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

TOMOYO Linux Cross Reference
Linux/mm/memory_hotplug.c

Version: ~ [ linux-5.15-rc7 ] ~ [ linux-5.14.14 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.75 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.155 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.213 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.252 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.287 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.289 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ 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 // SPDX-License-Identifier: GPL-2.0-only
  2 /*
  3  *  linux/mm/memory_hotplug.c
  4  *
  5  *  Copyright (C)
  6  */
  7 
  8 #include <linux/stddef.h>
  9 #include <linux/mm.h>
 10 #include <linux/sched/signal.h>
 11 #include <linux/swap.h>
 12 #include <linux/interrupt.h>
 13 #include <linux/pagemap.h>
 14 #include <linux/compiler.h>
 15 #include <linux/export.h>
 16 #include <linux/pagevec.h>
 17 #include <linux/writeback.h>
 18 #include <linux/slab.h>
 19 #include <linux/sysctl.h>
 20 #include <linux/cpu.h>
 21 #include <linux/memory.h>
 22 #include <linux/memremap.h>
 23 #include <linux/memory_hotplug.h>
 24 #include <linux/highmem.h>
 25 #include <linux/vmalloc.h>
 26 #include <linux/ioport.h>
 27 #include <linux/delay.h>
 28 #include <linux/migrate.h>
 29 #include <linux/page-isolation.h>
 30 #include <linux/pfn.h>
 31 #include <linux/suspend.h>
 32 #include <linux/mm_inline.h>
 33 #include <linux/firmware-map.h>
 34 #include <linux/stop_machine.h>
 35 #include <linux/hugetlb.h>
 36 #include <linux/memblock.h>
 37 #include <linux/compaction.h>
 38 #include <linux/rmap.h>
 39 
 40 #include <asm/tlbflush.h>
 41 
 42 #include "internal.h"
 43 #include "shuffle.h"
 44 
 45 /*
 46  * online_page_callback contains pointer to current page onlining function.
 47  * Initially it is generic_online_page(). If it is required it could be
 48  * changed by calling set_online_page_callback() for callback registration
 49  * and restore_online_page_callback() for generic callback restore.
 50  */
 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 u64 max_mem_size = U64_MAX;
100 
101 /* add this memory to iomem resource */
102 static struct resource *register_memory_resource(u64 start, u64 size)
103 {
104         struct resource *res;
105         unsigned long flags =  IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
106         char *resource_name = "System RAM";
107 
108         if (start + size > max_mem_size)
109                 return ERR_PTR(-E2BIG);
110 
111         /*
112          * Request ownership of the new memory range.  This might be
113          * a child of an existing resource that was present but
114          * not marked as busy.
115          */
116         res = __request_region(&iomem_resource, start, size,
117                                resource_name, flags);
118 
119         if (!res) {
120                 pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n",
121                                 start, start + size);
122                 return ERR_PTR(-EEXIST);
123         }
124         return res;
125 }
126 
127 static void release_memory_resource(struct resource *res)
128 {
129         if (!res)
130                 return;
131         release_resource(res);
132         kfree(res);
133 }
134 
135 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
136 void get_page_bootmem(unsigned long info,  struct page *page,
137                       unsigned long type)
138 {
139         page->freelist = (void *)type;
140         SetPagePrivate(page);
141         set_page_private(page, info);
142         page_ref_inc(page);
143 }
144 
145 void put_page_bootmem(struct page *page)
146 {
147         unsigned long type;
148 
149         type = (unsigned long) page->freelist;
150         BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
151                type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
152 
153         if (page_ref_dec_return(page) == 1) {
154                 page->freelist = NULL;
155                 ClearPagePrivate(page);
156                 set_page_private(page, 0);
157                 INIT_LIST_HEAD(&page->lru);
158                 free_reserved_page(page);
159         }
160 }
161 
162 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
163 #ifndef CONFIG_SPARSEMEM_VMEMMAP
164 static void register_page_bootmem_info_section(unsigned long start_pfn)
165 {
166         unsigned long mapsize, section_nr, i;
167         struct mem_section *ms;
168         struct page *page, *memmap;
169         struct mem_section_usage *usage;
170 
171         section_nr = pfn_to_section_nr(start_pfn);
172         ms = __nr_to_section(section_nr);
173 
174         /* Get section's memmap address */
175         memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
176 
177         /*
178          * Get page for the memmap's phys address
179          * XXX: need more consideration for sparse_vmemmap...
180          */
181         page = virt_to_page(memmap);
182         mapsize = sizeof(struct page) * PAGES_PER_SECTION;
183         mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
184 
185         /* remember memmap's page */
186         for (i = 0; i < mapsize; i++, page++)
187                 get_page_bootmem(section_nr, page, SECTION_INFO);
188 
189         usage = ms->usage;
190         page = virt_to_page(usage);
191 
192         mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT;
193 
194         for (i = 0; i < mapsize; i++, page++)
195                 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
196 
197 }
198 #else /* CONFIG_SPARSEMEM_VMEMMAP */
199 static void register_page_bootmem_info_section(unsigned long start_pfn)
200 {
201         unsigned long mapsize, section_nr, i;
202         struct mem_section *ms;
203         struct page *page, *memmap;
204         struct mem_section_usage *usage;
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         usage = ms->usage;
214         page = virt_to_page(usage);
215 
216         mapsize = PAGE_ALIGN(mem_section_usage_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 check_pfn_span(unsigned long pfn, unsigned long nr_pages,
253                 const char *reason)
254 {
255         /*
256          * Disallow all operations smaller than a sub-section and only
257          * allow operations smaller than a section for
258          * SPARSEMEM_VMEMMAP. Note that check_hotplug_memory_range()
259          * enforces a larger memory_block_size_bytes() granularity for
260          * memory that will be marked online, so this check should only
261          * fire for direct arch_{add,remove}_memory() users outside of
262          * add_memory_resource().
263          */
264         unsigned long min_align;
265 
266         if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
267                 min_align = PAGES_PER_SUBSECTION;
268         else
269                 min_align = PAGES_PER_SECTION;
270         if (!IS_ALIGNED(pfn, min_align)
271                         || !IS_ALIGNED(nr_pages, min_align)) {
272                 WARN(1, "Misaligned __%s_pages start: %#lx end: #%lx\n",
273                                 reason, pfn, pfn + nr_pages - 1);
274                 return -EINVAL;
275         }
276         return 0;
277 }
278 
279 static int check_hotplug_memory_addressable(unsigned long pfn,
280                                             unsigned long nr_pages)
281 {
282         const u64 max_addr = PFN_PHYS(pfn + nr_pages) - 1;
283 
284         if (max_addr >> MAX_PHYSMEM_BITS) {
285                 const u64 max_allowed = (1ull << (MAX_PHYSMEM_BITS + 1)) - 1;
286                 WARN(1,
287                      "Hotplugged memory exceeds maximum addressable address, range=%#llx-%#llx, maximum=%#llx\n",
288                      (u64)PFN_PHYS(pfn), max_addr, max_allowed);
289                 return -E2BIG;
290         }
291 
292         return 0;
293 }
294 
295 /*
296  * Reasonably generic function for adding memory.  It is
297  * expected that archs that support memory hotplug will
298  * call this function after deciding the zone to which to
299  * add the new pages.
300  */
301 int __ref __add_pages(int nid, unsigned long pfn, unsigned long nr_pages,
302                 struct mhp_restrictions *restrictions)
303 {
304         int err;
305         unsigned long nr, start_sec, end_sec;
306         struct vmem_altmap *altmap = restrictions->altmap;
307 
308         err = check_hotplug_memory_addressable(pfn, nr_pages);
309         if (err)
310                 return err;
311 
312         if (altmap) {
313                 /*
314                  * Validate altmap is within bounds of the total request
315                  */
316                 if (altmap->base_pfn != pfn
317                                 || vmem_altmap_offset(altmap) > nr_pages) {
318                         pr_warn_once("memory add fail, invalid altmap\n");
319                         return -EINVAL;
320                 }
321                 altmap->alloc = 0;
322         }
323 
324         err = check_pfn_span(pfn, nr_pages, "add");
325         if (err)
326                 return err;
327 
328         start_sec = pfn_to_section_nr(pfn);
329         end_sec = pfn_to_section_nr(pfn + nr_pages - 1);
330         for (nr = start_sec; nr <= end_sec; nr++) {
331                 unsigned long pfns;
332 
333                 pfns = min(nr_pages, PAGES_PER_SECTION
334                                 - (pfn & ~PAGE_SECTION_MASK));
335                 err = sparse_add_section(nid, pfn, pfns, altmap);
336                 if (err)
337                         break;
338                 pfn += pfns;
339                 nr_pages -= pfns;
340                 cond_resched();
341         }
342         vmemmap_populate_print_last();
343         return err;
344 }
345 
346 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
347 static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
348                                      unsigned long start_pfn,
349                                      unsigned long end_pfn)
350 {
351         for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) {
352                 if (unlikely(!pfn_to_online_page(start_pfn)))
353                         continue;
354 
355                 if (unlikely(pfn_to_nid(start_pfn) != nid))
356                         continue;
357 
358                 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
359                         continue;
360 
361                 return start_pfn;
362         }
363 
364         return 0;
365 }
366 
367 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
368 static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
369                                     unsigned long start_pfn,
370                                     unsigned long end_pfn)
371 {
372         unsigned long pfn;
373 
374         /* pfn is the end pfn of a memory section. */
375         pfn = end_pfn - 1;
376         for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) {
377                 if (unlikely(!pfn_to_online_page(pfn)))
378                         continue;
379 
380                 if (unlikely(pfn_to_nid(pfn) != nid))
381                         continue;
382 
383                 if (zone && zone != page_zone(pfn_to_page(pfn)))
384                         continue;
385 
386                 return pfn;
387         }
388 
389         return 0;
390 }
391 
392 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
393                              unsigned long end_pfn)
394 {
395         unsigned long zone_start_pfn = zone->zone_start_pfn;
396         unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
397         unsigned long zone_end_pfn = z;
398         unsigned long pfn;
399         int nid = zone_to_nid(zone);
400 
401         zone_span_writelock(zone);
402         if (zone_start_pfn == start_pfn) {
403                 /*
404                  * If the section is smallest section in the zone, it need
405                  * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
406                  * In this case, we find second smallest valid mem_section
407                  * for shrinking zone.
408                  */
409                 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
410                                                 zone_end_pfn);
411                 if (pfn) {
412                         zone->zone_start_pfn = pfn;
413                         zone->spanned_pages = zone_end_pfn - pfn;
414                 }
415         } else if (zone_end_pfn == end_pfn) {
416                 /*
417                  * If the section is biggest section in the zone, it need
418                  * shrink zone->spanned_pages.
419                  * In this case, we find second biggest valid mem_section for
420                  * shrinking zone.
421                  */
422                 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
423                                                start_pfn);
424                 if (pfn)
425                         zone->spanned_pages = pfn - zone_start_pfn + 1;
426         }
427 
428         /*
429          * The section is not biggest or smallest mem_section in the zone, it
430          * only creates a hole in the zone. So in this case, we need not
431          * change the zone. But perhaps, the zone has only hole data. Thus
432          * it check the zone has only hole or not.
433          */
434         pfn = zone_start_pfn;
435         for (; pfn < zone_end_pfn; pfn += PAGES_PER_SUBSECTION) {
436                 if (unlikely(!pfn_to_online_page(pfn)))
437                         continue;
438 
439                 if (page_zone(pfn_to_page(pfn)) != zone)
440                         continue;
441 
442                 /* Skip range to be removed */
443                 if (pfn >= start_pfn && pfn < end_pfn)
444                         continue;
445 
446                 /* If we find valid section, we have nothing to do */
447                 zone_span_writeunlock(zone);
448                 return;
449         }
450 
451         /* The zone has no valid section */
452         zone->zone_start_pfn = 0;
453         zone->spanned_pages = 0;
454         zone_span_writeunlock(zone);
455 }
456 
457 static void update_pgdat_span(struct pglist_data *pgdat)
458 {
459         unsigned long node_start_pfn = 0, node_end_pfn = 0;
460         struct zone *zone;
461 
462         for (zone = pgdat->node_zones;
463              zone < pgdat->node_zones + MAX_NR_ZONES; zone++) {
464                 unsigned long zone_end_pfn = zone->zone_start_pfn +
465                                              zone->spanned_pages;
466 
467                 /* No need to lock the zones, they can't change. */
468                 if (!zone->spanned_pages)
469                         continue;
470                 if (!node_end_pfn) {
471                         node_start_pfn = zone->zone_start_pfn;
472                         node_end_pfn = zone_end_pfn;
473                         continue;
474                 }
475 
476                 if (zone_end_pfn > node_end_pfn)
477                         node_end_pfn = zone_end_pfn;
478                 if (zone->zone_start_pfn < node_start_pfn)
479                         node_start_pfn = zone->zone_start_pfn;
480         }
481 
482         pgdat->node_start_pfn = node_start_pfn;
483         pgdat->node_spanned_pages = node_end_pfn - node_start_pfn;
484 }
485 
486 void __ref remove_pfn_range_from_zone(struct zone *zone,
487                                       unsigned long start_pfn,
488                                       unsigned long nr_pages)
489 {
490         struct pglist_data *pgdat = zone->zone_pgdat;
491         unsigned long flags;
492 
493 #ifdef CONFIG_ZONE_DEVICE
494         /*
495          * Zone shrinking code cannot properly deal with ZONE_DEVICE. So
496          * we will not try to shrink the zones - which is okay as
497          * set_zone_contiguous() cannot deal with ZONE_DEVICE either way.
498          */
499         if (zone_idx(zone) == ZONE_DEVICE)
500                 return;
501 #endif
502 
503         clear_zone_contiguous(zone);
504 
505         pgdat_resize_lock(zone->zone_pgdat, &flags);
506         shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
507         update_pgdat_span(pgdat);
508         pgdat_resize_unlock(zone->zone_pgdat, &flags);
509 
510         set_zone_contiguous(zone);
511 }
512 
513 static void __remove_section(unsigned long pfn, unsigned long nr_pages,
514                              unsigned long map_offset,
515                              struct vmem_altmap *altmap)
516 {
517         struct mem_section *ms = __nr_to_section(pfn_to_section_nr(pfn));
518 
519         if (WARN_ON_ONCE(!valid_section(ms)))
520                 return;
521 
522         sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap);
523 }
524 
525 /**
526  * __remove_pages() - remove sections of pages
527  * @pfn: starting pageframe (must be aligned to start of a section)
528  * @nr_pages: number of pages to remove (must be multiple of section size)
529  * @altmap: alternative device page map or %NULL if default memmap is used
530  *
531  * Generic helper function to remove section mappings and sysfs entries
532  * for the section of the memory we are removing. Caller needs to make
533  * sure that pages are marked reserved and zones are adjust properly by
534  * calling offline_pages().
535  */
536 void __remove_pages(unsigned long pfn, unsigned long nr_pages,
537                     struct vmem_altmap *altmap)
538 {
539         unsigned long map_offset = 0;
540         unsigned long nr, start_sec, end_sec;
541 
542         map_offset = vmem_altmap_offset(altmap);
543 
544         if (check_pfn_span(pfn, nr_pages, "remove"))
545                 return;
546 
547         start_sec = pfn_to_section_nr(pfn);
548         end_sec = pfn_to_section_nr(pfn + nr_pages - 1);
549         for (nr = start_sec; nr <= end_sec; nr++) {
550                 unsigned long pfns;
551 
552                 cond_resched();
553                 pfns = min(nr_pages, PAGES_PER_SECTION
554                                 - (pfn & ~PAGE_SECTION_MASK));
555                 __remove_section(pfn, pfns, map_offset, altmap);
556                 pfn += pfns;
557                 nr_pages -= pfns;
558                 map_offset = 0;
559         }
560 }
561 
562 int set_online_page_callback(online_page_callback_t callback)
563 {
564         int rc = -EINVAL;
565 
566         get_online_mems();
567         mutex_lock(&online_page_callback_lock);
568 
569         if (online_page_callback == generic_online_page) {
570                 online_page_callback = callback;
571                 rc = 0;
572         }
573 
574         mutex_unlock(&online_page_callback_lock);
575         put_online_mems();
576 
577         return rc;
578 }
579 EXPORT_SYMBOL_GPL(set_online_page_callback);
580 
581 int restore_online_page_callback(online_page_callback_t callback)
582 {
583         int rc = -EINVAL;
584 
585         get_online_mems();
586         mutex_lock(&online_page_callback_lock);
587 
588         if (online_page_callback == callback) {
589                 online_page_callback = generic_online_page;
590                 rc = 0;
591         }
592 
593         mutex_unlock(&online_page_callback_lock);
594         put_online_mems();
595 
596         return rc;
597 }
598 EXPORT_SYMBOL_GPL(restore_online_page_callback);
599 
600 void generic_online_page(struct page *page, unsigned int order)
601 {
602         /*
603          * Freeing the page with debug_pagealloc enabled will try to unmap it,
604          * so we should map it first. This is better than introducing a special
605          * case in page freeing fast path.
606          */
607         if (debug_pagealloc_enabled_static())
608                 kernel_map_pages(page, 1 << order, 1);
609         __free_pages_core(page, order);
610         totalram_pages_add(1UL << order);
611 #ifdef CONFIG_HIGHMEM
612         if (PageHighMem(page))
613                 totalhigh_pages_add(1UL << order);
614 #endif
615 }
616 EXPORT_SYMBOL_GPL(generic_online_page);
617 
618 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
619                         void *arg)
620 {
621         const unsigned long end_pfn = start_pfn + nr_pages;
622         unsigned long pfn;
623         int order;
624 
625         /*
626          * Online the pages. The callback might decide to keep some pages
627          * PG_reserved (to add them to the buddy later), but we still account
628          * them as being online/belonging to this zone ("present").
629          */
630         for (pfn = start_pfn; pfn < end_pfn; pfn += 1ul << order) {
631                 order = min(MAX_ORDER - 1, get_order(PFN_PHYS(end_pfn - pfn)));
632                 /* __free_pages_core() wants pfns to be aligned to the order */
633                 if (WARN_ON_ONCE(!IS_ALIGNED(pfn, 1ul << order)))
634                         order = 0;
635                 (*online_page_callback)(pfn_to_page(pfn), order);
636         }
637 
638         /* mark all involved sections as online */
639         online_mem_sections(start_pfn, end_pfn);
640 
641         *(unsigned long *)arg += nr_pages;
642         return 0;
643 }
644 
645 /* check which state of node_states will be changed when online memory */
646 static void node_states_check_changes_online(unsigned long nr_pages,
647         struct zone *zone, struct memory_notify *arg)
648 {
649         int nid = zone_to_nid(zone);
650 
651         arg->status_change_nid = NUMA_NO_NODE;
652         arg->status_change_nid_normal = NUMA_NO_NODE;
653         arg->status_change_nid_high = NUMA_NO_NODE;
654 
655         if (!node_state(nid, N_MEMORY))
656                 arg->status_change_nid = nid;
657         if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
658                 arg->status_change_nid_normal = nid;
659 #ifdef CONFIG_HIGHMEM
660         if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY))
661                 arg->status_change_nid_high = nid;
662 #endif
663 }
664 
665 static void node_states_set_node(int node, struct memory_notify *arg)
666 {
667         if (arg->status_change_nid_normal >= 0)
668                 node_set_state(node, N_NORMAL_MEMORY);
669 
670         if (arg->status_change_nid_high >= 0)
671                 node_set_state(node, N_HIGH_MEMORY);
672 
673         if (arg->status_change_nid >= 0)
674                 node_set_state(node, N_MEMORY);
675 }
676 
677 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
678                 unsigned long nr_pages)
679 {
680         unsigned long old_end_pfn = zone_end_pfn(zone);
681 
682         if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
683                 zone->zone_start_pfn = start_pfn;
684 
685         zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
686 }
687 
688 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
689                                      unsigned long nr_pages)
690 {
691         unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
692 
693         if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
694                 pgdat->node_start_pfn = start_pfn;
695 
696         pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
697 
698 }
699 /*
700  * Associate the pfn range with the given zone, initializing the memmaps
701  * and resizing the pgdat/zone data to span the added pages. After this
702  * call, all affected pages are PG_reserved.
703  */
704 void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
705                 unsigned long nr_pages, struct vmem_altmap *altmap)
706 {
707         struct pglist_data *pgdat = zone->zone_pgdat;
708         int nid = pgdat->node_id;
709         unsigned long flags;
710 
711         clear_zone_contiguous(zone);
712 
713         /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
714         pgdat_resize_lock(pgdat, &flags);
715         zone_span_writelock(zone);
716         if (zone_is_empty(zone))
717                 init_currently_empty_zone(zone, start_pfn, nr_pages);
718         resize_zone_range(zone, start_pfn, nr_pages);
719         zone_span_writeunlock(zone);
720         resize_pgdat_range(pgdat, start_pfn, nr_pages);
721         pgdat_resize_unlock(pgdat, &flags);
722 
723         /*
724          * TODO now we have a visible range of pages which are not associated
725          * with their zone properly. Not nice but set_pfnblock_flags_mask
726          * expects the zone spans the pfn range. All the pages in the range
727          * are reserved so nobody should be touching them so we should be safe
728          */
729         memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
730                         MEMMAP_HOTPLUG, altmap);
731 
732         set_zone_contiguous(zone);
733 }
734 
735 /*
736  * Returns a default kernel memory zone for the given pfn range.
737  * If no kernel zone covers this pfn range it will automatically go
738  * to the ZONE_NORMAL.
739  */
740 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
741                 unsigned long nr_pages)
742 {
743         struct pglist_data *pgdat = NODE_DATA(nid);
744         int zid;
745 
746         for (zid = 0; zid <= ZONE_NORMAL; zid++) {
747                 struct zone *zone = &pgdat->node_zones[zid];
748 
749                 if (zone_intersects(zone, start_pfn, nr_pages))
750                         return zone;
751         }
752 
753         return &pgdat->node_zones[ZONE_NORMAL];
754 }
755 
756 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
757                 unsigned long nr_pages)
758 {
759         struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
760                         nr_pages);
761         struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
762         bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
763         bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
764 
765         /*
766          * We inherit the existing zone in a simple case where zones do not
767          * overlap in the given range
768          */
769         if (in_kernel ^ in_movable)
770                 return (in_kernel) ? kernel_zone : movable_zone;
771 
772         /*
773          * If the range doesn't belong to any zone or two zones overlap in the
774          * given range then we use movable zone only if movable_node is
775          * enabled because we always online to a kernel zone by default.
776          */
777         return movable_node_enabled ? movable_zone : kernel_zone;
778 }
779 
780 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
781                 unsigned long nr_pages)
782 {
783         if (online_type == MMOP_ONLINE_KERNEL)
784                 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
785 
786         if (online_type == MMOP_ONLINE_MOVABLE)
787                 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
788 
789         return default_zone_for_pfn(nid, start_pfn, nr_pages);
790 }
791 
792 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
793 {
794         unsigned long flags;
795         unsigned long onlined_pages = 0;
796         struct zone *zone;
797         int need_zonelists_rebuild = 0;
798         int nid;
799         int ret;
800         struct memory_notify arg;
801         struct memory_block *mem;
802 
803         mem_hotplug_begin();
804 
805         /*
806          * We can't use pfn_to_nid() because nid might be stored in struct page
807          * which is not yet initialized. Instead, we find nid from memory block.
808          */
809         mem = find_memory_block(__pfn_to_section(pfn));
810         nid = mem->nid;
811         put_device(&mem->dev);
812 
813         /* associate pfn range with the zone */
814         zone = zone_for_pfn_range(online_type, nid, pfn, nr_pages);
815         move_pfn_range_to_zone(zone, pfn, nr_pages, NULL);
816 
817         arg.start_pfn = pfn;
818         arg.nr_pages = nr_pages;
819         node_states_check_changes_online(nr_pages, zone, &arg);
820 
821         ret = memory_notify(MEM_GOING_ONLINE, &arg);
822         ret = notifier_to_errno(ret);
823         if (ret)
824                 goto failed_addition;
825 
826         /*
827          * If this zone is not populated, then it is not in zonelist.
828          * This means the page allocator ignores this zone.
829          * So, zonelist must be updated after online.
830          */
831         if (!populated_zone(zone)) {
832                 need_zonelists_rebuild = 1;
833                 setup_zone_pageset(zone);
834         }
835 
836         ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
837                 online_pages_range);
838         if (ret) {
839                 /* not a single memory resource was applicable */
840                 if (need_zonelists_rebuild)
841                         zone_pcp_reset(zone);
842                 goto failed_addition;
843         }
844 
845         zone->present_pages += onlined_pages;
846 
847         pgdat_resize_lock(zone->zone_pgdat, &flags);
848         zone->zone_pgdat->node_present_pages += onlined_pages;
849         pgdat_resize_unlock(zone->zone_pgdat, &flags);
850 
851         shuffle_zone(zone);
852 
853         node_states_set_node(nid, &arg);
854         if (need_zonelists_rebuild)
855                 build_all_zonelists(NULL);
856         else
857                 zone_pcp_update(zone);
858 
859         init_per_zone_wmark_min();
860 
861         kswapd_run(nid);
862         kcompactd_run(nid);
863 
864         vm_total_pages = nr_free_pagecache_pages();
865 
866         writeback_set_ratelimit();
867 
868         memory_notify(MEM_ONLINE, &arg);
869         mem_hotplug_done();
870         return 0;
871 
872 failed_addition:
873         pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
874                  (unsigned long long) pfn << PAGE_SHIFT,
875                  (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
876         memory_notify(MEM_CANCEL_ONLINE, &arg);
877         remove_pfn_range_from_zone(zone, pfn, nr_pages);
878         mem_hotplug_done();
879         return ret;
880 }
881 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
882 
883 static void reset_node_present_pages(pg_data_t *pgdat)
884 {
885         struct zone *z;
886 
887         for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
888                 z->present_pages = 0;
889 
890         pgdat->node_present_pages = 0;
891 }
892 
893 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
894 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
895 {
896         struct pglist_data *pgdat;
897         unsigned long start_pfn = PFN_DOWN(start);
898 
899         pgdat = NODE_DATA(nid);
900         if (!pgdat) {
901                 pgdat = arch_alloc_nodedata(nid);
902                 if (!pgdat)
903                         return NULL;
904 
905                 pgdat->per_cpu_nodestats =
906                         alloc_percpu(struct per_cpu_nodestat);
907                 arch_refresh_nodedata(nid, pgdat);
908         } else {
909                 int cpu;
910                 /*
911                  * Reset the nr_zones, order and classzone_idx before reuse.
912                  * Note that kswapd will init kswapd_classzone_idx properly
913                  * when it starts in the near future.
914                  */
915                 pgdat->nr_zones = 0;
916                 pgdat->kswapd_order = 0;
917                 pgdat->kswapd_classzone_idx = 0;
918                 for_each_online_cpu(cpu) {
919                         struct per_cpu_nodestat *p;
920 
921                         p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
922                         memset(p, 0, sizeof(*p));
923                 }
924         }
925 
926         /* we can use NODE_DATA(nid) from here */
927 
928         pgdat->node_id = nid;
929         pgdat->node_start_pfn = start_pfn;
930 
931         /* init node's zones as empty zones, we don't have any present pages.*/
932         free_area_init_core_hotplug(nid);
933 
934         /*
935          * The node we allocated has no zone fallback lists. For avoiding
936          * to access not-initialized zonelist, build here.
937          */
938         build_all_zonelists(pgdat);
939 
940         /*
941          * When memory is hot-added, all the memory is in offline state. So
942          * clear all zones' present_pages because they will be updated in
943          * online_pages() and offline_pages().
944          */
945         reset_node_managed_pages(pgdat);
946         reset_node_present_pages(pgdat);
947 
948         return pgdat;
949 }
950 
951 static void rollback_node_hotadd(int nid)
952 {
953         pg_data_t *pgdat = NODE_DATA(nid);
954 
955         arch_refresh_nodedata(nid, NULL);
956         free_percpu(pgdat->per_cpu_nodestats);
957         arch_free_nodedata(pgdat);
958 }
959 
960 
961 /**
962  * try_online_node - online a node if offlined
963  * @nid: the node ID
964  * @start: start addr of the node
965  * @set_node_online: Whether we want to online the node
966  * called by cpu_up() to online a node without onlined memory.
967  *
968  * Returns:
969  * 1 -> a new node has been allocated
970  * 0 -> the node is already online
971  * -ENOMEM -> the node could not be allocated
972  */
973 static int __try_online_node(int nid, u64 start, bool set_node_online)
974 {
975         pg_data_t *pgdat;
976         int ret = 1;
977 
978         if (node_online(nid))
979                 return 0;
980 
981         pgdat = hotadd_new_pgdat(nid, start);
982         if (!pgdat) {
983                 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
984                 ret = -ENOMEM;
985                 goto out;
986         }
987 
988         if (set_node_online) {
989                 node_set_online(nid);
990                 ret = register_one_node(nid);
991                 BUG_ON(ret);
992         }
993 out:
994         return ret;
995 }
996 
997 /*
998  * Users of this function always want to online/register the node
999  */
1000 int try_online_node(int nid)
1001 {
1002         int ret;
1003 
1004         mem_hotplug_begin();
1005         ret =  __try_online_node(nid, 0, true);
1006         mem_hotplug_done();
1007         return ret;
1008 }
1009 
1010 static int check_hotplug_memory_range(u64 start, u64 size)
1011 {
1012         /* memory range must be block size aligned */
1013         if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) ||
1014             !IS_ALIGNED(size, memory_block_size_bytes())) {
1015                 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1016                        memory_block_size_bytes(), start, size);
1017                 return -EINVAL;
1018         }
1019 
1020         return 0;
1021 }
1022 
1023 static int online_memory_block(struct memory_block *mem, void *arg)
1024 {
1025         return device_online(&mem->dev);
1026 }
1027 
1028 /*
1029  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1030  * and online/offline operations (triggered e.g. by sysfs).
1031  *
1032  * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1033  */
1034 int __ref add_memory_resource(int nid, struct resource *res)
1035 {
1036         struct mhp_restrictions restrictions = {};
1037         u64 start, size;
1038         bool new_node = false;
1039         int ret;
1040 
1041         start = res->start;
1042         size = resource_size(res);
1043 
1044         ret = check_hotplug_memory_range(start, size);
1045         if (ret)
1046                 return ret;
1047 
1048         mem_hotplug_begin();
1049 
1050         /*
1051          * Add new range to memblock so that when hotadd_new_pgdat() is called
1052          * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1053          * this new range and calculate total pages correctly.  The range will
1054          * be removed at hot-remove time.
1055          */
1056         memblock_add_node(start, size, nid);
1057 
1058         ret = __try_online_node(nid, start, false);
1059         if (ret < 0)
1060                 goto error;
1061         new_node = ret;
1062 
1063         /* call arch's memory hotadd */
1064         ret = arch_add_memory(nid, start, size, &restrictions);
1065         if (ret < 0)
1066                 goto error;
1067 
1068         /* create memory block devices after memory was added */
1069         ret = create_memory_block_devices(start, size);
1070         if (ret) {
1071                 arch_remove_memory(nid, start, size, NULL);
1072                 goto error;
1073         }
1074 
1075         if (new_node) {
1076                 /* If sysfs file of new node can't be created, cpu on the node
1077                  * can't be hot-added. There is no rollback way now.
1078                  * So, check by BUG_ON() to catch it reluctantly..
1079                  * We online node here. We can't roll back from here.
1080                  */
1081                 node_set_online(nid);
1082                 ret = __register_one_node(nid);
1083                 BUG_ON(ret);
1084         }
1085 
1086         /* link memory sections under this node.*/
1087         ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1));
1088         BUG_ON(ret);
1089 
1090         /* create new memmap entry */
1091         firmware_map_add_hotplug(start, start + size, "System RAM");
1092 
1093         /* device_online() will take the lock when calling online_pages() */
1094         mem_hotplug_done();
1095 
1096         /* online pages if requested */
1097         if (memhp_auto_online)
1098                 walk_memory_blocks(start, size, NULL, online_memory_block);
1099 
1100         return ret;
1101 error:
1102         /* rollback pgdat allocation and others */
1103         if (new_node)
1104                 rollback_node_hotadd(nid);
1105         memblock_remove(start, size);
1106         mem_hotplug_done();
1107         return ret;
1108 }
1109 
1110 /* requires device_hotplug_lock, see add_memory_resource() */
1111 int __ref __add_memory(int nid, u64 start, u64 size)
1112 {
1113         struct resource *res;
1114         int ret;
1115 
1116         res = register_memory_resource(start, size);
1117         if (IS_ERR(res))
1118                 return PTR_ERR(res);
1119 
1120         ret = add_memory_resource(nid, res);
1121         if (ret < 0)
1122                 release_memory_resource(res);
1123         return ret;
1124 }
1125 
1126 int add_memory(int nid, u64 start, u64 size)
1127 {
1128         int rc;
1129 
1130         lock_device_hotplug();
1131         rc = __add_memory(nid, start, size);
1132         unlock_device_hotplug();
1133 
1134         return rc;
1135 }
1136 EXPORT_SYMBOL_GPL(add_memory);
1137 
1138 #ifdef CONFIG_MEMORY_HOTREMOVE
1139 /*
1140  * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1141  * set and the size of the free page is given by page_order(). Using this,
1142  * the function determines if the pageblock contains only free pages.
1143  * Due to buddy contraints, a free page at least the size of a pageblock will
1144  * be located at the start of the pageblock
1145  */
1146 static inline int pageblock_free(struct page *page)
1147 {
1148         return PageBuddy(page) && page_order(page) >= pageblock_order;
1149 }
1150 
1151 /* Return the pfn of the start of the next active pageblock after a given pfn */
1152 static unsigned long next_active_pageblock(unsigned long pfn)
1153 {
1154         struct page *page = pfn_to_page(pfn);
1155 
1156         /* Ensure the starting page is pageblock-aligned */
1157         BUG_ON(pfn & (pageblock_nr_pages - 1));
1158 
1159         /* If the entire pageblock is free, move to the end of free page */
1160         if (pageblock_free(page)) {
1161                 int order;
1162                 /* be careful. we don't have locks, page_order can be changed.*/
1163                 order = page_order(page);
1164                 if ((order < MAX_ORDER) && (order >= pageblock_order))
1165                         return pfn + (1 << order);
1166         }
1167 
1168         return pfn + pageblock_nr_pages;
1169 }
1170 
1171 static bool is_pageblock_removable_nolock(unsigned long pfn)
1172 {
1173         struct page *page = pfn_to_page(pfn);
1174         struct zone *zone;
1175 
1176         /*
1177          * We have to be careful here because we are iterating over memory
1178          * sections which are not zone aware so we might end up outside of
1179          * the zone but still within the section.
1180          * We have to take care about the node as well. If the node is offline
1181          * its NODE_DATA will be NULL - see page_zone.
1182          */
1183         if (!node_online(page_to_nid(page)))
1184                 return false;
1185 
1186         zone = page_zone(page);
1187         pfn = page_to_pfn(page);
1188         if (!zone_spans_pfn(zone, pfn))
1189                 return false;
1190 
1191         return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE,
1192                                     MEMORY_OFFLINE);
1193 }
1194 
1195 /* Checks if this range of memory is likely to be hot-removable. */
1196 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1197 {
1198         unsigned long end_pfn, pfn;
1199 
1200         end_pfn = min(start_pfn + nr_pages,
1201                         zone_end_pfn(page_zone(pfn_to_page(start_pfn))));
1202 
1203         /* Check the starting page of each pageblock within the range */
1204         for (pfn = start_pfn; pfn < end_pfn; pfn = next_active_pageblock(pfn)) {
1205                 if (!is_pageblock_removable_nolock(pfn))
1206                         return false;
1207                 cond_resched();
1208         }
1209 
1210         /* All pageblocks in the memory block are likely to be hot-removable */
1211         return true;
1212 }
1213 
1214 /*
1215  * Confirm all pages in a range [start, end) belong to the same zone.
1216  * When true, return its valid [start, end).
1217  */
1218 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1219                          unsigned long *valid_start, unsigned long *valid_end)
1220 {
1221         unsigned long pfn, sec_end_pfn;
1222         unsigned long start, end;
1223         struct zone *zone = NULL;
1224         struct page *page;
1225         int i;
1226         for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1227              pfn < end_pfn;
1228              pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1229                 /* Make sure the memory section is present first */
1230                 if (!present_section_nr(pfn_to_section_nr(pfn)))
1231                         continue;
1232                 for (; pfn < sec_end_pfn && pfn < end_pfn;
1233                      pfn += MAX_ORDER_NR_PAGES) {
1234                         i = 0;
1235                         /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1236                         while ((i < MAX_ORDER_NR_PAGES) &&
1237                                 !pfn_valid_within(pfn + i))
1238                                 i++;
1239                         if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1240                                 continue;
1241                         /* Check if we got outside of the zone */
1242                         if (zone && !zone_spans_pfn(zone, pfn + i))
1243                                 return 0;
1244                         page = pfn_to_page(pfn + i);
1245                         if (zone && page_zone(page) != zone)
1246                                 return 0;
1247                         if (!zone)
1248                                 start = pfn + i;
1249                         zone = page_zone(page);
1250                         end = pfn + MAX_ORDER_NR_PAGES;
1251                 }
1252         }
1253 
1254         if (zone) {
1255                 *valid_start = start;
1256                 *valid_end = min(end, end_pfn);
1257                 return 1;
1258         } else {
1259                 return 0;
1260         }
1261 }
1262 
1263 /*
1264  * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1265  * non-lru movable pages and hugepages). We scan pfn because it's much
1266  * easier than scanning over linked list. This function returns the pfn
1267  * of the first found movable page if it's found, otherwise 0.
1268  */
1269 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1270 {
1271         unsigned long pfn;
1272 
1273         for (pfn = start; pfn < end; pfn++) {
1274                 struct page *page, *head;
1275                 unsigned long skip;
1276 
1277                 if (!pfn_valid(pfn))
1278                         continue;
1279                 page = pfn_to_page(pfn);
1280                 if (PageLRU(page))
1281                         return pfn;
1282                 if (__PageMovable(page))
1283                         return pfn;
1284 
1285                 if (!PageHuge(page))
1286                         continue;
1287                 head = compound_head(page);
1288                 if (page_huge_active(head))
1289                         return pfn;
1290                 skip = compound_nr(head) - (page - head);
1291                 pfn += skip - 1;
1292         }
1293         return 0;
1294 }
1295 
1296 static struct page *new_node_page(struct page *page, unsigned long private)
1297 {
1298         int nid = page_to_nid(page);
1299         nodemask_t nmask = node_states[N_MEMORY];
1300 
1301         /*
1302          * try to allocate from a different node but reuse this node if there
1303          * are no other online nodes to be used (e.g. we are offlining a part
1304          * of the only existing node)
1305          */
1306         node_clear(nid, nmask);
1307         if (nodes_empty(nmask))
1308                 node_set(nid, nmask);
1309 
1310         return new_page_nodemask(page, nid, &nmask);
1311 }
1312 
1313 static int
1314 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1315 {
1316         unsigned long pfn;
1317         struct page *page;
1318         int ret = 0;
1319         LIST_HEAD(source);
1320 
1321         for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1322                 if (!pfn_valid(pfn))
1323                         continue;
1324                 page = pfn_to_page(pfn);
1325 
1326                 if (PageHuge(page)) {
1327                         struct page *head = compound_head(page);
1328                         pfn = page_to_pfn(head) + compound_nr(head) - 1;
1329                         isolate_huge_page(head, &source);
1330                         continue;
1331                 } else if (PageTransHuge(page))
1332                         pfn = page_to_pfn(compound_head(page))
1333                                 + hpage_nr_pages(page) - 1;
1334 
1335                 /*
1336                  * HWPoison pages have elevated reference counts so the migration would
1337                  * fail on them. It also doesn't make any sense to migrate them in the
1338                  * first place. Still try to unmap such a page in case it is still mapped
1339                  * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1340                  * the unmap as the catch all safety net).
1341                  */
1342                 if (PageHWPoison(page)) {
1343                         if (WARN_ON(PageLRU(page)))
1344                                 isolate_lru_page(page);
1345                         if (page_mapped(page))
1346                                 try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1347                         continue;
1348                 }
1349 
1350                 if (!get_page_unless_zero(page))
1351                         continue;
1352                 /*
1353                  * We can skip free pages. And we can deal with pages on
1354                  * LRU and non-lru movable pages.
1355                  */
1356                 if (PageLRU(page))
1357                         ret = isolate_lru_page(page);
1358                 else
1359                         ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1360                 if (!ret) { /* Success */
1361                         list_add_tail(&page->lru, &source);
1362                         if (!__PageMovable(page))
1363                                 inc_node_page_state(page, NR_ISOLATED_ANON +
1364                                                     page_is_file_cache(page));
1365 
1366                 } else {
1367                         pr_warn("failed to isolate pfn %lx\n", pfn);
1368                         dump_page(page, "isolation failed");
1369                 }
1370                 put_page(page);
1371         }
1372         if (!list_empty(&source)) {
1373                 /* Allocate a new page from the nearest neighbor node */
1374                 ret = migrate_pages(&source, new_node_page, NULL, 0,
1375                                         MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1376                 if (ret) {
1377                         list_for_each_entry(page, &source, lru) {
1378                                 pr_warn("migrating pfn %lx failed ret:%d ",
1379                                        page_to_pfn(page), ret);
1380                                 dump_page(page, "migration failure");
1381                         }
1382                         putback_movable_pages(&source);
1383                 }
1384         }
1385 
1386         return ret;
1387 }
1388 
1389 /* Mark all sections offline and remove all free pages from the buddy. */
1390 static int
1391 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1392                         void *data)
1393 {
1394         unsigned long *offlined_pages = (unsigned long *)data;
1395 
1396         *offlined_pages += __offline_isolated_pages(start, start + nr_pages);
1397         return 0;
1398 }
1399 
1400 /*
1401  * Check all pages in range, recoreded as memory resource, are isolated.
1402  */
1403 static int
1404 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1405                         void *data)
1406 {
1407         return test_pages_isolated(start_pfn, start_pfn + nr_pages,
1408                                    MEMORY_OFFLINE);
1409 }
1410 
1411 static int __init cmdline_parse_movable_node(char *p)
1412 {
1413 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1414         movable_node_enabled = true;
1415 #else
1416         pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1417 #endif
1418         return 0;
1419 }
1420 early_param("movable_node", cmdline_parse_movable_node);
1421 
1422 /* check which state of node_states will be changed when offline memory */
1423 static void node_states_check_changes_offline(unsigned long nr_pages,
1424                 struct zone *zone, struct memory_notify *arg)
1425 {
1426         struct pglist_data *pgdat = zone->zone_pgdat;
1427         unsigned long present_pages = 0;
1428         enum zone_type zt;
1429 
1430         arg->status_change_nid = NUMA_NO_NODE;
1431         arg->status_change_nid_normal = NUMA_NO_NODE;
1432         arg->status_change_nid_high = NUMA_NO_NODE;
1433 
1434         /*
1435          * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1436          * If the memory to be offline is within the range
1437          * [0..ZONE_NORMAL], and it is the last present memory there,
1438          * the zones in that range will become empty after the offlining,
1439          * thus we can determine that we need to clear the node from
1440          * node_states[N_NORMAL_MEMORY].
1441          */
1442         for (zt = 0; zt <= ZONE_NORMAL; zt++)
1443                 present_pages += pgdat->node_zones[zt].present_pages;
1444         if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
1445                 arg->status_change_nid_normal = zone_to_nid(zone);
1446 
1447 #ifdef CONFIG_HIGHMEM
1448         /*
1449          * node_states[N_HIGH_MEMORY] contains nodes which
1450          * have normal memory or high memory.
1451          * Here we add the present_pages belonging to ZONE_HIGHMEM.
1452          * If the zone is within the range of [0..ZONE_HIGHMEM), and
1453          * we determine that the zones in that range become empty,
1454          * we need to clear the node for N_HIGH_MEMORY.
1455          */
1456         present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1457         if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
1458                 arg->status_change_nid_high = zone_to_nid(zone);
1459 #endif
1460 
1461         /*
1462          * We have accounted the pages from [0..ZONE_NORMAL), and
1463          * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1464          * as well.
1465          * Here we count the possible pages from ZONE_MOVABLE.
1466          * If after having accounted all the pages, we see that the nr_pages
1467          * to be offlined is over or equal to the accounted pages,
1468          * we know that the node will become empty, and so, we can clear
1469          * it for N_MEMORY as well.
1470          */
1471         present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
1472 
1473         if (nr_pages >= present_pages)
1474                 arg->status_change_nid = zone_to_nid(zone);
1475 }
1476 
1477 static void node_states_clear_node(int node, struct memory_notify *arg)
1478 {
1479         if (arg->status_change_nid_normal >= 0)
1480                 node_clear_state(node, N_NORMAL_MEMORY);
1481 
1482         if (arg->status_change_nid_high >= 0)
1483                 node_clear_state(node, N_HIGH_MEMORY);
1484 
1485         if (arg->status_change_nid >= 0)
1486                 node_clear_state(node, N_MEMORY);
1487 }
1488 
1489 static int count_system_ram_pages_cb(unsigned long start_pfn,
1490                                      unsigned long nr_pages, void *data)
1491 {
1492         unsigned long *nr_system_ram_pages = data;
1493 
1494         *nr_system_ram_pages += nr_pages;
1495         return 0;
1496 }
1497 
1498 static int __ref __offline_pages(unsigned long start_pfn,
1499                   unsigned long end_pfn)
1500 {
1501         unsigned long pfn, nr_pages = 0;
1502         unsigned long offlined_pages = 0;
1503         int ret, node, nr_isolate_pageblock;
1504         unsigned long flags;
1505         unsigned long valid_start, valid_end;
1506         struct zone *zone;
1507         struct memory_notify arg;
1508         char *reason;
1509 
1510         mem_hotplug_begin();
1511 
1512         /*
1513          * Don't allow to offline memory blocks that contain holes.
1514          * Consequently, memory blocks with holes can never get onlined
1515          * via the hotplug path - online_pages() - as hotplugged memory has
1516          * no holes. This way, we e.g., don't have to worry about marking
1517          * memory holes PG_reserved, don't need pfn_valid() checks, and can
1518          * avoid using walk_system_ram_range() later.
1519          */
1520         walk_system_ram_range(start_pfn, end_pfn - start_pfn, &nr_pages,
1521                               count_system_ram_pages_cb);
1522         if (nr_pages != end_pfn - start_pfn) {
1523                 ret = -EINVAL;
1524                 reason = "memory holes";
1525                 goto failed_removal;
1526         }
1527 
1528         /* This makes hotplug much easier...and readable.
1529            we assume this for now. .*/
1530         if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start,
1531                                   &valid_end)) {
1532                 ret = -EINVAL;
1533                 reason = "multizone range";
1534                 goto failed_removal;
1535         }
1536 
1537         zone = page_zone(pfn_to_page(valid_start));
1538         node = zone_to_nid(zone);
1539 
1540         /* set above range as isolated */
1541         ret = start_isolate_page_range(start_pfn, end_pfn,
1542                                        MIGRATE_MOVABLE,
1543                                        MEMORY_OFFLINE | REPORT_FAILURE);
1544         if (ret < 0) {
1545                 reason = "failure to isolate range";
1546                 goto failed_removal;
1547         }
1548         nr_isolate_pageblock = ret;
1549 
1550         arg.start_pfn = start_pfn;
1551         arg.nr_pages = nr_pages;
1552         node_states_check_changes_offline(nr_pages, zone, &arg);
1553 
1554         ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1555         ret = notifier_to_errno(ret);
1556         if (ret) {
1557                 reason = "notifier failure";
1558                 goto failed_removal_isolated;
1559         }
1560 
1561         do {
1562                 for (pfn = start_pfn; pfn;) {
1563                         if (signal_pending(current)) {
1564                                 ret = -EINTR;
1565                                 reason = "signal backoff";
1566                                 goto failed_removal_isolated;
1567                         }
1568 
1569                         cond_resched();
1570                         lru_add_drain_all();
1571 
1572                         pfn = scan_movable_pages(pfn, end_pfn);
1573                         if (pfn) {
1574                                 /*
1575                                  * TODO: fatal migration failures should bail
1576                                  * out
1577                                  */
1578                                 do_migrate_range(pfn, end_pfn);
1579                         }
1580                 }
1581 
1582                 /*
1583                  * Dissolve free hugepages in the memory block before doing
1584                  * offlining actually in order to make hugetlbfs's object
1585                  * counting consistent.
1586                  */
1587                 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1588                 if (ret) {
1589                         reason = "failure to dissolve huge pages";
1590                         goto failed_removal_isolated;
1591                 }
1592                 /* check again */
1593                 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1594                                             NULL, check_pages_isolated_cb);
1595         } while (ret);
1596 
1597         /* Ok, all of our target is isolated.
1598            We cannot do rollback at this point. */
1599         walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1600                               &offlined_pages, offline_isolated_pages_cb);
1601         pr_info("Offlined Pages %ld\n", offlined_pages);
1602         /*
1603          * Onlining will reset pagetype flags and makes migrate type
1604          * MOVABLE, so just need to decrease the number of isolated
1605          * pageblocks zone counter here.
1606          */
1607         spin_lock_irqsave(&zone->lock, flags);
1608         zone->nr_isolate_pageblock -= nr_isolate_pageblock;
1609         spin_unlock_irqrestore(&zone->lock, flags);
1610 
1611         /* removal success */
1612         adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1613         zone->present_pages -= offlined_pages;
1614 
1615         pgdat_resize_lock(zone->zone_pgdat, &flags);
1616         zone->zone_pgdat->node_present_pages -= offlined_pages;
1617         pgdat_resize_unlock(zone->zone_pgdat, &flags);
1618 
1619         init_per_zone_wmark_min();
1620 
1621         if (!populated_zone(zone)) {
1622                 zone_pcp_reset(zone);
1623                 build_all_zonelists(NULL);
1624         } else
1625                 zone_pcp_update(zone);
1626 
1627         node_states_clear_node(node, &arg);
1628         if (arg.status_change_nid >= 0) {
1629                 kswapd_stop(node);
1630                 kcompactd_stop(node);
1631         }
1632 
1633         vm_total_pages = nr_free_pagecache_pages();
1634         writeback_set_ratelimit();
1635 
1636         memory_notify(MEM_OFFLINE, &arg);
1637         remove_pfn_range_from_zone(zone, start_pfn, nr_pages);
1638         mem_hotplug_done();
1639         return 0;
1640 
1641 failed_removal_isolated:
1642         undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1643         memory_notify(MEM_CANCEL_OFFLINE, &arg);
1644 failed_removal:
1645         pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1646                  (unsigned long long) start_pfn << PAGE_SHIFT,
1647                  ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1648                  reason);
1649         /* pushback to free area */
1650         mem_hotplug_done();
1651         return ret;
1652 }
1653 
1654 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1655 {
1656         return __offline_pages(start_pfn, start_pfn + nr_pages);
1657 }
1658 
1659 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1660 {
1661         int ret = !is_memblock_offlined(mem);
1662 
1663         if (unlikely(ret)) {
1664                 phys_addr_t beginpa, endpa;
1665 
1666                 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1667                 endpa = beginpa + memory_block_size_bytes() - 1;
1668                 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1669                         &beginpa, &endpa);
1670 
1671                 return -EBUSY;
1672         }
1673         return 0;
1674 }
1675 
1676 static int check_cpu_on_node(pg_data_t *pgdat)
1677 {
1678         int cpu;
1679 
1680         for_each_present_cpu(cpu) {
1681                 if (cpu_to_node(cpu) == pgdat->node_id)
1682                         /*
1683                          * the cpu on this node isn't removed, and we can't
1684                          * offline this node.
1685                          */
1686                         return -EBUSY;
1687         }
1688 
1689         return 0;
1690 }
1691 
1692 static int check_no_memblock_for_node_cb(struct memory_block *mem, void *arg)
1693 {
1694         int nid = *(int *)arg;
1695 
1696         /*
1697          * If a memory block belongs to multiple nodes, the stored nid is not
1698          * reliable. However, such blocks are always online (e.g., cannot get
1699          * offlined) and, therefore, are still spanned by the node.
1700          */
1701         return mem->nid == nid ? -EEXIST : 0;
1702 }
1703 
1704 /**
1705  * try_offline_node
1706  * @nid: the node ID
1707  *
1708  * Offline a node if all memory sections and cpus of the node are removed.
1709  *
1710  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1711  * and online/offline operations before this call.
1712  */
1713 void try_offline_node(int nid)
1714 {
1715         pg_data_t *pgdat = NODE_DATA(nid);
1716         int rc;
1717 
1718         /*
1719          * If the node still spans pages (especially ZONE_DEVICE), don't
1720          * offline it. A node spans memory after move_pfn_range_to_zone(),
1721          * e.g., after the memory block was onlined.
1722          */
1723         if (pgdat->node_spanned_pages)
1724                 return;
1725 
1726         /*
1727          * Especially offline memory blocks might not be spanned by the
1728          * node. They will get spanned by the node once they get onlined.
1729          * However, they link to the node in sysfs and can get onlined later.
1730          */
1731         rc = for_each_memory_block(&nid, check_no_memblock_for_node_cb);
1732         if (rc)
1733                 return;
1734 
1735         if (check_cpu_on_node(pgdat))
1736                 return;
1737 
1738         /*
1739          * all memory/cpu of this node are removed, we can offline this
1740          * node now.
1741          */
1742         node_set_offline(nid);
1743         unregister_one_node(nid);
1744 }
1745 EXPORT_SYMBOL(try_offline_node);
1746 
1747 static void __release_memory_resource(resource_size_t start,
1748                                       resource_size_t size)
1749 {
1750         int ret;
1751 
1752         /*
1753          * When removing memory in the same granularity as it was added,
1754          * this function never fails. It might only fail if resources
1755          * have to be adjusted or split. We'll ignore the error, as
1756          * removing of memory cannot fail.
1757          */
1758         ret = release_mem_region_adjustable(&iomem_resource, start, size);
1759         if (ret) {
1760                 resource_size_t endres = start + size - 1;
1761 
1762                 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
1763                         &start, &endres, ret);
1764         }
1765 }
1766 
1767 static int __ref try_remove_memory(int nid, u64 start, u64 size)
1768 {
1769         int rc = 0;
1770 
1771         BUG_ON(check_hotplug_memory_range(start, size));
1772 
1773         /*
1774          * All memory blocks must be offlined before removing memory.  Check
1775          * whether all memory blocks in question are offline and return error
1776          * if this is not the case.
1777          */
1778         rc = walk_memory_blocks(start, size, NULL, check_memblock_offlined_cb);
1779         if (rc)
1780                 goto done;
1781 
1782         /* remove memmap entry */
1783         firmware_map_remove(start, start + size, "System RAM");
1784 
1785         /*
1786          * Memory block device removal under the device_hotplug_lock is
1787          * a barrier against racing online attempts.
1788          */
1789         remove_memory_block_devices(start, size);
1790 
1791         mem_hotplug_begin();
1792 
1793         arch_remove_memory(nid, start, size, NULL);
1794         memblock_free(start, size);
1795         memblock_remove(start, size);
1796         __release_memory_resource(start, size);
1797 
1798         try_offline_node(nid);
1799 
1800 done:
1801         mem_hotplug_done();
1802         return rc;
1803 }
1804 
1805 /**
1806  * remove_memory
1807  * @nid: the node ID
1808  * @start: physical address of the region to remove
1809  * @size: size of the region to remove
1810  *
1811  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1812  * and online/offline operations before this call, as required by
1813  * try_offline_node().
1814  */
1815 void __remove_memory(int nid, u64 start, u64 size)
1816 {
1817 
1818         /*
1819          * trigger BUG() if some memory is not offlined prior to calling this
1820          * function
1821          */
1822         if (try_remove_memory(nid, start, size))
1823                 BUG();
1824 }
1825 
1826 /*
1827  * Remove memory if every memory block is offline, otherwise return -EBUSY is
1828  * some memory is not offline
1829  */
1830 int remove_memory(int nid, u64 start, u64 size)
1831 {
1832         int rc;
1833 
1834         lock_device_hotplug();
1835         rc  = try_remove_memory(nid, start, size);
1836         unlock_device_hotplug();
1837 
1838         return rc;
1839 }
1840 EXPORT_SYMBOL_GPL(remove_memory);
1841 #endif /* CONFIG_MEMORY_HOTREMOVE */
1842 

~ [ 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