~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/mm/memory_hotplug.c

Version: ~ [ linux-4.15-rc8 ] ~ [ linux-4.14.13 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.76 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.111 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.48 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.91 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.53 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.98 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.27.62 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

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

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp