mm/page_cgroup.c

   1 #include <linux/mm.h>
   2 #include <linux/mmzone.h>
   3 #include <linux/bootmem.h>
   4 #include <linux/bit_spinlock.h>
   5 #include <linux/page_cgroup.h>
   6 #include <linux/hash.h>
   7 #include <linux/slab.h>
   8 #include <linux/memory.h>
   9 #include <linux/vmalloc.h>
  10 #include <linux/cgroup.h>
  11
  12 static void __meminit
  13 __init_page_cgroup(struct page_cgroup *pc, unsigned long pfn)
  14 {
  15         pc->flags = 0;
  16         pc->mem_cgroup = NULL;
  17         pc->page = pfn_to_page(pfn);
  18 }
  19 static unsigned long total_usage;
  20
  21 #if !defined(CONFIG_SPARSEMEM)
  22
  23
  24 void __meminit pgdat_page_cgroup_init(struct pglist_data *pgdat)
  25 {
  26         pgdat->node_page_cgroup = NULL;
  27 }
  28
  29 struct page_cgroup *lookup_page_cgroup(struct page *page)
  30 {
  31         unsigned long pfn = page_to_pfn(page);
  32         unsigned long offset;
  33         struct page_cgroup *base;
  34
  35         base = NODE_DATA(page_to_nid(page))->node_page_cgroup;
  36         if (unlikely(!base))
  37                 return NULL;
  38
  39         offset = pfn - NODE_DATA(page_to_nid(page))->node_start_pfn;
  40         return base + offset;
  41 }
  42
  43 static int __init alloc_node_page_cgroup(int nid)
  44 {
  45         struct page_cgroup *base, *pc;
  46         unsigned long table_size;
  47         unsigned long start_pfn, nr_pages, index;
  48
  49         start_pfn = NODE_DATA(nid)->node_start_pfn;
  50         nr_pages = NODE_DATA(nid)->node_spanned_pages;
  51
  52         if (!nr_pages)
  53                 return 0;
  54
  55         table_size = sizeof(struct page_cgroup) * nr_pages;
  56
  57         base = __alloc_bootmem_node_nopanic(NODE_DATA(nid),
  58                         table_size, PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
  59         if (!base)
  60                 return -ENOMEM;
  61         for (index = 0; index < nr_pages; index++) {
  62                 pc = base + index;
  63                 __init_page_cgroup(pc, start_pfn + index);
  64         }
  65         NODE_DATA(nid)->node_page_cgroup = base;
  66         total_usage += table_size;
  67         return 0;
  68 }
  69
  70 void __init page_cgroup_init(void)
  71 {
  72
  73         int nid, fail;
  74
  75         if (mem_cgroup_subsys.disabled)
  76                 return;
  77
  78         for_each_online_node(nid)  {
  79                 fail = alloc_node_page_cgroup(nid);
  80                 if (fail)
  81                         goto fail;
  82         }
  83         printk(KERN_INFO "allocated %ld bytes of page_cgroup\n", total_usage);
  84         printk(KERN_INFO "please try cgroup_disable=memory option if you"
  85         " don't want\n");
  86         return;
  87 fail:
  88         printk(KERN_CRIT "allocation of page_cgroup was failed.\n");
  89         printk(KERN_CRIT "please try cgroup_disable=memory boot option\n");
  90         panic("Out of memory");
  91 }
  92
  93 #else /* CONFIG_FLAT_NODE_MEM_MAP */
  94
  95 struct page_cgroup *lookup_page_cgroup(struct page *page)
  96 {
  97         unsigned long pfn = page_to_pfn(page);
  98         struct mem_section *section = __pfn_to_section(pfn);
  99
 100         return section->page_cgroup + pfn;
 101 }
 102
 103 /* __alloc_bootmem...() is protected by !slab_available() */
 104 static int __init_refok init_section_page_cgroup(unsigned long pfn)
 105 {
 106         struct mem_section *section;
 107         struct page_cgroup *base, *pc;
 108         unsigned long table_size;
 109         int nid, index;
 110
 111         section = __pfn_to_section(pfn);
 112
 113         if (!section->page_cgroup) {
 114                 nid = page_to_nid(pfn_to_page(pfn));
 115                 table_size = sizeof(struct page_cgroup) * PAGES_PER_SECTION;
 116                 if (slab_is_available()) {
 117                         base = kmalloc_node(table_size, GFP_KERNEL, nid);
 118                         if (!base)
 119                                 base = vmalloc_node(table_size, nid);
 120                 } else {
 121                         base = __alloc_bootmem_node_nopanic(NODE_DATA(nid),
 122                                 table_size,
 123                                 PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
 124                 }
 125         } else {
 126                 /*
 127                  * We don't have to allocate page_cgroup again, but
 128                  * address of memmap may be changed. So, we have to initialize
 129                  * again.
 130                  */
 131                 base = section->page_cgroup + pfn;
 132                 table_size = 0;
 133                 /* check address of memmap is changed or not. */
 134                 if (base->page == pfn_to_page(pfn))
 135                         return 0;
 136         }
 137
 138         if (!base) {
 139                 printk(KERN_ERR "page cgroup allocation failure\n");
 140                 return -ENOMEM;
 141         }
 142
 143         for (index = 0; index < PAGES_PER_SECTION; index++) {
 144                 pc = base + index;
 145                 __init_page_cgroup(pc, pfn + index);
 146         }
 147
 148         section = __pfn_to_section(pfn);
 149         section->page_cgroup = base - pfn;
 150         total_usage += table_size;
 151         return 0;
 152 }
 153 #ifdef CONFIG_MEMORY_HOTPLUG
 154 void __free_page_cgroup(unsigned long pfn)
 155 {
 156         struct mem_section *ms;
 157         struct page_cgroup *base;
 158
 159         ms = __pfn_to_section(pfn);
 160         if (!ms || !ms->page_cgroup)
 161                 return;
 162         base = ms->page_cgroup + pfn;
 163         if (is_vmalloc_addr(base)) {
 164                 vfree(base);
 165                 ms->page_cgroup = NULL;
 166         } else {
 167                 struct page *page = virt_to_page(base);
 168                 if (!PageReserved(page)) { /* Is bootmem ? */
 169                         kfree(base);
 170                         ms->page_cgroup = NULL;
 171                 }
 172         }
 173 }
 174
 175 int __meminit online_page_cgroup(unsigned long start_pfn,
 176                         unsigned long nr_pages,
 177                         int nid)
 178 {
 179         unsigned long start, end, pfn;
 180         int fail = 0;
 181
 182         start = start_pfn & ~(PAGES_PER_SECTION - 1);
 183         end = ALIGN(start_pfn + nr_pages, PAGES_PER_SECTION);
 184
 185         for (pfn = start; !fail && pfn < end; pfn += PAGES_PER_SECTION) {
 186                 if (!pfn_present(pfn))
 187                         continue;
 188                 fail = init_section_page_cgroup(pfn);
 189         }
 190         if (!fail)
 191                 return 0;
 192
 193         /* rollback */
 194         for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION)
 195                 __free_page_cgroup(pfn);
 196
 197         return -ENOMEM;
 198 }
 199
 200 int __meminit offline_page_cgroup(unsigned long start_pfn,
 201                 unsigned long nr_pages, int nid)
 202 {
 203         unsigned long start, end, pfn;
 204
 205         start = start_pfn & ~(PAGES_PER_SECTION - 1);
 206         end = ALIGN(start_pfn + nr_pages, PAGES_PER_SECTION);
 207
 208         for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION)
 209                 __free_page_cgroup(pfn);
 210         return 0;
 211
 212 }
 213
 214 static int __meminit page_cgroup_callback(struct notifier_block *self,
 215                                unsigned long action, void *arg)
 216 {
 217         struct memory_notify *mn = arg;
 218         int ret = 0;
 219         switch (action) {
 220         case MEM_GOING_ONLINE:
 221                 ret = online_page_cgroup(mn->start_pfn,
 222                                    mn->nr_pages, mn->status_change_nid);
 223                 break;
 224         case MEM_OFFLINE:
 225                 offline_page_cgroup(mn->start_pfn,
 226                                 mn->nr_pages, mn->status_change_nid);
 227                 break;
 228         case MEM_CANCEL_ONLINE:
 229         case MEM_GOING_OFFLINE:
 230                 break;
 231         case MEM_ONLINE:
 232         case MEM_CANCEL_OFFLINE:
 233                 break;
 234         }
 235
 236         if (ret)
 237                 ret = notifier_from_errno(ret);
 238         else
 239                 ret = NOTIFY_OK;
 240
 241         return ret;
 242 }
 243
 244 #endif
 245
 246 void __init page_cgroup_init(void)
 247 {
 248         unsigned long pfn;
 249         int fail = 0;
 250
 251         if (mem_cgroup_subsys.disabled)
 252                 return;
 253
 254         for (pfn = 0; !fail && pfn < max_pfn; pfn += PAGES_PER_SECTION) {
 255                 if (!pfn_present(pfn))
 256                         continue;
 257                 fail = init_section_page_cgroup(pfn);
 258         }
 259         if (fail) {
 260                 printk(KERN_CRIT "try cgroup_disable=memory boot option\n");
 261                 panic("Out of memory");
 262         } else {
 263                 hotplug_memory_notifier(page_cgroup_callback, 0);
 264         }
 265         printk(KERN_INFO "allocated %ld bytes of page_cgroup\n", total_usage);
 266         printk(KERN_INFO "please try cgroup_disable=memory option if you don't"
 267         " want\n");
 268 }
 269
 270 void __meminit pgdat_page_cgroup_init(struct pglist_data *pgdat)
 271 {
 272         return;
 273 }
 274
 275 #endif