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/writeback.h>
  17 #include <linux/slab.h>
  18 #include <linux/sysctl.h>
  19 #include <linux/cpu.h>
  20 #include <linux/memory.h>
  21 #include <linux/memory_hotplug.h>
  22 #include <linux/highmem.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/ioport.h>
  25 #include <linux/cpuset.h>
  26
  27 #include <asm/tlbflush.h>
  28
  29 extern void zonetable_add(struct zone *zone, int nid, int zid, unsigned long pfn,
  30                           unsigned long size);
  31 static int __add_zone(struct zone *zone, unsigned long phys_start_pfn)
  32 {
  33         struct pglist_data *pgdat = zone->zone_pgdat;
  34         int nr_pages = PAGES_PER_SECTION;
  35         int nid = pgdat->node_id;
  36         int zone_type;
  37
  38         zone_type = zone - pgdat->node_zones;
  39         if (!populated_zone(zone)) {
  40                 int ret = 0;
  41                 ret = init_currently_empty_zone(zone, phys_start_pfn, nr_pages);
  42                 if (ret < 0)
  43                         return ret;
  44         }
  45         memmap_init_zone(nr_pages, nid, zone_type, phys_start_pfn);
  46         zonetable_add(zone, nid, zone_type, phys_start_pfn, nr_pages);
  47         return 0;
  48 }
  49
  50 extern int sparse_add_one_section(struct zone *zone, unsigned long start_pfn,
  51                                   int nr_pages);
  52 static int __add_section(struct zone *zone, unsigned long phys_start_pfn)
  53 {
  54         int nr_pages = PAGES_PER_SECTION;
  55         int ret;
  56
  57         if (pfn_valid(phys_start_pfn))
  58                 return -EEXIST;
  59
  60         ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
  61
  62         if (ret < 0)
  63                 return ret;
  64
  65         ret = __add_zone(zone, phys_start_pfn);
  66
  67         if (ret < 0)
  68                 return ret;
  69
  70         return register_new_memory(__pfn_to_section(phys_start_pfn));
  71 }
  72
  73 /*
  74  * Reasonably generic function for adding memory.  It is
  75  * expected that archs that support memory hotplug will
  76  * call this function after deciding the zone to which to
  77  * add the new pages.
  78  */
  79 int __add_pages(struct zone *zone, unsigned long phys_start_pfn,
  80                  unsigned long nr_pages)
  81 {
  82         unsigned long i;
  83         int err = 0;
  84         int start_sec, end_sec;
  85         /* during initialize mem_map, align hot-added range to section */
  86         start_sec = pfn_to_section_nr(phys_start_pfn);
  87         end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
  88
  89         for (i = start_sec; i <= end_sec; i++) {
  90                 err = __add_section(zone, i << PFN_SECTION_SHIFT);
  91
  92                 /*
  93                  * EEXIST is finally dealed with by ioresource collision
  94                  * check. see add_memory() => register_memory_resource()
  95                  * Warning will be printed if there is collision.
  96                  */
  97                 if (err && (err != -EEXIST))
  98                         break;
  99                 err = 0;
 100         }
 101
 102         return err;
 103 }
 104 EXPORT_SYMBOL_GPL(__add_pages);
 105
 106 static void grow_zone_span(struct zone *zone,
 107                 unsigned long start_pfn, unsigned long end_pfn)
 108 {
 109         unsigned long old_zone_end_pfn;
 110
 111         zone_span_writelock(zone);
 112
 113         old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
 114         if (start_pfn < zone->zone_start_pfn)
 115                 zone->zone_start_pfn = start_pfn;
 116
 117         zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
 118                                 zone->zone_start_pfn;
 119
 120         zone_span_writeunlock(zone);
 121 }
 122
 123 static void grow_pgdat_span(struct pglist_data *pgdat,
 124                 unsigned long start_pfn, unsigned long end_pfn)
 125 {
 126         unsigned long old_pgdat_end_pfn =
 127                 pgdat->node_start_pfn + pgdat->node_spanned_pages;
 128
 129         if (start_pfn < pgdat->node_start_pfn)
 130                 pgdat->node_start_pfn = start_pfn;
 131
 132         pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
 133                                         pgdat->node_start_pfn;
 134 }
 135
 136 int online_pages(unsigned long pfn, unsigned long nr_pages)
 137 {
 138         unsigned long i;
 139         unsigned long flags;
 140         unsigned long onlined_pages = 0;
 141         struct resource res;
 142         u64 section_end;
 143         unsigned long start_pfn;
 144         struct zone *zone;
 145         int need_zonelists_rebuild = 0;
 146
 147         /*
 148          * This doesn't need a lock to do pfn_to_page().
 149          * The section can't be removed here because of the
 150          * memory_block->state_sem.
 151          */
 152         zone = page_zone(pfn_to_page(pfn));
 153         pgdat_resize_lock(zone->zone_pgdat, &flags);
 154         grow_zone_span(zone, pfn, pfn + nr_pages);
 155         grow_pgdat_span(zone->zone_pgdat, pfn, pfn + nr_pages);
 156         pgdat_resize_unlock(zone->zone_pgdat, &flags);
 157
 158         /*
 159          * If this zone is not populated, then it is not in zonelist.
 160          * This means the page allocator ignores this zone.
 161          * So, zonelist must be updated after online.
 162          */
 163         if (!populated_zone(zone))
 164                 need_zonelists_rebuild = 1;
 165
 166         res.start = (u64)pfn << PAGE_SHIFT;
 167         res.end = res.start + ((u64)nr_pages << PAGE_SHIFT) - 1;
 168         res.flags = IORESOURCE_MEM; /* we just need system ram */
 169         section_end = res.end;
 170
 171         while ((res.start < res.end) && (find_next_system_ram(&res) >= 0)) {
 172                 start_pfn = (unsigned long)(res.start >> PAGE_SHIFT);
 173                 nr_pages = (unsigned long)
 174                            ((res.end + 1 - res.start) >> PAGE_SHIFT);
 175
 176                 if (PageReserved(pfn_to_page(start_pfn))) {
 177                         /* this region's page is not onlined now */
 178                         for (i = 0; i < nr_pages; i++) {
 179                                 struct page *page = pfn_to_page(start_pfn + i);
 180                                 online_page(page);
 181                                 onlined_pages++;
 182                         }
 183                 }
 184
 185                 res.start = res.end + 1;
 186                 res.end = section_end;
 187         }
 188         zone->present_pages += onlined_pages;
 189         zone->zone_pgdat->node_present_pages += onlined_pages;
 190
 191         setup_per_zone_pages_min();
 192
 193         if (need_zonelists_rebuild)
 194                 build_all_zonelists();
 195         vm_total_pages = nr_free_pagecache_pages();
 196         writeback_set_ratelimit();
 197         return 0;
 198 }
 199
 200 static pg_data_t *hotadd_new_pgdat(int nid, u64 start)
 201 {
 202         struct pglist_data *pgdat;
 203         unsigned long zones_size[MAX_NR_ZONES] = {0};
 204         unsigned long zholes_size[MAX_NR_ZONES] = {0};
 205         unsigned long start_pfn = start >> PAGE_SHIFT;
 206
 207         pgdat = arch_alloc_nodedata(nid);
 208         if (!pgdat)
 209                 return NULL;
 210
 211         arch_refresh_nodedata(nid, pgdat);
 212
 213         /* we can use NODE_DATA(nid) from here */
 214
 215         /* init node's zones as empty zones, we don't have any present pages.*/
 216         free_area_init_node(nid, pgdat, zones_size, start_pfn, zholes_size);
 217
 218         return pgdat;
 219 }
 220
 221 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
 222 {
 223         arch_refresh_nodedata(nid, NULL);
 224         arch_free_nodedata(pgdat);
 225         return;
 226 }
 227
 228 /* add this memory to iomem resource */
 229 static struct resource *register_memory_resource(u64 start, u64 size)
 230 {
 231         struct resource *res;
 232         res = kzalloc(sizeof(struct resource), GFP_KERNEL);
 233         BUG_ON(!res);
 234
 235         res->name = "System RAM";
 236         res->start = start;
 237         res->end = start + size - 1;
 238         res->flags = IORESOURCE_MEM;
 239         if (request_resource(&iomem_resource, res) < 0) {
 240                 printk("System RAM resource %llx - %llx cannot be added\n",
 241                 (unsigned long long)res->start, (unsigned long long)res->end);
 242                 kfree(res);
 243                 res = NULL;
 244         }
 245         return res;
 246 }
 247
 248 static void release_memory_resource(struct resource *res)
 249 {
 250         if (!res)
 251                 return;
 252         release_resource(res);
 253         kfree(res);
 254         return;
 255 }
 256
 257
 258
 259 int add_memory(int nid, u64 start, u64 size)
 260 {
 261         pg_data_t *pgdat = NULL;
 262         int new_pgdat = 0;
 263         struct resource *res;
 264         int ret;
 265
 266         res = register_memory_resource(start, size);
 267         if (!res)
 268                 return -EEXIST;
 269
 270         if (!node_online(nid)) {
 271                 pgdat = hotadd_new_pgdat(nid, start);
 272                 if (!pgdat)
 273                         return -ENOMEM;
 274                 new_pgdat = 1;
 275                 ret = kswapd_run(nid);
 276                 if (ret)
 277                         goto error;
 278         }
 279
 280         /* call arch's memory hotadd */
 281         ret = arch_add_memory(nid, start, size);
 282
 283         if (ret < 0)
 284                 goto error;
 285
 286         /* we online node here. we can't roll back from here. */
 287         node_set_online(nid);
 288
 289         cpuset_track_online_nodes();
 290
 291         if (new_pgdat) {
 292                 ret = register_one_node(nid);
 293                 /*
 294                  * If sysfs file of new node can't create, cpu on the node
 295                  * can't be hot-added. There is no rollback way now.
 296                  * So, check by BUG_ON() to catch it reluctantly..
 297                  */
 298                 BUG_ON(ret);
 299         }
 300
 301         return ret;
 302 error:
 303         /* rollback pgdat allocation and others */
 304         if (new_pgdat)
 305                 rollback_node_hotadd(nid, pgdat);
 306         if (res)
 307                 release_memory_resource(res);
 308
 309         return ret;
 310 }
 311 EXPORT_SYMBOL_GPL(add_memory);