mm/memory_hotplug.c

   1 /*
   2  *  linux/mm/memory_hotplug.c
   3  *
   4  *  Copyright (C)
   5  */
   6
   7 #include <linux/stddef.h>
   8 #include <linux/mm.h>
   9 #include <linux/swap.h>
  10 #include <linux/interrupt.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/bootmem.h>
  13 #include <linux/compiler.h>
  14 #include <linux/module.h>
  15 #include <linux/pagevec.h>
  16 #include <linux/slab.h>
  17 #include <linux/sysctl.h>
  18 #include <linux/cpu.h>
  19 #include <linux/memory.h>
  20 #include <linux/memory_hotplug.h>
  21 #include <linux/highmem.h>
  22 #include <linux/vmalloc.h>
  23 #include <linux/ioport.h>
  24
  25 #include <asm/tlbflush.h>
  26
  27 extern void zonetable_add(struct zone *zone, int nid, int zid, unsigned long pfn,
  28                           unsigned long size);
  29 static int __add_zone(struct zone *zone, unsigned long phys_start_pfn)
  30 {
  31         struct pglist_data *pgdat = zone->zone_pgdat;
  32         int nr_pages = PAGES_PER_SECTION;
  33         int nid = pgdat->node_id;
  34         int zone_type;
  35
  36         zone_type = zone - pgdat->node_zones;
  37         if (!populated_zone(zone)) {
  38                 int ret = 0;
  39                 ret = init_currently_empty_zone(zone, phys_start_pfn, nr_pages);
  40                 if (ret < 0)
  41                         return ret;
  42         }
  43         memmap_init_zone(nr_pages, nid, zone_type, phys_start_pfn);
  44         zonetable_add(zone, nid, zone_type, phys_start_pfn, nr_pages);
  45         return 0;
  46 }
  47
  48 extern int sparse_add_one_section(struct zone *zone, unsigned long start_pfn,
  49                                   int nr_pages);
  50 static int __add_section(struct zone *zone, unsigned long phys_start_pfn)
  51 {
  52         int nr_pages = PAGES_PER_SECTION;
  53         int ret;
  54
  55         if (pfn_valid(phys_start_pfn))
  56                 return -EEXIST;
  57
  58         ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
  59
  60         if (ret < 0)
  61                 return ret;
  62
  63         ret = __add_zone(zone, phys_start_pfn);
  64
  65         if (ret < 0)
  66                 return ret;
  67
  68         return register_new_memory(__pfn_to_section(phys_start_pfn));
  69 }
  70
  71 /*
  72  * Reasonably generic function for adding memory.  It is
  73  * expected that archs that support memory hotplug will
  74  * call this function after deciding the zone to which to
  75  * add the new pages.
  76  */
  77 int __add_pages(struct zone *zone, unsigned long phys_start_pfn,
  78                  unsigned long nr_pages)
  79 {
  80         unsigned long i;
  81         int err = 0;
  82         int start_sec, end_sec;
  83         /* during initialize mem_map, align hot-added range to section */
  84         start_sec = pfn_to_section_nr(phys_start_pfn);
  85         end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
  86
  87         for (i = start_sec; i <= end_sec; i++) {
  88                 err = __add_section(zone, i << PFN_SECTION_SHIFT);
  89
  90                 /*
  91                  * EEXIST is finally dealed with by ioresource collision
  92                  * check. see add_memory() => register_memory_resource()
  93                  * Warning will be printed if there is collision.
  94                  */
  95                 if (err && (err != -EEXIST))
  96                         break;
  97                 err = 0;
  98         }
  99
 100         return err;
 101 }
 102 EXPORT_SYMBOL_GPL(__add_pages);
 103
 104 static void grow_zone_span(struct zone *zone,
 105                 unsigned long start_pfn, unsigned long end_pfn)
 106 {
 107         unsigned long old_zone_end_pfn;
 108
 109         zone_span_writelock(zone);
 110
 111         old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
 112         if (start_pfn < zone->zone_start_pfn)
 113                 zone->zone_start_pfn = start_pfn;
 114
 115         zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
 116                                 zone->zone_start_pfn;
 117
 118         zone_span_writeunlock(zone);
 119 }
 120
 121 static void grow_pgdat_span(struct pglist_data *pgdat,
 122                 unsigned long start_pfn, unsigned long end_pfn)
 123 {
 124         unsigned long old_pgdat_end_pfn =
 125                 pgdat->node_start_pfn + pgdat->node_spanned_pages;
 126
 127         if (start_pfn < pgdat->node_start_pfn)
 128                 pgdat->node_start_pfn = start_pfn;
 129
 130         pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
 131                                         pgdat->node_start_pfn;
 132 }
 133
 134 int online_pages(unsigned long pfn, unsigned long nr_pages)
 135 {
 136         unsigned long i;
 137         unsigned long flags;
 138         unsigned long onlined_pages = 0;
 139         struct resource res;
 140         u64 section_end;
 141         unsigned long start_pfn;
 142         struct zone *zone;
 143         int need_zonelists_rebuild = 0;
 144
 145         /*
 146          * This doesn't need a lock to do pfn_to_page().
 147          * The section can't be removed here because of the
 148          * memory_block->state_sem.
 149          */
 150         zone = page_zone(pfn_to_page(pfn));
 151         pgdat_resize_lock(zone->zone_pgdat, &flags);
 152         grow_zone_span(zone, pfn, pfn + nr_pages);
 153         grow_pgdat_span(zone->zone_pgdat, pfn, pfn + nr_pages);
 154         pgdat_resize_unlock(zone->zone_pgdat, &flags);
 155
 156         /*
 157          * If this zone is not populated, then it is not in zonelist.
 158          * This means the page allocator ignores this zone.
 159          * So, zonelist must be updated after online.
 160          */
 161         if (!populated_zone(zone))
 162                 need_zonelists_rebuild = 1;
 163
 164         res.start = (u64)pfn << PAGE_SHIFT;
 165         res.end = res.start + ((u64)nr_pages << PAGE_SHIFT) - 1;
 166         res.flags = IORESOURCE_MEM; /* we just need system ram */
 167         section_end = res.end;
 168
 169         while ((res.start < res.end) && (find_next_system_ram(&res) >= 0)) {
 170                 start_pfn = (unsigned long)(res.start >> PAGE_SHIFT);
 171                 nr_pages = (unsigned long)
 172                            ((res.end + 1 - res.start) >> PAGE_SHIFT);
 173
 174                 if (PageReserved(pfn_to_page(start_pfn))) {
 175                         /* this region's page is not onlined now */
 176                         for (i = 0; i < nr_pages; i++) {
 177                                 struct page *page = pfn_to_page(start_pfn + i);
 178                                 online_page(page);
 179                                 onlined_pages++;
 180                         }
 181                 }
 182
 183                 res.start = res.end + 1;
 184                 res.end = section_end;
 185         }
 186         zone->present_pages += onlined_pages;
 187         zone->zone_pgdat->node_present_pages += onlined_pages;
 188
 189         setup_per_zone_pages_min();
 190
 191         if (need_zonelists_rebuild)
 192                 build_all_zonelists();
 193         vm_total_pages = nr_free_pagecache_pages();
 194         return 0;
 195 }
 196
 197 static pg_data_t *hotadd_new_pgdat(int nid, u64 start)
 198 {
 199         struct pglist_data *pgdat;
 200         unsigned long zones_size[MAX_NR_ZONES] = {0};
 201         unsigned long zholes_size[MAX_NR_ZONES] = {0};
 202         unsigned long start_pfn = start >> PAGE_SHIFT;
 203
 204         pgdat = arch_alloc_nodedata(nid);
 205         if (!pgdat)
 206                 return NULL;
 207
 208         arch_refresh_nodedata(nid, pgdat);
 209
 210         /* we can use NODE_DATA(nid) from here */
 211
 212         /* init node's zones as empty zones, we don't have any present pages.*/
 213         free_area_init_node(nid, pgdat, zones_size, start_pfn, zholes_size);
 214
 215         return pgdat;
 216 }
 217
 218 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
 219 {
 220         arch_refresh_nodedata(nid, NULL);
 221         arch_free_nodedata(pgdat);
 222         return;
 223 }
 224
 225 /* add this memory to iomem resource */
 226 static struct resource *register_memory_resource(u64 start, u64 size)
 227 {
 228         struct resource *res;
 229         res = kzalloc(sizeof(struct resource), GFP_KERNEL);
 230         BUG_ON(!res);
 231
 232         res->name = "System RAM";
 233         res->start = start;
 234         res->end = start + size - 1;
 235         res->flags = IORESOURCE_MEM;
 236         if (request_resource(&iomem_resource, res) < 0) {
 237                 printk("System RAM resource %llx - %llx cannot be added\n",
 238                 (unsigned long long)res->start, (unsigned long long)res->end);
 239                 kfree(res);
 240                 res = NULL;
 241         }
 242         return res;
 243 }
 244
 245 static void release_memory_resource(struct resource *res)
 246 {
 247         if (!res)
 248                 return;
 249         release_resource(res);
 250         kfree(res);
 251         return;
 252 }
 253
 254
 255
 256 int add_memory(int nid, u64 start, u64 size)
 257 {
 258         pg_data_t *pgdat = NULL;
 259         int new_pgdat = 0;
 260         struct resource *res;
 261         int ret;
 262
 263         res = register_memory_resource(start, size);
 264         if (!res)
 265                 return -EEXIST;
 266
 267         if (!node_online(nid)) {
 268                 pgdat = hotadd_new_pgdat(nid, start);
 269                 if (!pgdat)
 270                         return -ENOMEM;
 271                 new_pgdat = 1;
 272                 ret = kswapd_run(nid);
 273                 if (ret)
 274                         goto error;
 275         }
 276
 277         /* call arch's memory hotadd */
 278         ret = arch_add_memory(nid, start, size);
 279
 280         if (ret < 0)
 281                 goto error;
 282
 283         /* we online node here. we can't roll back from here. */
 284         node_set_online(nid);
 285
 286         if (new_pgdat) {
 287                 ret = register_one_node(nid);
 288                 /*
 289                  * If sysfs file of new node can't create, cpu on the node
 290                  * can't be hot-added. There is no rollback way now.
 291                  * So, check by BUG_ON() to catch it reluctantly..
 292                  */
 293                 BUG_ON(ret);
 294         }
 295
 296         return ret;
 297 error:
 298         /* rollback pgdat allocation and others */
 299         if (new_pgdat)
 300                 rollback_node_hotadd(nid, pgdat);
 301         if (res)
 302                 release_memory_resource(res);
 303
 304         return ret;
 305 }
 306 EXPORT_SYMBOL_GPL(add_memory);