1787 * If we don't have HIGHMEM nor movable node,
1788 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1789 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1791 if (N_MEMORY == N_NORMAL_MEMORY)
1792 zone_last = ZONE_MOVABLE;
1795 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1796 * If the memory to be offline is in a zone of 0...zone_last,
1797 * and it is the last present memory, 0...zone_last will
1798 * become empty after offline , thus we can determind we will
1799 * need to clear the node from node_states[N_NORMAL_MEMORY].
1801 for (zt = 0; zt <= zone_last; zt++)
1802 present_pages += pgdat->node_zones[zt].present_pages;
1803 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1804 arg->status_change_nid_normal = zone_to_nid(zone);
1806 arg->status_change_nid_normal = -1;
1808 #ifdef CONFIG_HIGHMEM
1810 * If we have movable node, node_states[N_HIGH_MEMORY]
1811 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1812 * set zone_last to ZONE_HIGHMEM.
1814 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1815 * contains nodes which have zones of 0...ZONE_MOVABLE,
1816 * set zone_last to ZONE_MOVABLE.
1818 zone_last = ZONE_HIGHMEM;
1819 if (N_MEMORY == N_HIGH_MEMORY)
1820 zone_last = ZONE_MOVABLE;
1822 for (; zt <= zone_last; zt++)
1823 present_pages += pgdat->node_zones[zt].present_pages;
1824 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1825 arg->status_change_nid_high = zone_to_nid(zone);
1827 arg->status_change_nid_high = -1;
1829 arg->status_change_nid_high = arg->status_change_nid_normal;
1833 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1835 zone_last = ZONE_MOVABLE;
1838 * check whether node_states[N_HIGH_MEMORY] will be changed
1839 * If we try to offline the last present @nr_pages from the node,
1840 * we can determind we will need to clear the node from
1841 * node_states[N_HIGH_MEMORY].
1843 for (; zt <= zone_last; zt++)
1844 present_pages += pgdat->node_zones[zt].present_pages;
1845 if (nr_pages >= present_pages)
1846 arg->status_change_nid = zone_to_nid(zone);
1848 arg->status_change_nid = -1;
1851 static void node_states_clear_node(int node, struct memory_notify *arg)
1853 if (arg->status_change_nid_normal >= 0)
1854 node_clear_state(node, N_NORMAL_MEMORY);
1856 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1857 (arg->status_change_nid_high >= 0))
1858 node_clear_state(node, N_HIGH_MEMORY);
1860 if ((N_MEMORY != N_HIGH_MEMORY) &&
1861 (arg->status_change_nid >= 0))
1862 node_clear_state(node, N_MEMORY);
1865 static int __ref __offline_pages(unsigned long start_pfn,
1866 unsigned long end_pfn, unsigned long timeout)
1868 unsigned long pfn, nr_pages, expire;
1869 long offlined_pages;
1870 int ret, drain, retry_max, node;
1871 unsigned long flags;
1872 unsigned long valid_start, valid_end;
1874 struct memory_notify arg;
1876 /* at least, alignment against pageblock is necessary */
1877 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1879 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1881 /* This makes hotplug much easier...and readable.
1882 we assume this for now. .*/
1883 if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, &valid_end))
1886 zone = page_zone(pfn_to_page(valid_start));
1887 node = zone_to_nid(zone);
1888 nr_pages = end_pfn - start_pfn;
1890 if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1893 /* set above range as isolated */
1894 ret = start_isolate_page_range(start_pfn, end_pfn,
1895 MIGRATE_MOVABLE, true);
1899 arg.start_pfn = start_pfn;
1900 arg.nr_pages = nr_pages;
1901 node_states_check_changes_offline(nr_pages, zone, &arg);
1903 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1904 ret = notifier_to_errno(ret);
1906 goto failed_removal;
1909 expire = jiffies + timeout;
1913 /* start memory hot removal */
1915 if (time_after(jiffies, expire))
1916 goto failed_removal;
1918 if (signal_pending(current))
1919 goto failed_removal;
1922 lru_add_drain_all();
1924 drain_all_pages(zone);
1927 pfn = scan_movable_pages(start_pfn, end_pfn);
1928 if (pfn) { /* We have movable pages */
1929 ret = do_migrate_range(pfn, end_pfn);
1935 if (--retry_max == 0)
1936 goto failed_removal;
1942 /* drain all zone's lru pagevec, this is asynchronous... */
1943 lru_add_drain_all();
1945 /* drain pcp pages, this is synchronous. */
1946 drain_all_pages(zone);
1948 * dissolve free hugepages in the memory block before doing offlining
1949 * actually in order to make hugetlbfs's object counting consistent.
1951 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1953 goto failed_removal;
1955 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1956 if (offlined_pages < 0) {
1958 goto failed_removal;
1960 pr_info("Offlined Pages %ld\n", offlined_pages);
1961 /* Ok, all of our target is isolated.
1962 We cannot do rollback at this point. */
1963 offline_isolated_pages(start_pfn, end_pfn);
1964 /* reset pagetype flags and makes migrate type to be MOVABLE */
1965 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1966 /* removal success */
1967 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1968 zone->present_pages -= offlined_pages;
1970 pgdat_resize_lock(zone->zone_pgdat, &flags);
1971 zone->zone_pgdat->node_present_pages -= offlined_pages;
1972 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1974 init_per_zone_wmark_min();
1976 if (!populated_zone(zone)) {
1977 zone_pcp_reset(zone);
1978 mutex_lock(&zonelists_mutex);
1979 build_all_zonelists(NULL, NULL);
1980 mutex_unlock(&zonelists_mutex);
1982 zone_pcp_update(zone);
1984 node_states_clear_node(node, &arg);
1985 if (arg.status_change_nid >= 0) {
1987 kcompactd_stop(node);
1990 vm_total_pages = nr_free_pagecache_pages();
1991 writeback_set_ratelimit();
1993 memory_notify(MEM_OFFLINE, &arg);
1997 pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n",
1998 (unsigned long long) start_pfn << PAGE_SHIFT,
1999 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
2000 memory_notify(MEM_CANCEL_OFFLINE, &arg);
2001 /* pushback to free area */
2002 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
2006 /* Must be protected by mem_hotplug_begin() */
2007 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
2009 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
2011 #endif /* CONFIG_MEMORY_HOTREMOVE */
2014 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
2015 * @start_pfn: start pfn of the memory range
2016 * @end_pfn: end pfn of the memory range
2017 * @arg: argument passed to func
2018 * @func: callback for each memory section walked
2020 * This function walks through all present mem sections in range
2021 * [start_pfn, end_pfn) and call func on each mem section.
2023 * Returns the return value of func.
2025 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
2026 void *arg, int (*func)(struct memory_block *, void *))
2028 struct memory_block *mem = NULL;
2029 struct mem_section *section;
2030 unsigned long pfn, section_nr;
2033 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2034 section_nr = pfn_to_section_nr(pfn);
2035 if (!present_section_nr(section_nr))
2038 section = __nr_to_section(section_nr);
2039 /* same memblock? */
2041 if ((section_nr >= mem->start_section_nr) &&
2042 (section_nr <= mem->end_section_nr))
2045 mem = find_memory_block_hinted(section, mem);
2049 ret = func(mem, arg);
2051 kobject_put(&mem->dev.kobj);
2057 kobject_put(&mem->dev.kobj);
2062 #ifdef CONFIG_MEMORY_HOTREMOVE
2063 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
2065 int ret = !is_memblock_offlined(mem);
2067 if (unlikely(ret)) {
2068 phys_addr_t beginpa, endpa;
2070 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
2071 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
2072 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
2079 static int check_cpu_on_node(pg_data_t *pgdat)
2083 for_each_present_cpu(cpu) {
2084 if (cpu_to_node(cpu) == pgdat->node_id)
2086 * the cpu on this node isn't removed, and we can't
2087 * offline this node.
2095 static void unmap_cpu_on_node(pg_data_t *pgdat)
2097 #ifdef CONFIG_ACPI_NUMA
2100 for_each_possible_cpu(cpu)
2101 if (cpu_to_node(cpu) == pgdat->node_id)
2102 numa_clear_node(cpu);
2106 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
2110 ret = check_cpu_on_node(pgdat);
2115 * the node will be offlined when we come here, so we can clear
2116 * the cpu_to_node() now.
2119 unmap_cpu_on_node(pgdat);
2126 * Offline a node if all memory sections and cpus of the node are removed.
2128 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2129 * and online/offline operations before this call.
2131 void try_offline_node(int nid)
2133 pg_data_t *pgdat = NODE_DATA(nid);
2134 unsigned long start_pfn = pgdat->node_start_pfn;
2135 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
2138 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2139 unsigned long section_nr = pfn_to_section_nr(pfn);
2141 if (!present_section_nr(section_nr))
2144 if (pfn_to_nid(pfn) != nid)
2148 * some memory sections of this node are not removed, and we
2149 * can't offline node now.
2154 if (check_and_unmap_cpu_on_node(pgdat))
2158 * all memory/cpu of this node are removed, we can offline this
2161 node_set_offline(nid);
2162 unregister_one_node(nid);
2164 EXPORT_SYMBOL(try_offline_node);
2169 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2170 * and online/offline operations before this call, as required by
2171 * try_offline_node().
2173 void __ref remove_memory(int nid, u64 start, u64 size)
2177 BUG_ON(check_hotplug_memory_range(start, size));
2179 mem_hotplug_begin();
2182 * All memory blocks must be offlined before removing memory. Check
2183 * whether all memory blocks in question are offline and trigger a BUG()
2184 * if this is not the case.
2186 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
2187 check_memblock_offlined_cb);
2191 /* remove memmap entry */
2192 firmware_map_remove(start, start + size, "System RAM");
2193 memblock_free(start, size);
2194 memblock_remove(start, size);
2196 arch_remove_memory(start, size);
2198 try_offline_node(nid);
2202 EXPORT_SYMBOL_GPL(remove_memory);
2203 #endif /* CONFIG_MEMORY_HOTREMOVE */