Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame^] | 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * Basic Node interface support |
| 4 | */ |
| 5 | |
| 6 | #include <linux/module.h> |
| 7 | #include <linux/init.h> |
| 8 | #include <linux/mm.h> |
| 9 | #include <linux/memory.h> |
| 10 | #include <linux/vmstat.h> |
| 11 | #include <linux/notifier.h> |
| 12 | #include <linux/node.h> |
| 13 | #include <linux/hugetlb.h> |
| 14 | #include <linux/compaction.h> |
| 15 | #include <linux/cpumask.h> |
| 16 | #include <linux/topology.h> |
| 17 | #include <linux/nodemask.h> |
| 18 | #include <linux/cpu.h> |
| 19 | #include <linux/device.h> |
| 20 | #include <linux/swap.h> |
| 21 | #include <linux/slab.h> |
| 22 | |
| 23 | static struct bus_type node_subsys = { |
| 24 | .name = "node", |
| 25 | .dev_name = "node", |
| 26 | }; |
| 27 | |
| 28 | |
| 29 | static ssize_t node_read_cpumap(struct device *dev, bool list, char *buf) |
| 30 | { |
| 31 | ssize_t n; |
| 32 | cpumask_var_t mask; |
| 33 | struct node *node_dev = to_node(dev); |
| 34 | |
| 35 | /* 2008/04/07: buf currently PAGE_SIZE, need 9 chars per 32 bits. */ |
| 36 | BUILD_BUG_ON((NR_CPUS/32 * 9) > (PAGE_SIZE-1)); |
| 37 | |
| 38 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
| 39 | return 0; |
| 40 | |
| 41 | cpumask_and(mask, cpumask_of_node(node_dev->dev.id), cpu_online_mask); |
| 42 | n = cpumap_print_to_pagebuf(list, buf, mask); |
| 43 | free_cpumask_var(mask); |
| 44 | |
| 45 | return n; |
| 46 | } |
| 47 | |
| 48 | static inline ssize_t node_read_cpumask(struct device *dev, |
| 49 | struct device_attribute *attr, char *buf) |
| 50 | { |
| 51 | return node_read_cpumap(dev, false, buf); |
| 52 | } |
| 53 | static inline ssize_t node_read_cpulist(struct device *dev, |
| 54 | struct device_attribute *attr, char *buf) |
| 55 | { |
| 56 | return node_read_cpumap(dev, true, buf); |
| 57 | } |
| 58 | |
| 59 | static DEVICE_ATTR(cpumap, S_IRUGO, node_read_cpumask, NULL); |
| 60 | static DEVICE_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL); |
| 61 | |
| 62 | #define K(x) ((x) << (PAGE_SHIFT - 10)) |
| 63 | static ssize_t node_read_meminfo(struct device *dev, |
| 64 | struct device_attribute *attr, char *buf) |
| 65 | { |
| 66 | int n; |
| 67 | int nid = dev->id; |
| 68 | struct pglist_data *pgdat = NODE_DATA(nid); |
| 69 | struct sysinfo i; |
| 70 | |
| 71 | si_meminfo_node(&i, nid); |
| 72 | n = sprintf(buf, |
| 73 | "Node %d MemTotal: %8lu kB\n" |
| 74 | "Node %d MemFree: %8lu kB\n" |
| 75 | "Node %d MemUsed: %8lu kB\n" |
| 76 | "Node %d Active: %8lu kB\n" |
| 77 | "Node %d Inactive: %8lu kB\n" |
| 78 | "Node %d Active(anon): %8lu kB\n" |
| 79 | "Node %d Inactive(anon): %8lu kB\n" |
| 80 | "Node %d Active(file): %8lu kB\n" |
| 81 | "Node %d Inactive(file): %8lu kB\n" |
| 82 | "Node %d Unevictable: %8lu kB\n" |
| 83 | "Node %d Mlocked: %8lu kB\n", |
| 84 | nid, K(i.totalram), |
| 85 | nid, K(i.freeram), |
| 86 | nid, K(i.totalram - i.freeram), |
| 87 | nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) + |
| 88 | node_page_state(pgdat, NR_ACTIVE_FILE)), |
| 89 | nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) + |
| 90 | node_page_state(pgdat, NR_INACTIVE_FILE)), |
| 91 | nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)), |
| 92 | nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)), |
| 93 | nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)), |
| 94 | nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)), |
| 95 | nid, K(node_page_state(pgdat, NR_UNEVICTABLE)), |
| 96 | nid, K(sum_zone_node_page_state(nid, NR_MLOCK))); |
| 97 | |
| 98 | #ifdef CONFIG_HIGHMEM |
| 99 | n += sprintf(buf + n, |
| 100 | "Node %d HighTotal: %8lu kB\n" |
| 101 | "Node %d HighFree: %8lu kB\n" |
| 102 | "Node %d LowTotal: %8lu kB\n" |
| 103 | "Node %d LowFree: %8lu kB\n", |
| 104 | nid, K(i.totalhigh), |
| 105 | nid, K(i.freehigh), |
| 106 | nid, K(i.totalram - i.totalhigh), |
| 107 | nid, K(i.freeram - i.freehigh)); |
| 108 | #endif |
| 109 | n += sprintf(buf + n, |
| 110 | "Node %d Dirty: %8lu kB\n" |
| 111 | "Node %d Writeback: %8lu kB\n" |
| 112 | "Node %d FilePages: %8lu kB\n" |
| 113 | "Node %d Mapped: %8lu kB\n" |
| 114 | "Node %d AnonPages: %8lu kB\n" |
| 115 | "Node %d Shmem: %8lu kB\n" |
| 116 | "Node %d KernelStack: %8lu kB\n" |
| 117 | "Node %d PageTables: %8lu kB\n" |
| 118 | "Node %d NFS_Unstable: %8lu kB\n" |
| 119 | "Node %d Bounce: %8lu kB\n" |
| 120 | "Node %d WritebackTmp: %8lu kB\n" |
| 121 | "Node %d Slab: %8lu kB\n" |
| 122 | "Node %d SReclaimable: %8lu kB\n" |
| 123 | "Node %d SUnreclaim: %8lu kB\n" |
| 124 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 125 | "Node %d AnonHugePages: %8lu kB\n" |
| 126 | "Node %d ShmemHugePages: %8lu kB\n" |
| 127 | "Node %d ShmemPmdMapped: %8lu kB\n" |
| 128 | #endif |
| 129 | , |
| 130 | nid, K(node_page_state(pgdat, NR_FILE_DIRTY)), |
| 131 | nid, K(node_page_state(pgdat, NR_WRITEBACK)), |
| 132 | nid, K(node_page_state(pgdat, NR_FILE_PAGES)), |
| 133 | nid, K(node_page_state(pgdat, NR_FILE_MAPPED)), |
| 134 | nid, K(node_page_state(pgdat, NR_ANON_MAPPED)), |
| 135 | nid, K(i.sharedram), |
| 136 | nid, sum_zone_node_page_state(nid, NR_KERNEL_STACK_KB), |
| 137 | nid, K(sum_zone_node_page_state(nid, NR_PAGETABLE)), |
| 138 | nid, K(node_page_state(pgdat, NR_UNSTABLE_NFS)), |
| 139 | nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)), |
| 140 | nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)), |
| 141 | nid, K(node_page_state(pgdat, NR_SLAB_RECLAIMABLE) + |
| 142 | node_page_state(pgdat, NR_SLAB_UNRECLAIMABLE)), |
| 143 | nid, K(node_page_state(pgdat, NR_SLAB_RECLAIMABLE)), |
| 144 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 145 | nid, K(node_page_state(pgdat, NR_SLAB_UNRECLAIMABLE)), |
| 146 | nid, K(node_page_state(pgdat, NR_ANON_THPS) * |
| 147 | HPAGE_PMD_NR), |
| 148 | nid, K(node_page_state(pgdat, NR_SHMEM_THPS) * |
| 149 | HPAGE_PMD_NR), |
| 150 | nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) * |
| 151 | HPAGE_PMD_NR)); |
| 152 | #else |
| 153 | nid, K(node_page_state(pgdat, NR_SLAB_UNRECLAIMABLE))); |
| 154 | #endif |
| 155 | n += hugetlb_report_node_meminfo(nid, buf + n); |
| 156 | return n; |
| 157 | } |
| 158 | |
| 159 | #undef K |
| 160 | static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL); |
| 161 | |
| 162 | static ssize_t node_read_numastat(struct device *dev, |
| 163 | struct device_attribute *attr, char *buf) |
| 164 | { |
| 165 | return sprintf(buf, |
| 166 | "numa_hit %lu\n" |
| 167 | "numa_miss %lu\n" |
| 168 | "numa_foreign %lu\n" |
| 169 | "interleave_hit %lu\n" |
| 170 | "local_node %lu\n" |
| 171 | "other_node %lu\n", |
| 172 | sum_zone_numa_state(dev->id, NUMA_HIT), |
| 173 | sum_zone_numa_state(dev->id, NUMA_MISS), |
| 174 | sum_zone_numa_state(dev->id, NUMA_FOREIGN), |
| 175 | sum_zone_numa_state(dev->id, NUMA_INTERLEAVE_HIT), |
| 176 | sum_zone_numa_state(dev->id, NUMA_LOCAL), |
| 177 | sum_zone_numa_state(dev->id, NUMA_OTHER)); |
| 178 | } |
| 179 | static DEVICE_ATTR(numastat, S_IRUGO, node_read_numastat, NULL); |
| 180 | |
| 181 | static ssize_t node_read_vmstat(struct device *dev, |
| 182 | struct device_attribute *attr, char *buf) |
| 183 | { |
| 184 | int nid = dev->id; |
| 185 | struct pglist_data *pgdat = NODE_DATA(nid); |
| 186 | int i; |
| 187 | int n = 0; |
| 188 | |
| 189 | for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) |
| 190 | n += sprintf(buf+n, "%s %lu\n", vmstat_text[i], |
| 191 | sum_zone_node_page_state(nid, i)); |
| 192 | |
| 193 | #ifdef CONFIG_NUMA |
| 194 | for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++) |
| 195 | n += sprintf(buf+n, "%s %lu\n", |
| 196 | vmstat_text[i + NR_VM_ZONE_STAT_ITEMS], |
| 197 | sum_zone_numa_state(nid, i)); |
| 198 | #endif |
| 199 | |
| 200 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
| 201 | n += sprintf(buf+n, "%s %lu\n", |
| 202 | vmstat_text[i + NR_VM_ZONE_STAT_ITEMS + |
| 203 | NR_VM_NUMA_STAT_ITEMS], |
| 204 | node_page_state(pgdat, i)); |
| 205 | |
| 206 | return n; |
| 207 | } |
| 208 | static DEVICE_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL); |
| 209 | |
| 210 | static ssize_t node_read_distance(struct device *dev, |
| 211 | struct device_attribute *attr, char *buf) |
| 212 | { |
| 213 | int nid = dev->id; |
| 214 | int len = 0; |
| 215 | int i; |
| 216 | |
| 217 | /* |
| 218 | * buf is currently PAGE_SIZE in length and each node needs 4 chars |
| 219 | * at the most (distance + space or newline). |
| 220 | */ |
| 221 | BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE); |
| 222 | |
| 223 | for_each_online_node(i) |
| 224 | len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i)); |
| 225 | |
| 226 | len += sprintf(buf + len, "\n"); |
| 227 | return len; |
| 228 | } |
| 229 | static DEVICE_ATTR(distance, S_IRUGO, node_read_distance, NULL); |
| 230 | |
| 231 | static struct attribute *node_dev_attrs[] = { |
| 232 | &dev_attr_cpumap.attr, |
| 233 | &dev_attr_cpulist.attr, |
| 234 | &dev_attr_meminfo.attr, |
| 235 | &dev_attr_numastat.attr, |
| 236 | &dev_attr_distance.attr, |
| 237 | &dev_attr_vmstat.attr, |
| 238 | NULL |
| 239 | }; |
| 240 | ATTRIBUTE_GROUPS(node_dev); |
| 241 | |
| 242 | #ifdef CONFIG_HUGETLBFS |
| 243 | /* |
| 244 | * hugetlbfs per node attributes registration interface: |
| 245 | * When/if hugetlb[fs] subsystem initializes [sometime after this module], |
| 246 | * it will register its per node attributes for all online nodes with |
| 247 | * memory. It will also call register_hugetlbfs_with_node(), below, to |
| 248 | * register its attribute registration functions with this node driver. |
| 249 | * Once these hooks have been initialized, the node driver will call into |
| 250 | * the hugetlb module to [un]register attributes for hot-plugged nodes. |
| 251 | */ |
| 252 | static node_registration_func_t __hugetlb_register_node; |
| 253 | static node_registration_func_t __hugetlb_unregister_node; |
| 254 | |
| 255 | static inline bool hugetlb_register_node(struct node *node) |
| 256 | { |
| 257 | if (__hugetlb_register_node && |
| 258 | node_state(node->dev.id, N_MEMORY)) { |
| 259 | __hugetlb_register_node(node); |
| 260 | return true; |
| 261 | } |
| 262 | return false; |
| 263 | } |
| 264 | |
| 265 | static inline void hugetlb_unregister_node(struct node *node) |
| 266 | { |
| 267 | if (__hugetlb_unregister_node) |
| 268 | __hugetlb_unregister_node(node); |
| 269 | } |
| 270 | |
| 271 | void register_hugetlbfs_with_node(node_registration_func_t doregister, |
| 272 | node_registration_func_t unregister) |
| 273 | { |
| 274 | __hugetlb_register_node = doregister; |
| 275 | __hugetlb_unregister_node = unregister; |
| 276 | } |
| 277 | #else |
| 278 | static inline void hugetlb_register_node(struct node *node) {} |
| 279 | |
| 280 | static inline void hugetlb_unregister_node(struct node *node) {} |
| 281 | #endif |
| 282 | |
| 283 | static void node_device_release(struct device *dev) |
| 284 | { |
| 285 | struct node *node = to_node(dev); |
| 286 | |
| 287 | #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS) |
| 288 | /* |
| 289 | * We schedule the work only when a memory section is |
| 290 | * onlined/offlined on this node. When we come here, |
| 291 | * all the memory on this node has been offlined, |
| 292 | * so we won't enqueue new work to this work. |
| 293 | * |
| 294 | * The work is using node->node_work, so we should |
| 295 | * flush work before freeing the memory. |
| 296 | */ |
| 297 | flush_work(&node->node_work); |
| 298 | #endif |
| 299 | kfree(node); |
| 300 | } |
| 301 | |
| 302 | /* |
| 303 | * register_node - Setup a sysfs device for a node. |
| 304 | * @num - Node number to use when creating the device. |
| 305 | * |
| 306 | * Initialize and register the node device. |
| 307 | */ |
| 308 | static int register_node(struct node *node, int num) |
| 309 | { |
| 310 | int error; |
| 311 | |
| 312 | node->dev.id = num; |
| 313 | node->dev.bus = &node_subsys; |
| 314 | node->dev.release = node_device_release; |
| 315 | node->dev.groups = node_dev_groups; |
| 316 | error = device_register(&node->dev); |
| 317 | |
| 318 | if (error) |
| 319 | put_device(&node->dev); |
| 320 | else { |
| 321 | hugetlb_register_node(node); |
| 322 | |
| 323 | compaction_register_node(node); |
| 324 | } |
| 325 | return error; |
| 326 | } |
| 327 | |
| 328 | /** |
| 329 | * unregister_node - unregister a node device |
| 330 | * @node: node going away |
| 331 | * |
| 332 | * Unregisters a node device @node. All the devices on the node must be |
| 333 | * unregistered before calling this function. |
| 334 | */ |
| 335 | void unregister_node(struct node *node) |
| 336 | { |
| 337 | hugetlb_unregister_node(node); /* no-op, if memoryless node */ |
| 338 | |
| 339 | device_unregister(&node->dev); |
| 340 | } |
| 341 | |
| 342 | struct node *node_devices[MAX_NUMNODES]; |
| 343 | |
| 344 | /* |
| 345 | * register cpu under node |
| 346 | */ |
| 347 | int register_cpu_under_node(unsigned int cpu, unsigned int nid) |
| 348 | { |
| 349 | int ret; |
| 350 | struct device *obj; |
| 351 | |
| 352 | if (!node_online(nid)) |
| 353 | return 0; |
| 354 | |
| 355 | obj = get_cpu_device(cpu); |
| 356 | if (!obj) |
| 357 | return 0; |
| 358 | |
| 359 | ret = sysfs_create_link(&node_devices[nid]->dev.kobj, |
| 360 | &obj->kobj, |
| 361 | kobject_name(&obj->kobj)); |
| 362 | if (ret) |
| 363 | return ret; |
| 364 | |
| 365 | return sysfs_create_link(&obj->kobj, |
| 366 | &node_devices[nid]->dev.kobj, |
| 367 | kobject_name(&node_devices[nid]->dev.kobj)); |
| 368 | } |
| 369 | |
| 370 | int unregister_cpu_under_node(unsigned int cpu, unsigned int nid) |
| 371 | { |
| 372 | struct device *obj; |
| 373 | |
| 374 | if (!node_online(nid)) |
| 375 | return 0; |
| 376 | |
| 377 | obj = get_cpu_device(cpu); |
| 378 | if (!obj) |
| 379 | return 0; |
| 380 | |
| 381 | sysfs_remove_link(&node_devices[nid]->dev.kobj, |
| 382 | kobject_name(&obj->kobj)); |
| 383 | sysfs_remove_link(&obj->kobj, |
| 384 | kobject_name(&node_devices[nid]->dev.kobj)); |
| 385 | |
| 386 | return 0; |
| 387 | } |
| 388 | |
| 389 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
| 390 | static int __ref get_nid_for_pfn(unsigned long pfn) |
| 391 | { |
| 392 | if (!pfn_valid_within(pfn)) |
| 393 | return -1; |
| 394 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
| 395 | if (system_state < SYSTEM_RUNNING) |
| 396 | return early_pfn_to_nid(pfn); |
| 397 | #endif |
| 398 | return pfn_to_nid(pfn); |
| 399 | } |
| 400 | |
| 401 | /* register memory section under specified node if it spans that node */ |
| 402 | int register_mem_sect_under_node(struct memory_block *mem_blk, void *arg) |
| 403 | { |
| 404 | int ret, nid = *(int *)arg; |
| 405 | unsigned long pfn, sect_start_pfn, sect_end_pfn; |
| 406 | |
| 407 | mem_blk->nid = nid; |
| 408 | |
| 409 | sect_start_pfn = section_nr_to_pfn(mem_blk->start_section_nr); |
| 410 | sect_end_pfn = section_nr_to_pfn(mem_blk->end_section_nr); |
| 411 | sect_end_pfn += PAGES_PER_SECTION - 1; |
| 412 | for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) { |
| 413 | int page_nid; |
| 414 | |
| 415 | /* |
| 416 | * memory block could have several absent sections from start. |
| 417 | * skip pfn range from absent section |
| 418 | */ |
| 419 | if (!pfn_present(pfn)) { |
| 420 | pfn = round_down(pfn + PAGES_PER_SECTION, |
| 421 | PAGES_PER_SECTION) - 1; |
| 422 | continue; |
| 423 | } |
| 424 | |
| 425 | /* |
| 426 | * We need to check if page belongs to nid only for the boot |
| 427 | * case, during hotplug we know that all pages in the memory |
| 428 | * block belong to the same node. |
| 429 | */ |
| 430 | if (system_state == SYSTEM_BOOTING) { |
| 431 | page_nid = get_nid_for_pfn(pfn); |
| 432 | if (page_nid < 0) |
| 433 | continue; |
| 434 | if (page_nid != nid) |
| 435 | continue; |
| 436 | } |
| 437 | ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj, |
| 438 | &mem_blk->dev.kobj, |
| 439 | kobject_name(&mem_blk->dev.kobj)); |
| 440 | if (ret) |
| 441 | return ret; |
| 442 | |
| 443 | return sysfs_create_link_nowarn(&mem_blk->dev.kobj, |
| 444 | &node_devices[nid]->dev.kobj, |
| 445 | kobject_name(&node_devices[nid]->dev.kobj)); |
| 446 | } |
| 447 | /* mem section does not span the specified node */ |
| 448 | return 0; |
| 449 | } |
| 450 | |
| 451 | /* unregister memory section under all nodes that it spans */ |
| 452 | int unregister_mem_sect_under_nodes(struct memory_block *mem_blk, |
| 453 | unsigned long phys_index) |
| 454 | { |
| 455 | NODEMASK_ALLOC(nodemask_t, unlinked_nodes, GFP_KERNEL); |
| 456 | unsigned long pfn, sect_start_pfn, sect_end_pfn; |
| 457 | |
| 458 | if (!mem_blk) { |
| 459 | NODEMASK_FREE(unlinked_nodes); |
| 460 | return -EFAULT; |
| 461 | } |
| 462 | if (!unlinked_nodes) |
| 463 | return -ENOMEM; |
| 464 | nodes_clear(*unlinked_nodes); |
| 465 | |
| 466 | sect_start_pfn = section_nr_to_pfn(phys_index); |
| 467 | sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1; |
| 468 | for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) { |
| 469 | int nid; |
| 470 | |
| 471 | nid = get_nid_for_pfn(pfn); |
| 472 | if (nid < 0) |
| 473 | continue; |
| 474 | if (!node_online(nid)) |
| 475 | continue; |
| 476 | if (node_test_and_set(nid, *unlinked_nodes)) |
| 477 | continue; |
| 478 | sysfs_remove_link(&node_devices[nid]->dev.kobj, |
| 479 | kobject_name(&mem_blk->dev.kobj)); |
| 480 | sysfs_remove_link(&mem_blk->dev.kobj, |
| 481 | kobject_name(&node_devices[nid]->dev.kobj)); |
| 482 | } |
| 483 | NODEMASK_FREE(unlinked_nodes); |
| 484 | return 0; |
| 485 | } |
| 486 | |
| 487 | int link_mem_sections(int nid, unsigned long start_pfn, unsigned long end_pfn) |
| 488 | { |
| 489 | return walk_memory_range(start_pfn, end_pfn, (void *)&nid, |
| 490 | register_mem_sect_under_node); |
| 491 | } |
| 492 | |
| 493 | #ifdef CONFIG_HUGETLBFS |
| 494 | /* |
| 495 | * Handle per node hstate attribute [un]registration on transistions |
| 496 | * to/from memoryless state. |
| 497 | */ |
| 498 | static void node_hugetlb_work(struct work_struct *work) |
| 499 | { |
| 500 | struct node *node = container_of(work, struct node, node_work); |
| 501 | |
| 502 | /* |
| 503 | * We only get here when a node transitions to/from memoryless state. |
| 504 | * We can detect which transition occurred by examining whether the |
| 505 | * node has memory now. hugetlb_register_node() already check this |
| 506 | * so we try to register the attributes. If that fails, then the |
| 507 | * node has transitioned to memoryless, try to unregister the |
| 508 | * attributes. |
| 509 | */ |
| 510 | if (!hugetlb_register_node(node)) |
| 511 | hugetlb_unregister_node(node); |
| 512 | } |
| 513 | |
| 514 | static void init_node_hugetlb_work(int nid) |
| 515 | { |
| 516 | INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work); |
| 517 | } |
| 518 | |
| 519 | static int node_memory_callback(struct notifier_block *self, |
| 520 | unsigned long action, void *arg) |
| 521 | { |
| 522 | struct memory_notify *mnb = arg; |
| 523 | int nid = mnb->status_change_nid; |
| 524 | |
| 525 | switch (action) { |
| 526 | case MEM_ONLINE: |
| 527 | case MEM_OFFLINE: |
| 528 | /* |
| 529 | * offload per node hstate [un]registration to a work thread |
| 530 | * when transitioning to/from memoryless state. |
| 531 | */ |
| 532 | if (nid != NUMA_NO_NODE) |
| 533 | schedule_work(&node_devices[nid]->node_work); |
| 534 | break; |
| 535 | |
| 536 | case MEM_GOING_ONLINE: |
| 537 | case MEM_GOING_OFFLINE: |
| 538 | case MEM_CANCEL_ONLINE: |
| 539 | case MEM_CANCEL_OFFLINE: |
| 540 | default: |
| 541 | break; |
| 542 | } |
| 543 | |
| 544 | return NOTIFY_OK; |
| 545 | } |
| 546 | #endif /* CONFIG_HUGETLBFS */ |
| 547 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
| 548 | |
| 549 | #if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \ |
| 550 | !defined(CONFIG_HUGETLBFS) |
| 551 | static inline int node_memory_callback(struct notifier_block *self, |
| 552 | unsigned long action, void *arg) |
| 553 | { |
| 554 | return NOTIFY_OK; |
| 555 | } |
| 556 | |
| 557 | static void init_node_hugetlb_work(int nid) { } |
| 558 | |
| 559 | #endif |
| 560 | |
| 561 | int __register_one_node(int nid) |
| 562 | { |
| 563 | int error; |
| 564 | int cpu; |
| 565 | |
| 566 | node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL); |
| 567 | if (!node_devices[nid]) |
| 568 | return -ENOMEM; |
| 569 | |
| 570 | error = register_node(node_devices[nid], nid); |
| 571 | |
| 572 | /* link cpu under this node */ |
| 573 | for_each_present_cpu(cpu) { |
| 574 | if (cpu_to_node(cpu) == nid) |
| 575 | register_cpu_under_node(cpu, nid); |
| 576 | } |
| 577 | |
| 578 | /* initialize work queue for memory hot plug */ |
| 579 | init_node_hugetlb_work(nid); |
| 580 | |
| 581 | return error; |
| 582 | } |
| 583 | |
| 584 | void unregister_one_node(int nid) |
| 585 | { |
| 586 | if (!node_devices[nid]) |
| 587 | return; |
| 588 | |
| 589 | unregister_node(node_devices[nid]); |
| 590 | node_devices[nid] = NULL; |
| 591 | } |
| 592 | |
| 593 | /* |
| 594 | * node states attributes |
| 595 | */ |
| 596 | |
| 597 | static ssize_t print_nodes_state(enum node_states state, char *buf) |
| 598 | { |
| 599 | int n; |
| 600 | |
| 601 | n = scnprintf(buf, PAGE_SIZE - 1, "%*pbl", |
| 602 | nodemask_pr_args(&node_states[state])); |
| 603 | buf[n++] = '\n'; |
| 604 | buf[n] = '\0'; |
| 605 | return n; |
| 606 | } |
| 607 | |
| 608 | struct node_attr { |
| 609 | struct device_attribute attr; |
| 610 | enum node_states state; |
| 611 | }; |
| 612 | |
| 613 | static ssize_t show_node_state(struct device *dev, |
| 614 | struct device_attribute *attr, char *buf) |
| 615 | { |
| 616 | struct node_attr *na = container_of(attr, struct node_attr, attr); |
| 617 | return print_nodes_state(na->state, buf); |
| 618 | } |
| 619 | |
| 620 | #define _NODE_ATTR(name, state) \ |
| 621 | { __ATTR(name, 0444, show_node_state, NULL), state } |
| 622 | |
| 623 | static struct node_attr node_state_attr[] = { |
| 624 | [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE), |
| 625 | [N_ONLINE] = _NODE_ATTR(online, N_ONLINE), |
| 626 | [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY), |
| 627 | #ifdef CONFIG_HIGHMEM |
| 628 | [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY), |
| 629 | #endif |
| 630 | [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY), |
| 631 | [N_CPU] = _NODE_ATTR(has_cpu, N_CPU), |
| 632 | }; |
| 633 | |
| 634 | static struct attribute *node_state_attrs[] = { |
| 635 | &node_state_attr[N_POSSIBLE].attr.attr, |
| 636 | &node_state_attr[N_ONLINE].attr.attr, |
| 637 | &node_state_attr[N_NORMAL_MEMORY].attr.attr, |
| 638 | #ifdef CONFIG_HIGHMEM |
| 639 | &node_state_attr[N_HIGH_MEMORY].attr.attr, |
| 640 | #endif |
| 641 | &node_state_attr[N_MEMORY].attr.attr, |
| 642 | &node_state_attr[N_CPU].attr.attr, |
| 643 | NULL |
| 644 | }; |
| 645 | |
| 646 | static struct attribute_group memory_root_attr_group = { |
| 647 | .attrs = node_state_attrs, |
| 648 | }; |
| 649 | |
| 650 | static const struct attribute_group *cpu_root_attr_groups[] = { |
| 651 | &memory_root_attr_group, |
| 652 | NULL, |
| 653 | }; |
| 654 | |
| 655 | #define NODE_CALLBACK_PRI 2 /* lower than SLAB */ |
| 656 | static int __init register_node_type(void) |
| 657 | { |
| 658 | int ret; |
| 659 | |
| 660 | BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES); |
| 661 | BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES); |
| 662 | |
| 663 | ret = subsys_system_register(&node_subsys, cpu_root_attr_groups); |
| 664 | if (!ret) { |
| 665 | static struct notifier_block node_memory_callback_nb = { |
| 666 | .notifier_call = node_memory_callback, |
| 667 | .priority = NODE_CALLBACK_PRI, |
| 668 | }; |
| 669 | register_hotmemory_notifier(&node_memory_callback_nb); |
| 670 | } |
| 671 | |
| 672 | /* |
| 673 | * Note: we're not going to unregister the node class if we fail |
| 674 | * to register the node state class attribute files. |
| 675 | */ |
| 676 | return ret; |
| 677 | } |
| 678 | postcore_initcall(register_node_type); |