Version:  2.0.40 2.2.26 2.4.37 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 4.10

Linux/mm/memory_hotplug.c

  1 /*
  2  *  linux/mm/memory_hotplug.c
  3  *
  4  *  Copyright (C)
  5  */
  6 
  7 #include <linux/stddef.h>
  8 #include <linux/mm.h>
  9 #include <linux/swap.h>
 10 #include <linux/interrupt.h>
 11 #include <linux/pagemap.h>
 12 #include <linux/compiler.h>
 13 #include <linux/export.h>
 14 #include <linux/pagevec.h>
 15 #include <linux/writeback.h>
 16 #include <linux/slab.h>
 17 #include <linux/sysctl.h>
 18 #include <linux/cpu.h>
 19 #include <linux/memory.h>
 20 #include <linux/memremap.h>
 21 #include <linux/memory_hotplug.h>
 22 #include <linux/highmem.h>
 23 #include <linux/vmalloc.h>
 24 #include <linux/ioport.h>
 25 #include <linux/delay.h>
 26 #include <linux/migrate.h>
 27 #include <linux/page-isolation.h>
 28 #include <linux/pfn.h>
 29 #include <linux/suspend.h>
 30 #include <linux/mm_inline.h>
 31 #include <linux/firmware-map.h>
 32 #include <linux/stop_machine.h>
 33 #include <linux/hugetlb.h>
 34 #include <linux/memblock.h>
 35 #include <linux/bootmem.h>
 36 #include <linux/compaction.h>
 37 
 38 #include <asm/tlbflush.h>
 39 
 40 #include "internal.h"
 41 
 42 /*
 43  * online_page_callback contains pointer to current page onlining function.
 44  * Initially it is generic_online_page(). If it is required it could be
 45  * changed by calling set_online_page_callback() for callback registration
 46  * and restore_online_page_callback() for generic callback restore.
 47  */
 48 
 49 static void generic_online_page(struct page *page);
 50 
 51 static online_page_callback_t online_page_callback = generic_online_page;
 52 static DEFINE_MUTEX(online_page_callback_lock);
 53 
 54 /* The same as the cpu_hotplug lock, but for memory hotplug. */
 55 static struct {
 56         struct task_struct *active_writer;
 57         struct mutex lock; /* Synchronizes accesses to refcount, */
 58         /*
 59          * Also blocks the new readers during
 60          * an ongoing mem hotplug operation.
 61          */
 62         int refcount;
 63 
 64 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 65         struct lockdep_map dep_map;
 66 #endif
 67 } mem_hotplug = {
 68         .active_writer = NULL,
 69         .lock = __MUTEX_INITIALIZER(mem_hotplug.lock),
 70         .refcount = 0,
 71 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 72         .dep_map = {.name = "mem_hotplug.lock" },
 73 #endif
 74 };
 75 
 76 /* Lockdep annotations for get/put_online_mems() and mem_hotplug_begin/end() */
 77 #define memhp_lock_acquire_read() lock_map_acquire_read(&mem_hotplug.dep_map)
 78 #define memhp_lock_acquire()      lock_map_acquire(&mem_hotplug.dep_map)
 79 #define memhp_lock_release()      lock_map_release(&mem_hotplug.dep_map)
 80 
 81 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
 82 bool memhp_auto_online;
 83 #else
 84 bool memhp_auto_online = true;
 85 #endif
 86 EXPORT_SYMBOL_GPL(memhp_auto_online);
 87 
 88 static int __init setup_memhp_default_state(char *str)
 89 {
 90         if (!strcmp(str, "online"))
 91                 memhp_auto_online = true;
 92         else if (!strcmp(str, "offline"))
 93                 memhp_auto_online = false;
 94 
 95         return 1;
 96 }
 97 __setup("memhp_default_state=", setup_memhp_default_state);
 98 
 99 void get_online_mems(void)
100 {
101         might_sleep();
102         if (mem_hotplug.active_writer == current)
103                 return;
104         memhp_lock_acquire_read();
105         mutex_lock(&mem_hotplug.lock);
106         mem_hotplug.refcount++;
107         mutex_unlock(&mem_hotplug.lock);
108 
109 }
110 
111 void put_online_mems(void)
112 {
113         if (mem_hotplug.active_writer == current)
114                 return;
115         mutex_lock(&mem_hotplug.lock);
116 
117         if (WARN_ON(!mem_hotplug.refcount))
118                 mem_hotplug.refcount++; /* try to fix things up */
119 
120         if (!--mem_hotplug.refcount && unlikely(mem_hotplug.active_writer))
121                 wake_up_process(mem_hotplug.active_writer);
122         mutex_unlock(&mem_hotplug.lock);
123         memhp_lock_release();
124 
125 }
126 
127 void mem_hotplug_begin(void)
128 {
129         mem_hotplug.active_writer = current;
130 
131         memhp_lock_acquire();
132         for (;;) {
133                 mutex_lock(&mem_hotplug.lock);
134                 if (likely(!mem_hotplug.refcount))
135                         break;
136                 __set_current_state(TASK_UNINTERRUPTIBLE);
137                 mutex_unlock(&mem_hotplug.lock);
138                 schedule();
139         }
140 }
141 
142 void mem_hotplug_done(void)
143 {
144         mem_hotplug.active_writer = NULL;
145         mutex_unlock(&mem_hotplug.lock);
146         memhp_lock_release();
147 }
148 
149 /* add this memory to iomem resource */
150 static struct resource *register_memory_resource(u64 start, u64 size)
151 {
152         struct resource *res;
153         res = kzalloc(sizeof(struct resource), GFP_KERNEL);
154         if (!res)
155                 return ERR_PTR(-ENOMEM);
156 
157         res->name = "System RAM";
158         res->start = start;
159         res->end = start + size - 1;
160         res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
161         if (request_resource(&iomem_resource, res) < 0) {
162                 pr_debug("System RAM resource %pR cannot be added\n", res);
163                 kfree(res);
164                 return ERR_PTR(-EEXIST);
165         }
166         return res;
167 }
168 
169 static void release_memory_resource(struct resource *res)
170 {
171         if (!res)
172                 return;
173         release_resource(res);
174         kfree(res);
175         return;
176 }
177 
178 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
179 void get_page_bootmem(unsigned long info,  struct page *page,
180                       unsigned long type)
181 {
182         page->lru.next = (struct list_head *) type;
183         SetPagePrivate(page);
184         set_page_private(page, info);
185         page_ref_inc(page);
186 }
187 
188 void put_page_bootmem(struct page *page)
189 {
190         unsigned long type;
191 
192         type = (unsigned long) page->lru.next;
193         BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
194                type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
195 
196         if (page_ref_dec_return(page) == 1) {
197                 ClearPagePrivate(page);
198                 set_page_private(page, 0);
199                 INIT_LIST_HEAD(&page->lru);
200                 free_reserved_page(page);
201         }
202 }
203 
204 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
205 #ifndef CONFIG_SPARSEMEM_VMEMMAP
206 static void register_page_bootmem_info_section(unsigned long start_pfn)
207 {
208         unsigned long *usemap, mapsize, section_nr, i;
209         struct mem_section *ms;
210         struct page *page, *memmap;
211 
212         section_nr = pfn_to_section_nr(start_pfn);
213         ms = __nr_to_section(section_nr);
214 
215         /* Get section's memmap address */
216         memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
217 
218         /*
219          * Get page for the memmap's phys address
220          * XXX: need more consideration for sparse_vmemmap...
221          */
222         page = virt_to_page(memmap);
223         mapsize = sizeof(struct page) * PAGES_PER_SECTION;
224         mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
225 
226         /* remember memmap's page */
227         for (i = 0; i < mapsize; i++, page++)
228                 get_page_bootmem(section_nr, page, SECTION_INFO);
229 
230         usemap = __nr_to_section(section_nr)->pageblock_flags;
231         page = virt_to_page(usemap);
232 
233         mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
234 
235         for (i = 0; i < mapsize; i++, page++)
236                 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
237 
238 }
239 #else /* CONFIG_SPARSEMEM_VMEMMAP */
240 static void register_page_bootmem_info_section(unsigned long start_pfn)
241 {
242         unsigned long *usemap, mapsize, section_nr, i;
243         struct mem_section *ms;
244         struct page *page, *memmap;
245 
246         if (!pfn_valid(start_pfn))
247                 return;
248 
249         section_nr = pfn_to_section_nr(start_pfn);
250         ms = __nr_to_section(section_nr);
251 
252         memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
253 
254         register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
255 
256         usemap = __nr_to_section(section_nr)->pageblock_flags;
257         page = virt_to_page(usemap);
258 
259         mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
260 
261         for (i = 0; i < mapsize; i++, page++)
262                 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
263 }
264 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
265 
266 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
267 {
268         unsigned long i, pfn, end_pfn, nr_pages;
269         int node = pgdat->node_id;
270         struct page *page;
271 
272         nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
273         page = virt_to_page(pgdat);
274 
275         for (i = 0; i < nr_pages; i++, page++)
276                 get_page_bootmem(node, page, NODE_INFO);
277 
278         pfn = pgdat->node_start_pfn;
279         end_pfn = pgdat_end_pfn(pgdat);
280 
281         /* register section info */
282         for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
283                 /*
284                  * Some platforms can assign the same pfn to multiple nodes - on
285                  * node0 as well as nodeN.  To avoid registering a pfn against
286                  * multiple nodes we check that this pfn does not already
287                  * reside in some other nodes.
288                  */
289                 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
290                         register_page_bootmem_info_section(pfn);
291         }
292 }
293 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
294 
295 static void __meminit grow_zone_span(struct zone *zone, unsigned long start_pfn,
296                                      unsigned long end_pfn)
297 {
298         unsigned long old_zone_end_pfn;
299 
300         zone_span_writelock(zone);
301 
302         old_zone_end_pfn = zone_end_pfn(zone);
303         if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
304                 zone->zone_start_pfn = start_pfn;
305 
306         zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
307                                 zone->zone_start_pfn;
308 
309         zone_span_writeunlock(zone);
310 }
311 
312 static void resize_zone(struct zone *zone, unsigned long start_pfn,
313                 unsigned long end_pfn)
314 {
315         zone_span_writelock(zone);
316 
317         if (end_pfn - start_pfn) {
318                 zone->zone_start_pfn = start_pfn;
319                 zone->spanned_pages = end_pfn - start_pfn;
320         } else {
321                 /*
322                  * make it consist as free_area_init_core(),
323                  * if spanned_pages = 0, then keep start_pfn = 0
324                  */
325                 zone->zone_start_pfn = 0;
326                 zone->spanned_pages = 0;
327         }
328 
329         zone_span_writeunlock(zone);
330 }
331 
332 static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
333                 unsigned long end_pfn)
334 {
335         enum zone_type zid = zone_idx(zone);
336         int nid = zone->zone_pgdat->node_id;
337         unsigned long pfn;
338 
339         for (pfn = start_pfn; pfn < end_pfn; pfn++)
340                 set_page_links(pfn_to_page(pfn), zid, nid, pfn);
341 }
342 
343 /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
344  * alloc_bootmem_node_nopanic()/memblock_virt_alloc_node_nopanic() */
345 static int __ref ensure_zone_is_initialized(struct zone *zone,
346                         unsigned long start_pfn, unsigned long num_pages)
347 {
348         if (!zone_is_initialized(zone))
349                 return init_currently_empty_zone(zone, start_pfn, num_pages);
350 
351         return 0;
352 }
353 
354 static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
355                 unsigned long start_pfn, unsigned long end_pfn)
356 {
357         int ret;
358         unsigned long flags;
359         unsigned long z1_start_pfn;
360 
361         ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn);
362         if (ret)
363                 return ret;
364 
365         pgdat_resize_lock(z1->zone_pgdat, &flags);
366 
367         /* can't move pfns which are higher than @z2 */
368         if (end_pfn > zone_end_pfn(z2))
369                 goto out_fail;
370         /* the move out part must be at the left most of @z2 */
371         if (start_pfn > z2->zone_start_pfn)
372                 goto out_fail;
373         /* must included/overlap */
374         if (end_pfn <= z2->zone_start_pfn)
375                 goto out_fail;
376 
377         /* use start_pfn for z1's start_pfn if z1 is empty */
378         if (!zone_is_empty(z1))
379                 z1_start_pfn = z1->zone_start_pfn;
380         else
381                 z1_start_pfn = start_pfn;
382 
383         resize_zone(z1, z1_start_pfn, end_pfn);
384         resize_zone(z2, end_pfn, zone_end_pfn(z2));
385 
386         pgdat_resize_unlock(z1->zone_pgdat, &flags);
387 
388         fix_zone_id(z1, start_pfn, end_pfn);
389 
390         return 0;
391 out_fail:
392         pgdat_resize_unlock(z1->zone_pgdat, &flags);
393         return -1;
394 }
395 
396 static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
397                 unsigned long start_pfn, unsigned long end_pfn)
398 {
399         int ret;
400         unsigned long flags;
401         unsigned long z2_end_pfn;
402 
403         ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn);
404         if (ret)
405                 return ret;
406 
407         pgdat_resize_lock(z1->zone_pgdat, &flags);
408 
409         /* can't move pfns which are lower than @z1 */
410         if (z1->zone_start_pfn > start_pfn)
411                 goto out_fail;
412         /* the move out part mast at the right most of @z1 */
413         if (zone_end_pfn(z1) >  end_pfn)
414                 goto out_fail;
415         /* must included/overlap */
416         if (start_pfn >= zone_end_pfn(z1))
417                 goto out_fail;
418 
419         /* use end_pfn for z2's end_pfn if z2 is empty */
420         if (!zone_is_empty(z2))
421                 z2_end_pfn = zone_end_pfn(z2);
422         else
423                 z2_end_pfn = end_pfn;
424 
425         resize_zone(z1, z1->zone_start_pfn, start_pfn);
426         resize_zone(z2, start_pfn, z2_end_pfn);
427 
428         pgdat_resize_unlock(z1->zone_pgdat, &flags);
429 
430         fix_zone_id(z2, start_pfn, end_pfn);
431 
432         return 0;
433 out_fail:
434         pgdat_resize_unlock(z1->zone_pgdat, &flags);
435         return -1;
436 }
437 
438 static struct zone * __meminit move_pfn_range(int zone_shift,
439                 unsigned long start_pfn, unsigned long end_pfn)
440 {
441         struct zone *zone = page_zone(pfn_to_page(start_pfn));
442         int ret = 0;
443 
444         if (zone_shift < 0)
445                 ret = move_pfn_range_left(zone + zone_shift, zone,
446                                           start_pfn, end_pfn);
447         else if (zone_shift)
448                 ret = move_pfn_range_right(zone, zone + zone_shift,
449                                            start_pfn, end_pfn);
450 
451         if (ret)
452                 return NULL;
453 
454         return zone + zone_shift;
455 }
456 
457 static void __meminit grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
458                                       unsigned long end_pfn)
459 {
460         unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat);
461 
462         if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
463                 pgdat->node_start_pfn = start_pfn;
464 
465         pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
466                                         pgdat->node_start_pfn;
467 }
468 
469 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
470 {
471         struct pglist_data *pgdat = zone->zone_pgdat;
472         int nr_pages = PAGES_PER_SECTION;
473         int nid = pgdat->node_id;
474         int zone_type;
475         unsigned long flags, pfn;
476         int ret;
477 
478         zone_type = zone - pgdat->node_zones;
479         ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
480         if (ret)
481                 return ret;
482 
483         pgdat_resize_lock(zone->zone_pgdat, &flags);
484         grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
485         grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
486                         phys_start_pfn + nr_pages);
487         pgdat_resize_unlock(zone->zone_pgdat, &flags);
488         memmap_init_zone(nr_pages, nid, zone_type,
489                          phys_start_pfn, MEMMAP_HOTPLUG);
490 
491         /* online_page_range is called later and expects pages reserved */
492         for (pfn = phys_start_pfn; pfn < phys_start_pfn + nr_pages; pfn++) {
493                 if (!pfn_valid(pfn))
494                         continue;
495 
496                 SetPageReserved(pfn_to_page(pfn));
497         }
498         return 0;
499 }
500 
501 static int __meminit __add_section(int nid, struct zone *zone,
502                                         unsigned long phys_start_pfn)
503 {
504         int ret;
505 
506         if (pfn_valid(phys_start_pfn))
507                 return -EEXIST;
508 
509         ret = sparse_add_one_section(zone, phys_start_pfn);
510 
511         if (ret < 0)
512                 return ret;
513 
514         ret = __add_zone(zone, phys_start_pfn);
515 
516         if (ret < 0)
517                 return ret;
518 
519         return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
520 }
521 
522 /*
523  * Reasonably generic function for adding memory.  It is
524  * expected that archs that support memory hotplug will
525  * call this function after deciding the zone to which to
526  * add the new pages.
527  */
528 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
529                         unsigned long nr_pages)
530 {
531         unsigned long i;
532         int err = 0;
533         int start_sec, end_sec;
534         struct vmem_altmap *altmap;
535 
536         clear_zone_contiguous(zone);
537 
538         /* during initialize mem_map, align hot-added range to section */
539         start_sec = pfn_to_section_nr(phys_start_pfn);
540         end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
541 
542         altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn));
543         if (altmap) {
544                 /*
545                  * Validate altmap is within bounds of the total request
546                  */
547                 if (altmap->base_pfn != phys_start_pfn
548                                 || vmem_altmap_offset(altmap) > nr_pages) {
549                         pr_warn_once("memory add fail, invalid altmap\n");
550                         err = -EINVAL;
551                         goto out;
552                 }
553                 altmap->alloc = 0;
554         }
555 
556         for (i = start_sec; i <= end_sec; i++) {
557                 err = __add_section(nid, zone, section_nr_to_pfn(i));
558 
559                 /*
560                  * EEXIST is finally dealt with by ioresource collision
561                  * check. see add_memory() => register_memory_resource()
562                  * Warning will be printed if there is collision.
563                  */
564                 if (err && (err != -EEXIST))
565                         break;
566                 err = 0;
567         }
568         vmemmap_populate_print_last();
569 out:
570         set_zone_contiguous(zone);
571         return err;
572 }
573 EXPORT_SYMBOL_GPL(__add_pages);
574 
575 #ifdef CONFIG_MEMORY_HOTREMOVE
576 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
577 static int find_smallest_section_pfn(int nid, struct zone *zone,
578                                      unsigned long start_pfn,
579                                      unsigned long end_pfn)
580 {
581         struct mem_section *ms;
582 
583         for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
584                 ms = __pfn_to_section(start_pfn);
585 
586                 if (unlikely(!valid_section(ms)))
587                         continue;
588 
589                 if (unlikely(pfn_to_nid(start_pfn) != nid))
590                         continue;
591 
592                 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
593                         continue;
594 
595                 return start_pfn;
596         }
597 
598         return 0;
599 }
600 
601 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
602 static int find_biggest_section_pfn(int nid, struct zone *zone,
603                                     unsigned long start_pfn,
604                                     unsigned long end_pfn)
605 {
606         struct mem_section *ms;
607         unsigned long pfn;
608 
609         /* pfn is the end pfn of a memory section. */
610         pfn = end_pfn - 1;
611         for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
612                 ms = __pfn_to_section(pfn);
613 
614                 if (unlikely(!valid_section(ms)))
615                         continue;
616 
617                 if (unlikely(pfn_to_nid(pfn) != nid))
618                         continue;
619 
620                 if (zone && zone != page_zone(pfn_to_page(pfn)))
621                         continue;
622 
623                 return pfn;
624         }
625 
626         return 0;
627 }
628 
629 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
630                              unsigned long end_pfn)
631 {
632         unsigned long zone_start_pfn = zone->zone_start_pfn;
633         unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
634         unsigned long zone_end_pfn = z;
635         unsigned long pfn;
636         struct mem_section *ms;
637         int nid = zone_to_nid(zone);
638 
639         zone_span_writelock(zone);
640         if (zone_start_pfn == start_pfn) {
641                 /*
642                  * If the section is smallest section in the zone, it need
643                  * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
644                  * In this case, we find second smallest valid mem_section
645                  * for shrinking zone.
646                  */
647                 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
648                                                 zone_end_pfn);
649                 if (pfn) {
650                         zone->zone_start_pfn = pfn;
651                         zone->spanned_pages = zone_end_pfn - pfn;
652                 }
653         } else if (zone_end_pfn == end_pfn) {
654                 /*
655                  * If the section is biggest section in the zone, it need
656                  * shrink zone->spanned_pages.
657                  * In this case, we find second biggest valid mem_section for
658                  * shrinking zone.
659                  */
660                 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
661                                                start_pfn);
662                 if (pfn)
663                         zone->spanned_pages = pfn - zone_start_pfn + 1;
664         }
665 
666         /*
667          * The section is not biggest or smallest mem_section in the zone, it
668          * only creates a hole in the zone. So in this case, we need not
669          * change the zone. But perhaps, the zone has only hole data. Thus
670          * it check the zone has only hole or not.
671          */
672         pfn = zone_start_pfn;
673         for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
674                 ms = __pfn_to_section(pfn);
675 
676                 if (unlikely(!valid_section(ms)))
677                         continue;
678 
679                 if (page_zone(pfn_to_page(pfn)) != zone)
680                         continue;
681 
682                  /* If the section is current section, it continues the loop */
683                 if (start_pfn == pfn)
684                         continue;
685 
686                 /* If we find valid section, we have nothing to do */
687                 zone_span_writeunlock(zone);
688                 return;
689         }
690 
691         /* The zone has no valid section */
692         zone->zone_start_pfn = 0;
693         zone->spanned_pages = 0;
694         zone_span_writeunlock(zone);
695 }
696 
697 static void shrink_pgdat_span(struct pglist_data *pgdat,
698                               unsigned long start_pfn, unsigned long end_pfn)
699 {
700         unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
701         unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
702         unsigned long pgdat_end_pfn = p;
703         unsigned long pfn;
704         struct mem_section *ms;
705         int nid = pgdat->node_id;
706 
707         if (pgdat_start_pfn == start_pfn) {
708                 /*
709                  * If the section is smallest section in the pgdat, it need
710                  * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
711                  * In this case, we find second smallest valid mem_section
712                  * for shrinking zone.
713                  */
714                 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
715                                                 pgdat_end_pfn);
716                 if (pfn) {
717                         pgdat->node_start_pfn = pfn;
718                         pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
719                 }
720         } else if (pgdat_end_pfn == end_pfn) {
721                 /*
722                  * If the section is biggest section in the pgdat, it need
723                  * shrink pgdat->node_spanned_pages.
724                  * In this case, we find second biggest valid mem_section for
725                  * shrinking zone.
726                  */
727                 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
728                                                start_pfn);
729                 if (pfn)
730                         pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
731         }
732 
733         /*
734          * If the section is not biggest or smallest mem_section in the pgdat,
735          * it only creates a hole in the pgdat. So in this case, we need not
736          * change the pgdat.
737          * But perhaps, the pgdat has only hole data. Thus it check the pgdat
738          * has only hole or not.
739          */
740         pfn = pgdat_start_pfn;
741         for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
742                 ms = __pfn_to_section(pfn);
743 
744                 if (unlikely(!valid_section(ms)))
745                         continue;
746 
747                 if (pfn_to_nid(pfn) != nid)
748                         continue;
749 
750                  /* If the section is current section, it continues the loop */
751                 if (start_pfn == pfn)
752                         continue;
753 
754                 /* If we find valid section, we have nothing to do */
755                 return;
756         }
757 
758         /* The pgdat has no valid section */
759         pgdat->node_start_pfn = 0;
760         pgdat->node_spanned_pages = 0;
761 }
762 
763 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
764 {
765         struct pglist_data *pgdat = zone->zone_pgdat;
766         int nr_pages = PAGES_PER_SECTION;
767         int zone_type;
768         unsigned long flags;
769 
770         zone_type = zone - pgdat->node_zones;
771 
772         pgdat_resize_lock(zone->zone_pgdat, &flags);
773         shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
774         shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
775         pgdat_resize_unlock(zone->zone_pgdat, &flags);
776 }
777 
778 static int __remove_section(struct zone *zone, struct mem_section *ms,
779                 unsigned long map_offset)
780 {
781         unsigned long start_pfn;
782         int scn_nr;
783         int ret = -EINVAL;
784 
785         if (!valid_section(ms))
786                 return ret;
787 
788         ret = unregister_memory_section(ms);
789         if (ret)
790                 return ret;
791 
792         scn_nr = __section_nr(ms);
793         start_pfn = section_nr_to_pfn(scn_nr);
794         __remove_zone(zone, start_pfn);
795 
796         sparse_remove_one_section(zone, ms, map_offset);
797         return 0;
798 }
799 
800 /**
801  * __remove_pages() - remove sections of pages from a zone
802  * @zone: zone from which pages need to be removed
803  * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
804  * @nr_pages: number of pages to remove (must be multiple of section size)
805  *
806  * Generic helper function to remove section mappings and sysfs entries
807  * for the section of the memory we are removing. Caller needs to make
808  * sure that pages are marked reserved and zones are adjust properly by
809  * calling offline_pages().
810  */
811 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
812                  unsigned long nr_pages)
813 {
814         unsigned long i;
815         unsigned long map_offset = 0;
816         int sections_to_remove, ret = 0;
817 
818         /* In the ZONE_DEVICE case device driver owns the memory region */
819         if (is_dev_zone(zone)) {
820                 struct page *page = pfn_to_page(phys_start_pfn);
821                 struct vmem_altmap *altmap;
822 
823                 altmap = to_vmem_altmap((unsigned long) page);
824                 if (altmap)
825                         map_offset = vmem_altmap_offset(altmap);
826         } else {
827                 resource_size_t start, size;
828 
829                 start = phys_start_pfn << PAGE_SHIFT;
830                 size = nr_pages * PAGE_SIZE;
831 
832                 ret = release_mem_region_adjustable(&iomem_resource, start,
833                                         size);
834                 if (ret) {
835                         resource_size_t endres = start + size - 1;
836 
837                         pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
838                                         &start, &endres, ret);
839                 }
840         }
841 
842         clear_zone_contiguous(zone);
843 
844         /*
845          * We can only remove entire sections
846          */
847         BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
848         BUG_ON(nr_pages % PAGES_PER_SECTION);
849 
850         sections_to_remove = nr_pages / PAGES_PER_SECTION;
851         for (i = 0; i < sections_to_remove; i++) {
852                 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
853 
854                 ret = __remove_section(zone, __pfn_to_section(pfn), map_offset);
855                 map_offset = 0;
856                 if (ret)
857                         break;
858         }
859 
860         set_zone_contiguous(zone);
861 
862         return ret;
863 }
864 EXPORT_SYMBOL_GPL(__remove_pages);
865 #endif /* CONFIG_MEMORY_HOTREMOVE */
866 
867 int set_online_page_callback(online_page_callback_t callback)
868 {
869         int rc = -EINVAL;
870 
871         get_online_mems();
872         mutex_lock(&online_page_callback_lock);
873 
874         if (online_page_callback == generic_online_page) {
875                 online_page_callback = callback;
876                 rc = 0;
877         }
878 
879         mutex_unlock(&online_page_callback_lock);
880         put_online_mems();
881 
882         return rc;
883 }
884 EXPORT_SYMBOL_GPL(set_online_page_callback);
885 
886 int restore_online_page_callback(online_page_callback_t callback)
887 {
888         int rc = -EINVAL;
889 
890         get_online_mems();
891         mutex_lock(&online_page_callback_lock);
892 
893         if (online_page_callback == callback) {
894                 online_page_callback = generic_online_page;
895                 rc = 0;
896         }
897 
898         mutex_unlock(&online_page_callback_lock);
899         put_online_mems();
900 
901         return rc;
902 }
903 EXPORT_SYMBOL_GPL(restore_online_page_callback);
904 
905 void __online_page_set_limits(struct page *page)
906 {
907 }
908 EXPORT_SYMBOL_GPL(__online_page_set_limits);
909 
910 void __online_page_increment_counters(struct page *page)
911 {
912         adjust_managed_page_count(page, 1);
913 }
914 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
915 
916 void __online_page_free(struct page *page)
917 {
918         __free_reserved_page(page);
919 }
920 EXPORT_SYMBOL_GPL(__online_page_free);
921 
922 static void generic_online_page(struct page *page)
923 {
924         __online_page_set_limits(page);
925         __online_page_increment_counters(page);
926         __online_page_free(page);
927 }
928 
929 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
930                         void *arg)
931 {
932         unsigned long i;
933         unsigned long onlined_pages = *(unsigned long *)arg;
934         struct page *page;
935         if (PageReserved(pfn_to_page(start_pfn)))
936                 for (i = 0; i < nr_pages; i++) {
937                         page = pfn_to_page(start_pfn + i);
938                         (*online_page_callback)(page);
939                         onlined_pages++;
940                 }
941         *(unsigned long *)arg = onlined_pages;
942         return 0;
943 }
944 
945 #ifdef CONFIG_MOVABLE_NODE
946 /*
947  * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
948  * normal memory.
949  */
950 static bool can_online_high_movable(struct zone *zone)
951 {
952         return true;
953 }
954 #else /* CONFIG_MOVABLE_NODE */
955 /* ensure every online node has NORMAL memory */
956 static bool can_online_high_movable(struct zone *zone)
957 {
958         return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
959 }
960 #endif /* CONFIG_MOVABLE_NODE */
961 
962 /* check which state of node_states will be changed when online memory */
963 static void node_states_check_changes_online(unsigned long nr_pages,
964         struct zone *zone, struct memory_notify *arg)
965 {
966         int nid = zone_to_nid(zone);
967         enum zone_type zone_last = ZONE_NORMAL;
968 
969         /*
970          * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
971          * contains nodes which have zones of 0...ZONE_NORMAL,
972          * set zone_last to ZONE_NORMAL.
973          *
974          * If we don't have HIGHMEM nor movable node,
975          * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
976          * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
977          */
978         if (N_MEMORY == N_NORMAL_MEMORY)
979                 zone_last = ZONE_MOVABLE;
980 
981         /*
982          * if the memory to be online is in a zone of 0...zone_last, and
983          * the zones of 0...zone_last don't have memory before online, we will
984          * need to set the node to node_states[N_NORMAL_MEMORY] after
985          * the memory is online.
986          */
987         if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
988                 arg->status_change_nid_normal = nid;
989         else
990                 arg->status_change_nid_normal = -1;
991 
992 #ifdef CONFIG_HIGHMEM
993         /*
994          * If we have movable node, node_states[N_HIGH_MEMORY]
995          * contains nodes which have zones of 0...ZONE_HIGHMEM,
996          * set zone_last to ZONE_HIGHMEM.
997          *
998          * If we don't have movable node, node_states[N_NORMAL_MEMORY]
999          * contains nodes which have zones of 0...ZONE_MOVABLE,
1000          * set zone_last to ZONE_MOVABLE.
1001          */
1002         zone_last = ZONE_HIGHMEM;
1003         if (N_MEMORY == N_HIGH_MEMORY)
1004                 zone_last = ZONE_MOVABLE;
1005 
1006         if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
1007                 arg->status_change_nid_high = nid;
1008         else
1009                 arg->status_change_nid_high = -1;
1010 #else
1011         arg->status_change_nid_high = arg->status_change_nid_normal;
1012 #endif
1013 
1014         /*
1015          * if the node don't have memory befor online, we will need to
1016          * set the node to node_states[N_MEMORY] after the memory
1017          * is online.
1018          */
1019         if (!node_state(nid, N_MEMORY))
1020                 arg->status_change_nid = nid;
1021         else
1022                 arg->status_change_nid = -1;
1023 }
1024 
1025 static void node_states_set_node(int node, struct memory_notify *arg)
1026 {
1027         if (arg->status_change_nid_normal >= 0)
1028                 node_set_state(node, N_NORMAL_MEMORY);
1029 
1030         if (arg->status_change_nid_high >= 0)
1031                 node_set_state(node, N_HIGH_MEMORY);
1032 
1033         node_set_state(node, N_MEMORY);
1034 }
1035 
1036 bool zone_can_shift(unsigned long pfn, unsigned long nr_pages,
1037                    enum zone_type target, int *zone_shift)
1038 {
1039         struct zone *zone = page_zone(pfn_to_page(pfn));
1040         enum zone_type idx = zone_idx(zone);
1041         int i;
1042 
1043         *zone_shift = 0;
1044 
1045         if (idx < target) {
1046                 /* pages must be at end of current zone */
1047                 if (pfn + nr_pages != zone_end_pfn(zone))
1048                         return false;
1049 
1050                 /* no zones in use between current zone and target */
1051                 for (i = idx + 1; i < target; i++)
1052                         if (zone_is_initialized(zone - idx + i))
1053                                 return false;
1054         }
1055 
1056         if (target < idx) {
1057                 /* pages must be at beginning of current zone */
1058                 if (pfn != zone->zone_start_pfn)
1059                         return false;
1060 
1061                 /* no zones in use between current zone and target */
1062                 for (i = target + 1; i < idx; i++)
1063                         if (zone_is_initialized(zone - idx + i))
1064                                 return false;
1065         }
1066 
1067         *zone_shift = target - idx;
1068         return true;
1069 }
1070 
1071 /* Must be protected by mem_hotplug_begin() */
1072 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
1073 {
1074         unsigned long flags;
1075         unsigned long onlined_pages = 0;
1076         struct zone *zone;
1077         int need_zonelists_rebuild = 0;
1078         int nid;
1079         int ret;
1080         struct memory_notify arg;
1081         int zone_shift = 0;
1082 
1083         /*
1084          * This doesn't need a lock to do pfn_to_page().
1085          * The section can't be removed here because of the
1086          * memory_block->state_mutex.
1087          */
1088         zone = page_zone(pfn_to_page(pfn));
1089 
1090         if ((zone_idx(zone) > ZONE_NORMAL ||
1091             online_type == MMOP_ONLINE_MOVABLE) &&
1092             !can_online_high_movable(zone))
1093                 return -EINVAL;
1094 
1095         if (online_type == MMOP_ONLINE_KERNEL) {
1096                 if (!zone_can_shift(pfn, nr_pages, ZONE_NORMAL, &zone_shift))
1097                         return -EINVAL;
1098         } else if (online_type == MMOP_ONLINE_MOVABLE) {
1099                 if (!zone_can_shift(pfn, nr_pages, ZONE_MOVABLE, &zone_shift))
1100                         return -EINVAL;
1101         }
1102 
1103         zone = move_pfn_range(zone_shift, pfn, pfn + nr_pages);
1104         if (!zone)
1105                 return -EINVAL;
1106 
1107         arg.start_pfn = pfn;
1108         arg.nr_pages = nr_pages;
1109         node_states_check_changes_online(nr_pages, zone, &arg);
1110 
1111         nid = zone_to_nid(zone);
1112 
1113         ret = memory_notify(MEM_GOING_ONLINE, &arg);
1114         ret = notifier_to_errno(ret);
1115         if (ret)
1116                 goto failed_addition;
1117 
1118         /*
1119          * If this zone is not populated, then it is not in zonelist.
1120          * This means the page allocator ignores this zone.
1121          * So, zonelist must be updated after online.
1122          */
1123         mutex_lock(&zonelists_mutex);
1124         if (!populated_zone(zone)) {
1125                 need_zonelists_rebuild = 1;
1126                 build_all_zonelists(NULL, zone);
1127         }
1128 
1129         ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
1130                 online_pages_range);
1131         if (ret) {
1132                 if (need_zonelists_rebuild)
1133                         zone_pcp_reset(zone);
1134                 mutex_unlock(&zonelists_mutex);
1135                 goto failed_addition;
1136         }
1137 
1138         zone->present_pages += onlined_pages;
1139 
1140         pgdat_resize_lock(zone->zone_pgdat, &flags);
1141         zone->zone_pgdat->node_present_pages += onlined_pages;
1142         pgdat_resize_unlock(zone->zone_pgdat, &flags);
1143 
1144         if (onlined_pages) {
1145                 node_states_set_node(nid, &arg);
1146                 if (need_zonelists_rebuild)
1147                         build_all_zonelists(NULL, NULL);
1148                 else
1149                         zone_pcp_update(zone);
1150         }
1151 
1152         mutex_unlock(&zonelists_mutex);
1153 
1154         init_per_zone_wmark_min();
1155 
1156         if (onlined_pages) {
1157                 kswapd_run(nid);
1158                 kcompactd_run(nid);
1159         }
1160 
1161         vm_total_pages = nr_free_pagecache_pages();
1162 
1163         writeback_set_ratelimit();
1164 
1165         if (onlined_pages)
1166                 memory_notify(MEM_ONLINE, &arg);
1167         return 0;
1168 
1169 failed_addition:
1170         pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
1171                  (unsigned long long) pfn << PAGE_SHIFT,
1172                  (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
1173         memory_notify(MEM_CANCEL_ONLINE, &arg);
1174         return ret;
1175 }
1176 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1177 
1178 static void reset_node_present_pages(pg_data_t *pgdat)
1179 {
1180         struct zone *z;
1181 
1182         for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
1183                 z->present_pages = 0;
1184 
1185         pgdat->node_present_pages = 0;
1186 }
1187 
1188 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1189 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1190 {
1191         struct pglist_data *pgdat;
1192         unsigned long zones_size[MAX_NR_ZONES] = {0};
1193         unsigned long zholes_size[MAX_NR_ZONES] = {0};
1194         unsigned long start_pfn = PFN_DOWN(start);
1195 
1196         pgdat = NODE_DATA(nid);
1197         if (!pgdat) {
1198                 pgdat = arch_alloc_nodedata(nid);
1199                 if (!pgdat)
1200                         return NULL;
1201 
1202                 arch_refresh_nodedata(nid, pgdat);
1203         } else {
1204                 /* Reset the nr_zones, order and classzone_idx before reuse */
1205                 pgdat->nr_zones = 0;
1206                 pgdat->kswapd_order = 0;
1207                 pgdat->kswapd_classzone_idx = 0;
1208         }
1209 
1210         /* we can use NODE_DATA(nid) from here */
1211 
1212         /* init node's zones as empty zones, we don't have any present pages.*/
1213         free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1214         pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
1215 
1216         /*
1217          * The node we allocated has no zone fallback lists. For avoiding
1218          * to access not-initialized zonelist, build here.
1219          */
1220         mutex_lock(&zonelists_mutex);
1221         build_all_zonelists(pgdat, NULL);
1222         mutex_unlock(&zonelists_mutex);
1223 
1224         /*
1225          * zone->managed_pages is set to an approximate value in
1226          * free_area_init_core(), which will cause
1227          * /sys/device/system/node/nodeX/meminfo has wrong data.
1228          * So reset it to 0 before any memory is onlined.
1229          */
1230         reset_node_managed_pages(pgdat);
1231 
1232         /*
1233          * When memory is hot-added, all the memory is in offline state. So
1234          * clear all zones' present_pages because they will be updated in
1235          * online_pages() and offline_pages().
1236          */
1237         reset_node_present_pages(pgdat);
1238 
1239         return pgdat;
1240 }
1241 
1242 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1243 {
1244         arch_refresh_nodedata(nid, NULL);
1245         free_percpu(pgdat->per_cpu_nodestats);
1246         arch_free_nodedata(pgdat);
1247         return;
1248 }
1249 
1250 
1251 /**
1252  * try_online_node - online a node if offlined
1253  *
1254  * called by cpu_up() to online a node without onlined memory.
1255  */
1256 int try_online_node(int nid)
1257 {
1258         pg_data_t       *pgdat;
1259         int     ret;
1260 
1261         if (node_online(nid))
1262                 return 0;
1263 
1264         mem_hotplug_begin();
1265         pgdat = hotadd_new_pgdat(nid, 0);
1266         if (!pgdat) {
1267                 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1268                 ret = -ENOMEM;
1269                 goto out;
1270         }
1271         node_set_online(nid);
1272         ret = register_one_node(nid);
1273         BUG_ON(ret);
1274 
1275         if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
1276                 mutex_lock(&zonelists_mutex);
1277                 build_all_zonelists(NULL, NULL);
1278                 mutex_unlock(&zonelists_mutex);
1279         }
1280 
1281 out:
1282         mem_hotplug_done();
1283         return ret;
1284 }
1285 
1286 static int check_hotplug_memory_range(u64 start, u64 size)
1287 {
1288         u64 start_pfn = PFN_DOWN(start);
1289         u64 nr_pages = size >> PAGE_SHIFT;
1290 
1291         /* Memory range must be aligned with section */
1292         if ((start_pfn & ~PAGE_SECTION_MASK) ||
1293             (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1294                 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1295                                 (unsigned long long)start,
1296                                 (unsigned long long)size);
1297                 return -EINVAL;
1298         }
1299 
1300         return 0;
1301 }
1302 
1303 /*
1304  * If movable zone has already been setup, newly added memory should be check.
1305  * If its address is higher than movable zone, it should be added as movable.
1306  * Without this check, movable zone may overlap with other zone.
1307  */
1308 static int should_add_memory_movable(int nid, u64 start, u64 size)
1309 {
1310         unsigned long start_pfn = start >> PAGE_SHIFT;
1311         pg_data_t *pgdat = NODE_DATA(nid);
1312         struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE;
1313 
1314         if (zone_is_empty(movable_zone))
1315                 return 0;
1316 
1317         if (movable_zone->zone_start_pfn <= start_pfn)
1318                 return 1;
1319 
1320         return 0;
1321 }
1322 
1323 int zone_for_memory(int nid, u64 start, u64 size, int zone_default,
1324                 bool for_device)
1325 {
1326 #ifdef CONFIG_ZONE_DEVICE
1327         if (for_device)
1328                 return ZONE_DEVICE;
1329 #endif
1330         if (should_add_memory_movable(nid, start, size))
1331                 return ZONE_MOVABLE;
1332 
1333         return zone_default;
1334 }
1335 
1336 static int online_memory_block(struct memory_block *mem, void *arg)
1337 {
1338         return memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
1339 }
1340 
1341 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1342 int __ref add_memory_resource(int nid, struct resource *res, bool online)
1343 {
1344         u64 start, size;
1345         pg_data_t *pgdat = NULL;
1346         bool new_pgdat;
1347         bool new_node;
1348         int ret;
1349 
1350         start = res->start;
1351         size = resource_size(res);
1352 
1353         ret = check_hotplug_memory_range(start, size);
1354         if (ret)
1355                 return ret;
1356 
1357         {       /* Stupid hack to suppress address-never-null warning */
1358                 void *p = NODE_DATA(nid);
1359                 new_pgdat = !p;
1360         }
1361 
1362         mem_hotplug_begin();
1363 
1364         /*
1365          * Add new range to memblock so that when hotadd_new_pgdat() is called
1366          * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1367          * this new range and calculate total pages correctly.  The range will
1368          * be removed at hot-remove time.
1369          */
1370         memblock_add_node(start, size, nid);
1371 
1372         new_node = !node_online(nid);
1373         if (new_node) {
1374                 pgdat = hotadd_new_pgdat(nid, start);
1375                 ret = -ENOMEM;
1376                 if (!pgdat)
1377                         goto error;
1378         }
1379 
1380         /* call arch's memory hotadd */
1381         ret = arch_add_memory(nid, start, size, false);
1382 
1383         if (ret < 0)
1384                 goto error;
1385 
1386         /* we online node here. we can't roll back from here. */
1387         node_set_online(nid);
1388 
1389         if (new_node) {
1390                 ret = register_one_node(nid);
1391                 /*
1392                  * If sysfs file of new node can't create, cpu on the node
1393                  * can't be hot-added. There is no rollback way now.
1394                  * So, check by BUG_ON() to catch it reluctantly..
1395                  */
1396                 BUG_ON(ret);
1397         }
1398 
1399         /* create new memmap entry */
1400         firmware_map_add_hotplug(start, start + size, "System RAM");
1401 
1402         /* online pages if requested */
1403         if (online)
1404                 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1405                                   NULL, online_memory_block);
1406 
1407         goto out;
1408 
1409 error:
1410         /* rollback pgdat allocation and others */
1411         if (new_pgdat)
1412                 rollback_node_hotadd(nid, pgdat);
1413         memblock_remove(start, size);
1414 
1415 out:
1416         mem_hotplug_done();
1417         return ret;
1418 }
1419 EXPORT_SYMBOL_GPL(add_memory_resource);
1420 
1421 int __ref add_memory(int nid, u64 start, u64 size)
1422 {
1423         struct resource *res;
1424         int ret;
1425 
1426         res = register_memory_resource(start, size);
1427         if (IS_ERR(res))
1428                 return PTR_ERR(res);
1429 
1430         ret = add_memory_resource(nid, res, memhp_auto_online);
1431         if (ret < 0)
1432                 release_memory_resource(res);
1433         return ret;
1434 }
1435 EXPORT_SYMBOL_GPL(add_memory);
1436 
1437 #ifdef CONFIG_MEMORY_HOTREMOVE
1438 /*
1439  * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1440  * set and the size of the free page is given by page_order(). Using this,
1441  * the function determines if the pageblock contains only free pages.
1442  * Due to buddy contraints, a free page at least the size of a pageblock will
1443  * be located at the start of the pageblock
1444  */
1445 static inline int pageblock_free(struct page *page)
1446 {
1447         return PageBuddy(page) && page_order(page) >= pageblock_order;
1448 }
1449 
1450 /* Return the start of the next active pageblock after a given page */
1451 static struct page *next_active_pageblock(struct page *page)
1452 {
1453         /* Ensure the starting page is pageblock-aligned */
1454         BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1455 
1456         /* If the entire pageblock is free, move to the end of free page */
1457         if (pageblock_free(page)) {
1458                 int order;
1459                 /* be careful. we don't have locks, page_order can be changed.*/
1460                 order = page_order(page);
1461                 if ((order < MAX_ORDER) && (order >= pageblock_order))
1462                         return page + (1 << order);
1463         }
1464 
1465         return page + pageblock_nr_pages;
1466 }
1467 
1468 /* Checks if this range of memory is likely to be hot-removable. */
1469 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1470 {
1471         struct page *page = pfn_to_page(start_pfn);
1472         struct page *end_page = page + nr_pages;
1473 
1474         /* Check the starting page of each pageblock within the range */
1475         for (; page < end_page; page = next_active_pageblock(page)) {
1476                 if (!is_pageblock_removable_nolock(page))
1477                         return false;
1478                 cond_resched();
1479         }
1480 
1481         /* All pageblocks in the memory block are likely to be hot-removable */
1482         return true;
1483 }
1484 
1485 /*
1486  * Confirm all pages in a range [start, end) belong to the same zone.
1487  * When true, return its valid [start, end).
1488  */
1489 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1490                          unsigned long *valid_start, unsigned long *valid_end)
1491 {
1492         unsigned long pfn, sec_end_pfn;
1493         unsigned long start, end;
1494         struct zone *zone = NULL;
1495         struct page *page;
1496         int i;
1497         for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1498              pfn < end_pfn;
1499              pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1500                 /* Make sure the memory section is present first */
1501                 if (!present_section_nr(pfn_to_section_nr(pfn)))
1502                         continue;
1503                 for (; pfn < sec_end_pfn && pfn < end_pfn;
1504                      pfn += MAX_ORDER_NR_PAGES) {
1505                         i = 0;
1506                         /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1507                         while ((i < MAX_ORDER_NR_PAGES) &&
1508                                 !pfn_valid_within(pfn + i))
1509                                 i++;
1510                         if (i == MAX_ORDER_NR_PAGES)
1511                                 continue;
1512                         page = pfn_to_page(pfn + i);
1513                         if (zone && page_zone(page) != zone)
1514                                 return 0;
1515                         if (!zone)
1516                                 start = pfn + i;
1517                         zone = page_zone(page);
1518                         end = pfn + MAX_ORDER_NR_PAGES;
1519                 }
1520         }
1521 
1522         if (zone) {
1523                 *valid_start = start;
1524                 *valid_end = end;
1525                 return 1;
1526         } else {
1527                 return 0;
1528         }
1529 }
1530 
1531 /*
1532  * Scan pfn range [start,end) to find movable/migratable pages (LRU pages
1533  * and hugepages). We scan pfn because it's much easier than scanning over
1534  * linked list. This function returns the pfn of the first found movable
1535  * page if it's found, otherwise 0.
1536  */
1537 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1538 {
1539         unsigned long pfn;
1540         struct page *page;
1541         for (pfn = start; pfn < end; pfn++) {
1542                 if (pfn_valid(pfn)) {
1543                         page = pfn_to_page(pfn);
1544                         if (PageLRU(page))
1545                                 return pfn;
1546                         if (PageHuge(page)) {
1547                                 if (page_huge_active(page))
1548                                         return pfn;
1549                                 else
1550                                         pfn = round_up(pfn + 1,
1551                                                 1 << compound_order(page)) - 1;
1552                         }
1553                 }
1554         }
1555         return 0;
1556 }
1557 
1558 static struct page *new_node_page(struct page *page, unsigned long private,
1559                 int **result)
1560 {
1561         gfp_t gfp_mask = GFP_USER | __GFP_MOVABLE;
1562         int nid = page_to_nid(page);
1563         nodemask_t nmask = node_states[N_MEMORY];
1564         struct page *new_page = NULL;
1565 
1566         /*
1567          * TODO: allocate a destination hugepage from a nearest neighbor node,
1568          * accordance with memory policy of the user process if possible. For
1569          * now as a simple work-around, we use the next node for destination.
1570          */
1571         if (PageHuge(page))
1572                 return alloc_huge_page_node(page_hstate(compound_head(page)),
1573                                         next_node_in(nid, nmask));
1574 
1575         node_clear(nid, nmask);
1576 
1577         if (PageHighMem(page)
1578             || (zone_idx(page_zone(page)) == ZONE_MOVABLE))
1579                 gfp_mask |= __GFP_HIGHMEM;
1580 
1581         if (!nodes_empty(nmask))
1582                 new_page = __alloc_pages_nodemask(gfp_mask, 0,
1583                                         node_zonelist(nid, gfp_mask), &nmask);
1584         if (!new_page)
1585                 new_page = __alloc_pages(gfp_mask, 0,
1586                                         node_zonelist(nid, gfp_mask));
1587 
1588         return new_page;
1589 }
1590 
1591 #define NR_OFFLINE_AT_ONCE_PAGES        (256)
1592 static int
1593 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1594 {
1595         unsigned long pfn;
1596         struct page *page;
1597         int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1598         int not_managed = 0;
1599         int ret = 0;
1600         LIST_HEAD(source);
1601 
1602         for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1603                 if (!pfn_valid(pfn))
1604                         continue;
1605                 page = pfn_to_page(pfn);
1606 
1607                 if (PageHuge(page)) {
1608                         struct page *head = compound_head(page);
1609                         pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1610                         if (compound_order(head) > PFN_SECTION_SHIFT) {
1611                                 ret = -EBUSY;
1612                                 break;
1613                         }
1614                         if (isolate_huge_page(page, &source))
1615                                 move_pages -= 1 << compound_order(head);
1616                         continue;
1617                 }
1618 
1619                 if (!get_page_unless_zero(page))
1620                         continue;
1621                 /*
1622                  * We can skip free pages. And we can only deal with pages on
1623                  * LRU.
1624                  */
1625                 ret = isolate_lru_page(page);
1626                 if (!ret) { /* Success */
1627                         put_page(page);
1628                         list_add_tail(&page->lru, &source);
1629                         move_pages--;
1630                         inc_node_page_state(page, NR_ISOLATED_ANON +
1631                                             page_is_file_cache(page));
1632 
1633                 } else {
1634 #ifdef CONFIG_DEBUG_VM
1635                         pr_alert("removing pfn %lx from LRU failed\n", pfn);
1636                         dump_page(page, "failed to remove from LRU");
1637 #endif
1638                         put_page(page);
1639                         /* Because we don't have big zone->lock. we should
1640                            check this again here. */
1641                         if (page_count(page)) {
1642                                 not_managed++;
1643                                 ret = -EBUSY;
1644                                 break;
1645                         }
1646                 }
1647         }
1648         if (!list_empty(&source)) {
1649                 if (not_managed) {
1650                         putback_movable_pages(&source);
1651                         goto out;
1652                 }
1653 
1654                 /* Allocate a new page from the nearest neighbor node */
1655                 ret = migrate_pages(&source, new_node_page, NULL, 0,
1656                                         MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1657                 if (ret)
1658                         putback_movable_pages(&source);
1659         }
1660 out:
1661         return ret;
1662 }
1663 
1664 /*
1665  * remove from free_area[] and mark all as Reserved.
1666  */
1667 static int
1668 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1669                         void *data)
1670 {
1671         __offline_isolated_pages(start, start + nr_pages);
1672         return 0;
1673 }
1674 
1675 static void
1676 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1677 {
1678         walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1679                                 offline_isolated_pages_cb);
1680 }
1681 
1682 /*
1683  * Check all pages in range, recoreded as memory resource, are isolated.
1684  */
1685 static int
1686 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1687                         void *data)
1688 {
1689         int ret;
1690         long offlined = *(long *)data;
1691         ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1692         offlined = nr_pages;
1693         if (!ret)
1694                 *(long *)data += offlined;
1695         return ret;
1696 }
1697 
1698 static long
1699 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1700 {
1701         long offlined = 0;
1702         int ret;
1703 
1704         ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1705                         check_pages_isolated_cb);
1706         if (ret < 0)
1707                 offlined = (long)ret;
1708         return offlined;
1709 }
1710 
1711 #ifdef CONFIG_MOVABLE_NODE
1712 /*
1713  * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1714  * normal memory.
1715  */
1716 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1717 {
1718         return true;
1719 }
1720 #else /* CONFIG_MOVABLE_NODE */
1721 /* ensure the node has NORMAL memory if it is still online */
1722 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1723 {
1724         struct pglist_data *pgdat = zone->zone_pgdat;
1725         unsigned long present_pages = 0;
1726         enum zone_type zt;
1727 
1728         for (zt = 0; zt <= ZONE_NORMAL; zt++)
1729                 present_pages += pgdat->node_zones[zt].present_pages;
1730 
1731         if (present_pages > nr_pages)
1732                 return true;
1733 
1734         present_pages = 0;
1735         for (; zt <= ZONE_MOVABLE; zt++)
1736                 present_pages += pgdat->node_zones[zt].present_pages;
1737 
1738         /*
1739          * we can't offline the last normal memory until all
1740          * higher memory is offlined.
1741          */
1742         return present_pages == 0;
1743 }
1744 #endif /* CONFIG_MOVABLE_NODE */
1745 
1746 static int __init cmdline_parse_movable_node(char *p)
1747 {
1748 #ifdef CONFIG_MOVABLE_NODE
1749         movable_node_enabled = true;
1750 #else
1751         pr_warn("movable_node option not supported\n");
1752 #endif
1753         return 0;
1754 }
1755 early_param("movable_node", cmdline_parse_movable_node);
1756 
1757 /* check which state of node_states will be changed when offline memory */
1758 static void node_states_check_changes_offline(unsigned long nr_pages,
1759                 struct zone *zone, struct memory_notify *arg)
1760 {
1761         struct pglist_data *pgdat = zone->zone_pgdat;
1762         unsigned long present_pages = 0;
1763         enum zone_type zt, zone_last = ZONE_NORMAL;
1764 
1765         /*
1766          * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1767          * contains nodes which have zones of 0...ZONE_NORMAL,
1768          * set zone_last to ZONE_NORMAL.
1769          *
1770          * If we don't have HIGHMEM nor movable node,
1771          * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1772          * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1773          */
1774         if (N_MEMORY == N_NORMAL_MEMORY)
1775                 zone_last = ZONE_MOVABLE;
1776 
1777         /*
1778          * check whether node_states[N_NORMAL_MEMORY] will be changed.
1779          * If the memory to be offline is in a zone of 0...zone_last,
1780          * and it is the last present memory, 0...zone_last will
1781          * become empty after offline , thus we can determind we will
1782          * need to clear the node from node_states[N_NORMAL_MEMORY].
1783          */
1784         for (zt = 0; zt <= zone_last; zt++)
1785                 present_pages += pgdat->node_zones[zt].present_pages;
1786         if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1787                 arg->status_change_nid_normal = zone_to_nid(zone);
1788         else
1789                 arg->status_change_nid_normal = -1;
1790 
1791 #ifdef CONFIG_HIGHMEM
1792         /*
1793          * If we have movable node, node_states[N_HIGH_MEMORY]
1794          * contains nodes which have zones of 0...ZONE_HIGHMEM,
1795          * set zone_last to ZONE_HIGHMEM.
1796          *
1797          * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1798          * contains nodes which have zones of 0...ZONE_MOVABLE,
1799          * set zone_last to ZONE_MOVABLE.
1800          */
1801         zone_last = ZONE_HIGHMEM;
1802         if (N_MEMORY == N_HIGH_MEMORY)
1803                 zone_last = ZONE_MOVABLE;
1804 
1805         for (; zt <= zone_last; zt++)
1806                 present_pages += pgdat->node_zones[zt].present_pages;
1807         if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1808                 arg->status_change_nid_high = zone_to_nid(zone);
1809         else
1810                 arg->status_change_nid_high = -1;
1811 #else
1812         arg->status_change_nid_high = arg->status_change_nid_normal;
1813 #endif
1814 
1815         /*
1816          * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1817          */
1818         zone_last = ZONE_MOVABLE;
1819 
1820         /*
1821          * check whether node_states[N_HIGH_MEMORY] will be changed
1822          * If we try to offline the last present @nr_pages from the node,
1823          * we can determind we will need to clear the node from
1824          * node_states[N_HIGH_MEMORY].
1825          */
1826         for (; zt <= zone_last; zt++)
1827                 present_pages += pgdat->node_zones[zt].present_pages;
1828         if (nr_pages >= present_pages)
1829                 arg->status_change_nid = zone_to_nid(zone);
1830         else
1831                 arg->status_change_nid = -1;
1832 }
1833 
1834 static void node_states_clear_node(int node, struct memory_notify *arg)
1835 {
1836         if (arg->status_change_nid_normal >= 0)
1837                 node_clear_state(node, N_NORMAL_MEMORY);
1838 
1839         if ((N_MEMORY != N_NORMAL_MEMORY) &&
1840             (arg->status_change_nid_high >= 0))
1841                 node_clear_state(node, N_HIGH_MEMORY);
1842 
1843         if ((N_MEMORY != N_HIGH_MEMORY) &&
1844             (arg->status_change_nid >= 0))
1845                 node_clear_state(node, N_MEMORY);
1846 }
1847 
1848 static int __ref __offline_pages(unsigned long start_pfn,
1849                   unsigned long end_pfn, unsigned long timeout)
1850 {
1851         unsigned long pfn, nr_pages, expire;
1852         long offlined_pages;
1853         int ret, drain, retry_max, node;
1854         unsigned long flags;
1855         unsigned long valid_start, valid_end;
1856         struct zone *zone;
1857         struct memory_notify arg;
1858 
1859         /* at least, alignment against pageblock is necessary */
1860         if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1861                 return -EINVAL;
1862         if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1863                 return -EINVAL;
1864         /* This makes hotplug much easier...and readable.
1865            we assume this for now. .*/
1866         if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, &valid_end))
1867                 return -EINVAL;
1868 
1869         zone = page_zone(pfn_to_page(valid_start));
1870         node = zone_to_nid(zone);
1871         nr_pages = end_pfn - start_pfn;
1872 
1873         if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1874                 return -EINVAL;
1875 
1876         /* set above range as isolated */
1877         ret = start_isolate_page_range(start_pfn, end_pfn,
1878                                        MIGRATE_MOVABLE, true);
1879         if (ret)
1880                 return ret;
1881 
1882         arg.start_pfn = start_pfn;
1883         arg.nr_pages = nr_pages;
1884         node_states_check_changes_offline(nr_pages, zone, &arg);
1885 
1886         ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1887         ret = notifier_to_errno(ret);
1888         if (ret)
1889                 goto failed_removal;
1890 
1891         pfn = start_pfn;
1892         expire = jiffies + timeout;
1893         drain = 0;
1894         retry_max = 5;
1895 repeat:
1896         /* start memory hot removal */
1897         ret = -EAGAIN;
1898         if (time_after(jiffies, expire))
1899                 goto failed_removal;
1900         ret = -EINTR;
1901         if (signal_pending(current))
1902                 goto failed_removal;
1903         ret = 0;
1904         if (drain) {
1905                 lru_add_drain_all();
1906                 cond_resched();
1907                 drain_all_pages(zone);
1908         }
1909 
1910         pfn = scan_movable_pages(start_pfn, end_pfn);
1911         if (pfn) { /* We have movable pages */
1912                 ret = do_migrate_range(pfn, end_pfn);
1913                 if (!ret) {
1914                         drain = 1;
1915                         goto repeat;
1916                 } else {
1917                         if (ret < 0)
1918                                 if (--retry_max == 0)
1919                                         goto failed_removal;
1920                         yield();
1921                         drain = 1;
1922                         goto repeat;
1923                 }
1924         }
1925         /* drain all zone's lru pagevec, this is asynchronous... */
1926         lru_add_drain_all();
1927         yield();
1928         /* drain pcp pages, this is synchronous. */
1929         drain_all_pages(zone);
1930         /*
1931          * dissolve free hugepages in the memory block before doing offlining
1932          * actually in order to make hugetlbfs's object counting consistent.
1933          */
1934         ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1935         if (ret)
1936                 goto failed_removal;
1937         /* check again */
1938         offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1939         if (offlined_pages < 0) {
1940                 ret = -EBUSY;
1941                 goto failed_removal;
1942         }
1943         pr_info("Offlined Pages %ld\n", offlined_pages);
1944         /* Ok, all of our target is isolated.
1945            We cannot do rollback at this point. */
1946         offline_isolated_pages(start_pfn, end_pfn);
1947         /* reset pagetype flags and makes migrate type to be MOVABLE */
1948         undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1949         /* removal success */
1950         adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1951         zone->present_pages -= offlined_pages;
1952 
1953         pgdat_resize_lock(zone->zone_pgdat, &flags);
1954         zone->zone_pgdat->node_present_pages -= offlined_pages;
1955         pgdat_resize_unlock(zone->zone_pgdat, &flags);
1956 
1957         init_per_zone_wmark_min();
1958 
1959         if (!populated_zone(zone)) {
1960                 zone_pcp_reset(zone);
1961                 mutex_lock(&zonelists_mutex);
1962                 build_all_zonelists(NULL, NULL);
1963                 mutex_unlock(&zonelists_mutex);
1964         } else
1965                 zone_pcp_update(zone);
1966 
1967         node_states_clear_node(node, &arg);
1968         if (arg.status_change_nid >= 0) {
1969                 kswapd_stop(node);
1970                 kcompactd_stop(node);
1971         }
1972 
1973         vm_total_pages = nr_free_pagecache_pages();
1974         writeback_set_ratelimit();
1975 
1976         memory_notify(MEM_OFFLINE, &arg);
1977         return 0;
1978 
1979 failed_removal:
1980         pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n",
1981                  (unsigned long long) start_pfn << PAGE_SHIFT,
1982                  ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1983         memory_notify(MEM_CANCEL_OFFLINE, &arg);
1984         /* pushback to free area */
1985         undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1986         return ret;
1987 }
1988 
1989 /* Must be protected by mem_hotplug_begin() */
1990 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1991 {
1992         return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1993 }
1994 #endif /* CONFIG_MEMORY_HOTREMOVE */
1995 
1996 /**
1997  * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1998  * @start_pfn: start pfn of the memory range
1999  * @end_pfn: end pfn of the memory range
2000  * @arg: argument passed to func
2001  * @func: callback for each memory section walked
2002  *
2003  * This function walks through all present mem sections in range
2004  * [start_pfn, end_pfn) and call func on each mem section.
2005  *
2006  * Returns the return value of func.
2007  */
2008 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
2009                 void *arg, int (*func)(struct memory_block *, void *))
2010 {
2011         struct memory_block *mem = NULL;
2012         struct mem_section *section;
2013         unsigned long pfn, section_nr;
2014         int ret;
2015 
2016         for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2017                 section_nr = pfn_to_section_nr(pfn);
2018                 if (!present_section_nr(section_nr))
2019                         continue;
2020 
2021                 section = __nr_to_section(section_nr);
2022                 /* same memblock? */
2023                 if (mem)
2024                         if ((section_nr >= mem->start_section_nr) &&
2025                             (section_nr <= mem->end_section_nr))
2026                                 continue;
2027 
2028                 mem = find_memory_block_hinted(section, mem);
2029                 if (!mem)
2030                         continue;
2031 
2032                 ret = func(mem, arg);
2033                 if (ret) {
2034                         kobject_put(&mem->dev.kobj);
2035                         return ret;
2036                 }
2037         }
2038 
2039         if (mem)
2040                 kobject_put(&mem->dev.kobj);
2041 
2042         return 0;
2043 }
2044 
2045 #ifdef CONFIG_MEMORY_HOTREMOVE
2046 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
2047 {
2048         int ret = !is_memblock_offlined(mem);
2049 
2050         if (unlikely(ret)) {
2051                 phys_addr_t beginpa, endpa;
2052 
2053                 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
2054                 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
2055                 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
2056                         &beginpa, &endpa);
2057         }
2058 
2059         return ret;
2060 }
2061 
2062 static int check_cpu_on_node(pg_data_t *pgdat)
2063 {
2064         int cpu;
2065 
2066         for_each_present_cpu(cpu) {
2067                 if (cpu_to_node(cpu) == pgdat->node_id)
2068                         /*
2069                          * the cpu on this node isn't removed, and we can't
2070                          * offline this node.
2071                          */
2072                         return -EBUSY;
2073         }
2074 
2075         return 0;
2076 }
2077 
2078 static void unmap_cpu_on_node(pg_data_t *pgdat)
2079 {
2080 #ifdef CONFIG_ACPI_NUMA
2081         int cpu;
2082 
2083         for_each_possible_cpu(cpu)
2084                 if (cpu_to_node(cpu) == pgdat->node_id)
2085                         numa_clear_node(cpu);
2086 #endif
2087 }
2088 
2089 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
2090 {
2091         int ret;
2092 
2093         ret = check_cpu_on_node(pgdat);
2094         if (ret)
2095                 return ret;
2096 
2097         /*
2098          * the node will be offlined when we come here, so we can clear
2099          * the cpu_to_node() now.
2100          */
2101 
2102         unmap_cpu_on_node(pgdat);
2103         return 0;
2104 }
2105 
2106 /**
2107  * try_offline_node
2108  *
2109  * Offline a node if all memory sections and cpus of the node are removed.
2110  *
2111  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2112  * and online/offline operations before this call.
2113  */
2114 void try_offline_node(int nid)
2115 {
2116         pg_data_t *pgdat = NODE_DATA(nid);
2117         unsigned long start_pfn = pgdat->node_start_pfn;
2118         unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
2119         unsigned long pfn;
2120 
2121         for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2122                 unsigned long section_nr = pfn_to_section_nr(pfn);
2123 
2124                 if (!present_section_nr(section_nr))
2125                         continue;
2126 
2127                 if (pfn_to_nid(pfn) != nid)
2128                         continue;
2129 
2130                 /*
2131                  * some memory sections of this node are not removed, and we
2132                  * can't offline node now.
2133                  */
2134                 return;
2135         }
2136 
2137         if (check_and_unmap_cpu_on_node(pgdat))
2138                 return;
2139 
2140         /*
2141          * all memory/cpu of this node are removed, we can offline this
2142          * node now.
2143          */
2144         node_set_offline(nid);
2145         unregister_one_node(nid);
2146 }
2147 EXPORT_SYMBOL(try_offline_node);
2148 
2149 /**
2150  * remove_memory
2151  *
2152  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2153  * and online/offline operations before this call, as required by
2154  * try_offline_node().
2155  */
2156 void __ref remove_memory(int nid, u64 start, u64 size)
2157 {
2158         int ret;
2159 
2160         BUG_ON(check_hotplug_memory_range(start, size));
2161 
2162         mem_hotplug_begin();
2163 
2164         /*
2165          * All memory blocks must be offlined before removing memory.  Check
2166          * whether all memory blocks in question are offline and trigger a BUG()
2167          * if this is not the case.
2168          */
2169         ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
2170                                 check_memblock_offlined_cb);
2171         if (ret)
2172                 BUG();
2173 
2174         /* remove memmap entry */
2175         firmware_map_remove(start, start + size, "System RAM");
2176         memblock_free(start, size);
2177         memblock_remove(start, size);
2178 
2179         arch_remove_memory(start, size);
2180 
2181         try_offline_node(nid);
2182 
2183         mem_hotplug_done();
2184 }
2185 EXPORT_SYMBOL_GPL(remove_memory);
2186 #endif /* CONFIG_MEMORY_HOTREMOVE */
2187 

This page was automatically generated by LXR 0.3.1 (source).  •  Linux is a registered trademark of Linus Torvalds  •  Contact us