mm/memory_hotplug.c

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