drivers/base/memory.c

   1 /*
   2  * drivers/base/memory.c - basic Memory class support
   3  *
   4  * Written by Matt Tolentino <matthew.e.tolentino@intel.com>
   5  *            Dave Hansen <haveblue@us.ibm.com>
   6  *
   7  * This file provides the necessary infrastructure to represent
   8  * a SPARSEMEM-memory-model system's physical memory in /sysfs.
   9  * All arch-independent code that assumes MEMORY_HOTPLUG requires
  10  * SPARSEMEM should be contained here, or in mm/memory_hotplug.c.
  11  */
  12
  13 #include <linux/sysdev.h>
  14 #include <linux/module.h>
  15 #include <linux/init.h>
  16 #include <linux/topology.h>
  17 #include <linux/capability.h>
  18 #include <linux/device.h>
  19 #include <linux/memory.h>
  20 #include <linux/kobject.h>
  21 #include <linux/memory_hotplug.h>
  22 #include <linux/mm.h>
  23 #include <linux/mutex.h>
  24 #include <linux/stat.h>
  25
  26 #include <asm/atomic.h>
  27 #include <asm/uaccess.h>
  28
  29 #define MEMORY_CLASS_NAME       "memory"
  30
  31 static struct sysdev_class memory_sysdev_class = {
  32         .name = MEMORY_CLASS_NAME,
  33 };
  34
  35 static const char *memory_uevent_name(struct kset *kset, struct kobject *kobj)
  36 {
  37         return MEMORY_CLASS_NAME;
  38 }
  39
  40 static int memory_uevent(struct kset *kset, struct kobject *obj, struct kobj_uevent_env *env)
  41 {
  42         int retval = 0;
  43
  44         return retval;
  45 }
  46
  47 static struct kset_uevent_ops memory_uevent_ops = {
  48         .name           = memory_uevent_name,
  49         .uevent         = memory_uevent,
  50 };
  51
  52 static BLOCKING_NOTIFIER_HEAD(memory_chain);
  53
  54 int register_memory_notifier(struct notifier_block *nb)
  55 {
  56         return blocking_notifier_chain_register(&memory_chain, nb);
  57 }
  58 EXPORT_SYMBOL(register_memory_notifier);
  59
  60 void unregister_memory_notifier(struct notifier_block *nb)
  61 {
  62         blocking_notifier_chain_unregister(&memory_chain, nb);
  63 }
  64 EXPORT_SYMBOL(unregister_memory_notifier);
  65
  66 /*
  67  * register_memory - Setup a sysfs device for a memory block
  68  */
  69 static
  70 int register_memory(struct memory_block *memory, struct mem_section *section)
  71 {
  72         int error;
  73
  74         memory->sysdev.cls = &memory_sysdev_class;
  75         memory->sysdev.id = __section_nr(section);
  76
  77         error = sysdev_register(&memory->sysdev);
  78         return error;
  79 }
  80
  81 static void
  82 unregister_memory(struct memory_block *memory, struct mem_section *section)
  83 {
  84         BUG_ON(memory->sysdev.cls != &memory_sysdev_class);
  85         BUG_ON(memory->sysdev.id != __section_nr(section));
  86
  87         /* drop the ref. we got in remove_memory_block() */
  88         kobject_put(&memory->sysdev.kobj);
  89         sysdev_unregister(&memory->sysdev);
  90 }
  91
  92 /*
  93  * use this as the physical section index that this memsection
  94  * uses.
  95  */
  96
  97 static ssize_t show_mem_phys_index(struct sys_device *dev,
  98                         struct sysdev_attribute *attr, char *buf)
  99 {
 100         struct memory_block *mem =
 101                 container_of(dev, struct memory_block, sysdev);
 102         return sprintf(buf, "%08lx\n", mem->phys_index);
 103 }
 104
 105 /*
 106  * Show whether the section of memory is likely to be hot-removable
 107  */
 108 static ssize_t show_mem_removable(struct sys_device *dev,
 109                         struct sysdev_attribute *attr, char *buf)
 110 {
 111         unsigned long start_pfn;
 112         int ret;
 113         struct memory_block *mem =
 114                 container_of(dev, struct memory_block, sysdev);
 115
 116         start_pfn = section_nr_to_pfn(mem->phys_index);
 117         ret = is_mem_section_removable(start_pfn, PAGES_PER_SECTION);
 118         return sprintf(buf, "%d\n", ret);
 119 }
 120
 121 /*
 122  * online, offline, going offline, etc.
 123  */
 124 static ssize_t show_mem_state(struct sys_device *dev,
 125                         struct sysdev_attribute *attr, char *buf)
 126 {
 127         struct memory_block *mem =
 128                 container_of(dev, struct memory_block, sysdev);
 129         ssize_t len = 0;
 130
 131         /*
 132          * We can probably put these states in a nice little array
 133          * so that they're not open-coded
 134          */
 135         switch (mem->state) {
 136                 case MEM_ONLINE:
 137                         len = sprintf(buf, "online\n");
 138                         break;
 139                 case MEM_OFFLINE:
 140                         len = sprintf(buf, "offline\n");
 141                         break;
 142                 case MEM_GOING_OFFLINE:
 143                         len = sprintf(buf, "going-offline\n");
 144                         break;
 145                 default:
 146                         len = sprintf(buf, "ERROR-UNKNOWN-%ld\n",
 147                                         mem->state);
 148                         WARN_ON(1);
 149                         break;
 150         }
 151
 152         return len;
 153 }
 154
 155 int memory_notify(unsigned long val, void *v)
 156 {
 157         return blocking_notifier_call_chain(&memory_chain, val, v);
 158 }
 159
 160 /*
 161  * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is
 162  * OK to have direct references to sparsemem variables in here.
 163  */
 164 static int
 165 memory_block_action(struct memory_block *mem, unsigned long action)
 166 {
 167         int i;
 168         unsigned long psection;
 169         unsigned long start_pfn, start_paddr;
 170         struct page *first_page;
 171         int ret;
 172         int old_state = mem->state;
 173
 174         psection = mem->phys_index;
 175         first_page = pfn_to_page(psection << PFN_SECTION_SHIFT);
 176
 177         /*
 178          * The probe routines leave the pages reserved, just
 179          * as the bootmem code does.  Make sure they're still
 180          * that way.
 181          */
 182         if (action == MEM_ONLINE) {
 183                 for (i = 0; i < PAGES_PER_SECTION; i++) {
 184                         if (PageReserved(first_page+i))
 185                                 continue;
 186
 187                         printk(KERN_WARNING "section number %ld page number %d "
 188                                 "not reserved, was it already online? \n",
 189                                 psection, i);
 190                         return -EBUSY;
 191                 }
 192         }
 193
 194         switch (action) {
 195                 case MEM_ONLINE:
 196                         start_pfn = page_to_pfn(first_page);
 197                         ret = online_pages(start_pfn, PAGES_PER_SECTION);
 198                         break;
 199                 case MEM_OFFLINE:
 200                         mem->state = MEM_GOING_OFFLINE;
 201                         start_paddr = page_to_pfn(first_page) << PAGE_SHIFT;
 202                         ret = remove_memory(start_paddr,
 203                                             PAGES_PER_SECTION << PAGE_SHIFT);
 204                         if (ret) {
 205                                 mem->state = old_state;
 206                                 break;
 207                         }
 208                         break;
 209                 default:
 210                         WARN(1, KERN_WARNING "%s(%p, %ld) unknown action: %ld\n",
 211                                         __func__, mem, action, action);
 212                         ret = -EINVAL;
 213         }
 214
 215         return ret;
 216 }
 217
 218 static int memory_block_change_state(struct memory_block *mem,
 219                 unsigned long to_state, unsigned long from_state_req)
 220 {
 221         int ret = 0;
 222         mutex_lock(&mem->state_mutex);
 223
 224         if (mem->state != from_state_req) {
 225                 ret = -EINVAL;
 226                 goto out;
 227         }
 228
 229         ret = memory_block_action(mem, to_state);
 230         if (!ret)
 231                 mem->state = to_state;
 232
 233 out:
 234         mutex_unlock(&mem->state_mutex);
 235         return ret;
 236 }
 237
 238 static ssize_t
 239 store_mem_state(struct sys_device *dev,
 240                 struct sysdev_attribute *attr, const char *buf, size_t count)
 241 {
 242         struct memory_block *mem;
 243         unsigned int phys_section_nr;
 244         int ret = -EINVAL;
 245
 246         mem = container_of(dev, struct memory_block, sysdev);
 247         phys_section_nr = mem->phys_index;
 248
 249         if (!present_section_nr(phys_section_nr))
 250                 goto out;
 251
 252         if (!strncmp(buf, "online", min((int)count, 6)))
 253                 ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
 254         else if(!strncmp(buf, "offline", min((int)count, 7)))
 255                 ret = memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE);
 256 out:
 257         if (ret)
 258                 return ret;
 259         return count;
 260 }
 261
 262 /*
 263  * phys_device is a bad name for this.  What I really want
 264  * is a way to differentiate between memory ranges that
 265  * are part of physical devices that constitute
 266  * a complete removable unit or fru.
 267  * i.e. do these ranges belong to the same physical device,
 268  * s.t. if I offline all of these sections I can then
 269  * remove the physical device?
 270  */
 271 static ssize_t show_phys_device(struct sys_device *dev,
 272                                 struct sysdev_attribute *attr, char *buf)
 273 {
 274         struct memory_block *mem =
 275                 container_of(dev, struct memory_block, sysdev);
 276         return sprintf(buf, "%d\n", mem->phys_device);
 277 }
 278
 279 static SYSDEV_ATTR(phys_index, 0444, show_mem_phys_index, NULL);
 280 static SYSDEV_ATTR(state, 0644, show_mem_state, store_mem_state);
 281 static SYSDEV_ATTR(phys_device, 0444, show_phys_device, NULL);
 282 static SYSDEV_ATTR(removable, 0444, show_mem_removable, NULL);
 283
 284 #define mem_create_simple_file(mem, attr_name)  \
 285         sysdev_create_file(&mem->sysdev, &attr_##attr_name)
 286 #define mem_remove_simple_file(mem, attr_name)  \
 287         sysdev_remove_file(&mem->sysdev, &attr_##attr_name)
 288
 289 /*
 290  * Block size attribute stuff
 291  */
 292 static ssize_t
 293 print_block_size(struct class *class, char *buf)
 294 {
 295         return sprintf(buf, "%lx\n", (unsigned long)PAGES_PER_SECTION * PAGE_SIZE);
 296 }
 297
 298 static CLASS_ATTR(block_size_bytes, 0444, print_block_size, NULL);
 299
 300 static int block_size_init(void)
 301 {
 302         return sysfs_create_file(&memory_sysdev_class.kset.kobj,
 303                                 &class_attr_block_size_bytes.attr);
 304 }
 305
 306 /*
 307  * Some architectures will have custom drivers to do this, and
 308  * will not need to do it from userspace.  The fake hot-add code
 309  * as well as ppc64 will do all of their discovery in userspace
 310  * and will require this interface.
 311  */
 312 #ifdef CONFIG_ARCH_MEMORY_PROBE
 313 static ssize_t
 314 memory_probe_store(struct class *class, const char *buf, size_t count)
 315 {
 316         u64 phys_addr;
 317         int nid;
 318         int ret;
 319
 320         phys_addr = simple_strtoull(buf, NULL, 0);
 321
 322         nid = memory_add_physaddr_to_nid(phys_addr);
 323         ret = add_memory(nid, phys_addr, PAGES_PER_SECTION << PAGE_SHIFT);
 324
 325         if (ret)
 326                 count = ret;
 327
 328         return count;
 329 }
 330 static CLASS_ATTR(probe, S_IWUSR, NULL, memory_probe_store);
 331
 332 static int memory_probe_init(void)
 333 {
 334         return sysfs_create_file(&memory_sysdev_class.kset.kobj,
 335                                 &class_attr_probe.attr);
 336 }
 337 #else
 338 static inline int memory_probe_init(void)
 339 {
 340         return 0;
 341 }
 342 #endif
 343
 344 /*
 345  * Note that phys_device is optional.  It is here to allow for
 346  * differentiation between which *physical* devices each
 347  * section belongs to...
 348  */
 349
 350 static int add_memory_block(int nid, struct mem_section *section,
 351                         unsigned long state, int phys_device,
 352                         enum mem_add_context context)
 353 {
 354         struct memory_block *mem = kzalloc(sizeof(*mem), GFP_KERNEL);
 355         int ret = 0;
 356
 357         if (!mem)
 358                 return -ENOMEM;
 359
 360         mem->phys_index = __section_nr(section);
 361         mem->state = state;
 362         mutex_init(&mem->state_mutex);
 363         mem->phys_device = phys_device;
 364
 365         ret = register_memory(mem, section);
 366         if (!ret)
 367                 ret = mem_create_simple_file(mem, phys_index);
 368         if (!ret)
 369                 ret = mem_create_simple_file(mem, state);
 370         if (!ret)
 371                 ret = mem_create_simple_file(mem, phys_device);
 372         if (!ret)
 373                 ret = mem_create_simple_file(mem, removable);
 374         if (!ret) {
 375                 if (context == HOTPLUG)
 376                         ret = register_mem_sect_under_node(mem, nid);
 377         }
 378
 379         return ret;
 380 }
 381
 382 /*
 383  * For now, we have a linear search to go find the appropriate
 384  * memory_block corresponding to a particular phys_index. If
 385  * this gets to be a real problem, we can always use a radix
 386  * tree or something here.
 387  *
 388  * This could be made generic for all sysdev classes.
 389  */
 390 struct memory_block *find_memory_block(struct mem_section *section)
 391 {
 392         struct kobject *kobj;
 393         struct sys_device *sysdev;
 394         struct memory_block *mem;
 395         char name[sizeof(MEMORY_CLASS_NAME) + 9 + 1];
 396
 397         /*
 398          * This only works because we know that section == sysdev->id
 399          * slightly redundant with sysdev_register()
 400          */
 401         sprintf(&name[0], "%s%d", MEMORY_CLASS_NAME, __section_nr(section));
 402
 403         kobj = kset_find_obj(&memory_sysdev_class.kset, name);
 404         if (!kobj)
 405                 return NULL;
 406
 407         sysdev = container_of(kobj, struct sys_device, kobj);
 408         mem = container_of(sysdev, struct memory_block, sysdev);
 409
 410         return mem;
 411 }
 412
 413 int remove_memory_block(unsigned long node_id, struct mem_section *section,
 414                 int phys_device)
 415 {
 416         struct memory_block *mem;
 417
 418         mem = find_memory_block(section);
 419         unregister_mem_sect_under_nodes(mem);
 420         mem_remove_simple_file(mem, phys_index);
 421         mem_remove_simple_file(mem, state);
 422         mem_remove_simple_file(mem, phys_device);
 423         mem_remove_simple_file(mem, removable);
 424         unregister_memory(mem, section);
 425
 426         return 0;
 427 }
 428
 429 /*
 430  * need an interface for the VM to add new memory regions,
 431  * but without onlining it.
 432  */
 433 int register_new_memory(int nid, struct mem_section *section)
 434 {
 435         return add_memory_block(nid, section, MEM_OFFLINE, 0, HOTPLUG);
 436 }
 437
 438 int unregister_memory_section(struct mem_section *section)
 439 {
 440         if (!present_section(section))
 441                 return -EINVAL;
 442
 443         return remove_memory_block(0, section, 0);
 444 }
 445
 446 /*
 447  * Initialize the sysfs support for memory devices...
 448  */
 449 int __init memory_dev_init(void)
 450 {
 451         unsigned int i;
 452         int ret;
 453         int err;
 454
 455         memory_sysdev_class.kset.uevent_ops = &memory_uevent_ops;
 456         ret = sysdev_class_register(&memory_sysdev_class);
 457         if (ret)
 458                 goto out;
 459
 460         /*
 461          * Create entries for memory sections that were found
 462          * during boot and have been initialized
 463          */
 464         for (i = 0; i < NR_MEM_SECTIONS; i++) {
 465                 if (!present_section_nr(i))
 466                         continue;
 467                 err = add_memory_block(0, __nr_to_section(i), MEM_ONLINE,
 468                                         0, BOOT);
 469                 if (!ret)
 470                         ret = err;
 471         }
 472
 473         err = memory_probe_init();
 474         if (!ret)
 475                 ret = err;
 476         err = block_size_init();
 477         if (!ret)
 478                 ret = err;
 479 out:
 480         if (ret)
 481                 printk(KERN_ERR "%s() failed: %d\n", __func__, ret);
 482         return ret;
 483 }