]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
10fbcf4c | 2 | * Basic Node interface support |
1da177e4 LT |
3 | */ |
4 | ||
1da177e4 LT |
5 | #include <linux/module.h> |
6 | #include <linux/init.h> | |
7 | #include <linux/mm.h> | |
c04fc586 | 8 | #include <linux/memory.h> |
fa25c503 | 9 | #include <linux/vmstat.h> |
6e259e7d | 10 | #include <linux/notifier.h> |
1da177e4 LT |
11 | #include <linux/node.h> |
12 | #include <linux/hugetlb.h> | |
ed4a6d7f | 13 | #include <linux/compaction.h> |
1da177e4 LT |
14 | #include <linux/cpumask.h> |
15 | #include <linux/topology.h> | |
16 | #include <linux/nodemask.h> | |
76b67ed9 | 17 | #include <linux/cpu.h> |
bde631a5 | 18 | #include <linux/device.h> |
af936a16 | 19 | #include <linux/swap.h> |
18e5b539 | 20 | #include <linux/slab.h> |
1da177e4 | 21 | |
10fbcf4c | 22 | static struct bus_type node_subsys = { |
af5ca3f4 | 23 | .name = "node", |
10fbcf4c | 24 | .dev_name = "node", |
1da177e4 LT |
25 | }; |
26 | ||
27 | ||
5aaba363 | 28 | static ssize_t node_read_cpumap(struct device *dev, bool list, char *buf) |
1da177e4 LT |
29 | { |
30 | struct node *node_dev = to_node(dev); | |
10fbcf4c | 31 | const struct cpumask *mask = cpumask_of_node(node_dev->dev.id); |
1da177e4 | 32 | |
39106dcf MT |
33 | /* 2008/04/07: buf currently PAGE_SIZE, need 9 chars per 32 bits. */ |
34 | BUILD_BUG_ON((NR_CPUS/32 * 9) > (PAGE_SIZE-1)); | |
1da177e4 | 35 | |
5aaba363 | 36 | return cpumap_print_to_pagebuf(list, buf, mask); |
1da177e4 LT |
37 | } |
38 | ||
10fbcf4c KS |
39 | static inline ssize_t node_read_cpumask(struct device *dev, |
40 | struct device_attribute *attr, char *buf) | |
39106dcf | 41 | { |
5aaba363 | 42 | return node_read_cpumap(dev, false, buf); |
39106dcf | 43 | } |
10fbcf4c KS |
44 | static inline ssize_t node_read_cpulist(struct device *dev, |
45 | struct device_attribute *attr, char *buf) | |
39106dcf | 46 | { |
5aaba363 | 47 | return node_read_cpumap(dev, true, buf); |
39106dcf MT |
48 | } |
49 | ||
10fbcf4c KS |
50 | static DEVICE_ATTR(cpumap, S_IRUGO, node_read_cpumask, NULL); |
51 | static DEVICE_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL); | |
1da177e4 LT |
52 | |
53 | #define K(x) ((x) << (PAGE_SHIFT - 10)) | |
10fbcf4c KS |
54 | static ssize_t node_read_meminfo(struct device *dev, |
55 | struct device_attribute *attr, char *buf) | |
1da177e4 LT |
56 | { |
57 | int n; | |
58 | int nid = dev->id; | |
599d0c95 | 59 | struct pglist_data *pgdat = NODE_DATA(nid); |
1da177e4 | 60 | struct sysinfo i; |
1da177e4 LT |
61 | |
62 | si_meminfo_node(&i, nid); | |
7ee92255 | 63 | n = sprintf(buf, |
4f98a2fe RR |
64 | "Node %d MemTotal: %8lu kB\n" |
65 | "Node %d MemFree: %8lu kB\n" | |
66 | "Node %d MemUsed: %8lu kB\n" | |
67 | "Node %d Active: %8lu kB\n" | |
68 | "Node %d Inactive: %8lu kB\n" | |
69 | "Node %d Active(anon): %8lu kB\n" | |
70 | "Node %d Inactive(anon): %8lu kB\n" | |
71 | "Node %d Active(file): %8lu kB\n" | |
72 | "Node %d Inactive(file): %8lu kB\n" | |
5344b7e6 | 73 | "Node %d Unevictable: %8lu kB\n" |
7ee92255 KM |
74 | "Node %d Mlocked: %8lu kB\n", |
75 | nid, K(i.totalram), | |
76 | nid, K(i.freeram), | |
77 | nid, K(i.totalram - i.freeram), | |
599d0c95 MG |
78 | nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) + |
79 | node_page_state(pgdat, NR_ACTIVE_FILE)), | |
80 | nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) + | |
81 | node_page_state(pgdat, NR_INACTIVE_FILE)), | |
82 | nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)), | |
83 | nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)), | |
84 | nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)), | |
85 | nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)), | |
86 | nid, K(node_page_state(pgdat, NR_UNEVICTABLE)), | |
75ef7184 | 87 | nid, K(sum_zone_node_page_state(nid, NR_MLOCK))); |
7ee92255 | 88 | |
182e8e23 | 89 | #ifdef CONFIG_HIGHMEM |
7ee92255 | 90 | n += sprintf(buf + n, |
4f98a2fe RR |
91 | "Node %d HighTotal: %8lu kB\n" |
92 | "Node %d HighFree: %8lu kB\n" | |
93 | "Node %d LowTotal: %8lu kB\n" | |
7ee92255 KM |
94 | "Node %d LowFree: %8lu kB\n", |
95 | nid, K(i.totalhigh), | |
96 | nid, K(i.freehigh), | |
97 | nid, K(i.totalram - i.totalhigh), | |
98 | nid, K(i.freeram - i.freehigh)); | |
182e8e23 | 99 | #endif |
7ee92255 | 100 | n += sprintf(buf + n, |
4f98a2fe RR |
101 | "Node %d Dirty: %8lu kB\n" |
102 | "Node %d Writeback: %8lu kB\n" | |
103 | "Node %d FilePages: %8lu kB\n" | |
104 | "Node %d Mapped: %8lu kB\n" | |
105 | "Node %d AnonPages: %8lu kB\n" | |
4b02108a | 106 | "Node %d Shmem: %8lu kB\n" |
c6a7f572 | 107 | "Node %d KernelStack: %8lu kB\n" |
4f98a2fe RR |
108 | "Node %d PageTables: %8lu kB\n" |
109 | "Node %d NFS_Unstable: %8lu kB\n" | |
110 | "Node %d Bounce: %8lu kB\n" | |
111 | "Node %d WritebackTmp: %8lu kB\n" | |
112 | "Node %d Slab: %8lu kB\n" | |
113 | "Node %d SReclaimable: %8lu kB\n" | |
05b258e9 DR |
114 | "Node %d SUnreclaim: %8lu kB\n" |
115 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
116 | "Node %d AnonHugePages: %8lu kB\n" | |
65c45377 KS |
117 | "Node %d ShmemHugePages: %8lu kB\n" |
118 | "Node %d ShmemPmdMapped: %8lu kB\n" | |
05b258e9 DR |
119 | #endif |
120 | , | |
11fb9989 MG |
121 | nid, K(node_page_state(pgdat, NR_FILE_DIRTY)), |
122 | nid, K(node_page_state(pgdat, NR_WRITEBACK)), | |
123 | nid, K(node_page_state(pgdat, NR_FILE_PAGES)), | |
50658e2e | 124 | nid, K(node_page_state(pgdat, NR_FILE_MAPPED)), |
4b9d0fab | 125 | nid, K(node_page_state(pgdat, NR_ANON_MAPPED)), |
cc7452b6 | 126 | nid, K(i.sharedram), |
d30dd8be | 127 | nid, sum_zone_node_page_state(nid, NR_KERNEL_STACK_KB), |
75ef7184 | 128 | nid, K(sum_zone_node_page_state(nid, NR_PAGETABLE)), |
11fb9989 | 129 | nid, K(node_page_state(pgdat, NR_UNSTABLE_NFS)), |
75ef7184 | 130 | nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)), |
11fb9989 | 131 | nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)), |
75ef7184 MG |
132 | nid, K(sum_zone_node_page_state(nid, NR_SLAB_RECLAIMABLE) + |
133 | sum_zone_node_page_state(nid, NR_SLAB_UNRECLAIMABLE)), | |
134 | nid, K(sum_zone_node_page_state(nid, NR_SLAB_RECLAIMABLE)), | |
05b258e9 | 135 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
75ef7184 | 136 | nid, K(sum_zone_node_page_state(nid, NR_SLAB_UNRECLAIMABLE)), |
11fb9989 | 137 | nid, K(node_page_state(pgdat, NR_ANON_THPS) * |
65c45377 | 138 | HPAGE_PMD_NR), |
11fb9989 | 139 | nid, K(node_page_state(pgdat, NR_SHMEM_THPS) * |
65c45377 | 140 | HPAGE_PMD_NR), |
11fb9989 | 141 | nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) * |
65c45377 | 142 | HPAGE_PMD_NR)); |
91a13c28 | 143 | #else |
75ef7184 | 144 | nid, K(sum_zone_node_page_state(nid, NR_SLAB_UNRECLAIMABLE))); |
05b258e9 | 145 | #endif |
1da177e4 LT |
146 | n += hugetlb_report_node_meminfo(nid, buf + n); |
147 | return n; | |
148 | } | |
149 | ||
150 | #undef K | |
10fbcf4c | 151 | static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL); |
1da177e4 | 152 | |
10fbcf4c KS |
153 | static ssize_t node_read_numastat(struct device *dev, |
154 | struct device_attribute *attr, char *buf) | |
1da177e4 | 155 | { |
1da177e4 LT |
156 | return sprintf(buf, |
157 | "numa_hit %lu\n" | |
158 | "numa_miss %lu\n" | |
159 | "numa_foreign %lu\n" | |
160 | "interleave_hit %lu\n" | |
161 | "local_node %lu\n" | |
162 | "other_node %lu\n", | |
75ef7184 MG |
163 | sum_zone_node_page_state(dev->id, NUMA_HIT), |
164 | sum_zone_node_page_state(dev->id, NUMA_MISS), | |
165 | sum_zone_node_page_state(dev->id, NUMA_FOREIGN), | |
166 | sum_zone_node_page_state(dev->id, NUMA_INTERLEAVE_HIT), | |
167 | sum_zone_node_page_state(dev->id, NUMA_LOCAL), | |
168 | sum_zone_node_page_state(dev->id, NUMA_OTHER)); | |
1da177e4 | 169 | } |
10fbcf4c | 170 | static DEVICE_ATTR(numastat, S_IRUGO, node_read_numastat, NULL); |
1da177e4 | 171 | |
10fbcf4c KS |
172 | static ssize_t node_read_vmstat(struct device *dev, |
173 | struct device_attribute *attr, char *buf) | |
2ac39037 MR |
174 | { |
175 | int nid = dev->id; | |
75ef7184 | 176 | struct pglist_data *pgdat = NODE_DATA(nid); |
fa25c503 KM |
177 | int i; |
178 | int n = 0; | |
179 | ||
180 | for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) | |
181 | n += sprintf(buf+n, "%s %lu\n", vmstat_text[i], | |
75ef7184 MG |
182 | sum_zone_node_page_state(nid, i)); |
183 | ||
184 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) | |
185 | n += sprintf(buf+n, "%s %lu\n", | |
186 | vmstat_text[i + NR_VM_ZONE_STAT_ITEMS], | |
187 | node_page_state(pgdat, i)); | |
fa25c503 KM |
188 | |
189 | return n; | |
2ac39037 | 190 | } |
10fbcf4c | 191 | static DEVICE_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL); |
2ac39037 | 192 | |
10fbcf4c | 193 | static ssize_t node_read_distance(struct device *dev, |
518d3f38 | 194 | struct device_attribute *attr, char *buf) |
1da177e4 LT |
195 | { |
196 | int nid = dev->id; | |
197 | int len = 0; | |
198 | int i; | |
199 | ||
12ee3c0a DR |
200 | /* |
201 | * buf is currently PAGE_SIZE in length and each node needs 4 chars | |
202 | * at the most (distance + space or newline). | |
203 | */ | |
204 | BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE); | |
1da177e4 LT |
205 | |
206 | for_each_online_node(i) | |
207 | len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i)); | |
208 | ||
209 | len += sprintf(buf + len, "\n"); | |
210 | return len; | |
211 | } | |
10fbcf4c | 212 | static DEVICE_ATTR(distance, S_IRUGO, node_read_distance, NULL); |
1da177e4 | 213 | |
3c9b8aaf TI |
214 | static struct attribute *node_dev_attrs[] = { |
215 | &dev_attr_cpumap.attr, | |
216 | &dev_attr_cpulist.attr, | |
217 | &dev_attr_meminfo.attr, | |
218 | &dev_attr_numastat.attr, | |
219 | &dev_attr_distance.attr, | |
220 | &dev_attr_vmstat.attr, | |
221 | NULL | |
222 | }; | |
7ca7ec40 | 223 | ATTRIBUTE_GROUPS(node_dev); |
3c9b8aaf | 224 | |
9a305230 LS |
225 | #ifdef CONFIG_HUGETLBFS |
226 | /* | |
227 | * hugetlbfs per node attributes registration interface: | |
228 | * When/if hugetlb[fs] subsystem initializes [sometime after this module], | |
4faf8d95 LS |
229 | * it will register its per node attributes for all online nodes with |
230 | * memory. It will also call register_hugetlbfs_with_node(), below, to | |
9a305230 LS |
231 | * register its attribute registration functions with this node driver. |
232 | * Once these hooks have been initialized, the node driver will call into | |
233 | * the hugetlb module to [un]register attributes for hot-plugged nodes. | |
234 | */ | |
235 | static node_registration_func_t __hugetlb_register_node; | |
236 | static node_registration_func_t __hugetlb_unregister_node; | |
237 | ||
39da08cb | 238 | static inline bool hugetlb_register_node(struct node *node) |
9a305230 | 239 | { |
4faf8d95 | 240 | if (__hugetlb_register_node && |
8cebfcd0 | 241 | node_state(node->dev.id, N_MEMORY)) { |
9a305230 | 242 | __hugetlb_register_node(node); |
39da08cb LS |
243 | return true; |
244 | } | |
245 | return false; | |
9a305230 LS |
246 | } |
247 | ||
248 | static inline void hugetlb_unregister_node(struct node *node) | |
249 | { | |
250 | if (__hugetlb_unregister_node) | |
251 | __hugetlb_unregister_node(node); | |
252 | } | |
253 | ||
254 | void register_hugetlbfs_with_node(node_registration_func_t doregister, | |
255 | node_registration_func_t unregister) | |
256 | { | |
257 | __hugetlb_register_node = doregister; | |
258 | __hugetlb_unregister_node = unregister; | |
259 | } | |
260 | #else | |
261 | static inline void hugetlb_register_node(struct node *node) {} | |
262 | ||
263 | static inline void hugetlb_unregister_node(struct node *node) {} | |
264 | #endif | |
265 | ||
8c7b5b4e YI |
266 | static void node_device_release(struct device *dev) |
267 | { | |
268 | struct node *node = to_node(dev); | |
269 | ||
270 | #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS) | |
271 | /* | |
272 | * We schedule the work only when a memory section is | |
273 | * onlined/offlined on this node. When we come here, | |
274 | * all the memory on this node has been offlined, | |
275 | * so we won't enqueue new work to this work. | |
276 | * | |
277 | * The work is using node->node_work, so we should | |
278 | * flush work before freeing the memory. | |
279 | */ | |
280 | flush_work(&node->node_work); | |
281 | #endif | |
282 | kfree(node); | |
283 | } | |
1da177e4 LT |
284 | |
285 | /* | |
405ae7d3 | 286 | * register_node - Setup a sysfs device for a node. |
1da177e4 LT |
287 | * @num - Node number to use when creating the device. |
288 | * | |
289 | * Initialize and register the node device. | |
290 | */ | |
fa264375 | 291 | static int register_node(struct node *node, int num, struct node *parent) |
1da177e4 LT |
292 | { |
293 | int error; | |
294 | ||
10fbcf4c KS |
295 | node->dev.id = num; |
296 | node->dev.bus = &node_subsys; | |
8c7b5b4e | 297 | node->dev.release = node_device_release; |
7ca7ec40 | 298 | node->dev.groups = node_dev_groups; |
10fbcf4c | 299 | error = device_register(&node->dev); |
1da177e4 LT |
300 | |
301 | if (!error){ | |
9a305230 | 302 | hugetlb_register_node(node); |
ed4a6d7f MG |
303 | |
304 | compaction_register_node(node); | |
1da177e4 LT |
305 | } |
306 | return error; | |
307 | } | |
308 | ||
4b45099b KT |
309 | /** |
310 | * unregister_node - unregister a node device | |
311 | * @node: node going away | |
312 | * | |
313 | * Unregisters a node device @node. All the devices on the node must be | |
314 | * unregistered before calling this function. | |
315 | */ | |
316 | void unregister_node(struct node *node) | |
317 | { | |
4faf8d95 | 318 | hugetlb_unregister_node(node); /* no-op, if memoryless node */ |
af936a16 | 319 | |
10fbcf4c | 320 | device_unregister(&node->dev); |
4b45099b | 321 | } |
1da177e4 | 322 | |
8732794b | 323 | struct node *node_devices[MAX_NUMNODES]; |
0fc44159 | 324 | |
76b67ed9 KH |
325 | /* |
326 | * register cpu under node | |
327 | */ | |
328 | int register_cpu_under_node(unsigned int cpu, unsigned int nid) | |
329 | { | |
1830794a | 330 | int ret; |
8a25a2fd | 331 | struct device *obj; |
76b67ed9 | 332 | |
f8246f31 AC |
333 | if (!node_online(nid)) |
334 | return 0; | |
335 | ||
8a25a2fd | 336 | obj = get_cpu_device(cpu); |
f8246f31 AC |
337 | if (!obj) |
338 | return 0; | |
339 | ||
8732794b | 340 | ret = sysfs_create_link(&node_devices[nid]->dev.kobj, |
f8246f31 AC |
341 | &obj->kobj, |
342 | kobject_name(&obj->kobj)); | |
1830794a AC |
343 | if (ret) |
344 | return ret; | |
345 | ||
346 | return sysfs_create_link(&obj->kobj, | |
8732794b WC |
347 | &node_devices[nid]->dev.kobj, |
348 | kobject_name(&node_devices[nid]->dev.kobj)); | |
76b67ed9 KH |
349 | } |
350 | ||
351 | int unregister_cpu_under_node(unsigned int cpu, unsigned int nid) | |
352 | { | |
8a25a2fd | 353 | struct device *obj; |
b9d52dad AC |
354 | |
355 | if (!node_online(nid)) | |
356 | return 0; | |
357 | ||
8a25a2fd | 358 | obj = get_cpu_device(cpu); |
b9d52dad AC |
359 | if (!obj) |
360 | return 0; | |
361 | ||
8732794b | 362 | sysfs_remove_link(&node_devices[nid]->dev.kobj, |
b9d52dad | 363 | kobject_name(&obj->kobj)); |
1830794a | 364 | sysfs_remove_link(&obj->kobj, |
8732794b | 365 | kobject_name(&node_devices[nid]->dev.kobj)); |
b9d52dad | 366 | |
76b67ed9 KH |
367 | return 0; |
368 | } | |
369 | ||
c04fc586 | 370 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
bd721ea7 | 371 | static int __ref get_nid_for_pfn(unsigned long pfn) |
c04fc586 | 372 | { |
c04fc586 GH |
373 | if (!pfn_valid_within(pfn)) |
374 | return -1; | |
3a80a7fa | 375 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
8cdde385 | 376 | if (system_state < SYSTEM_RUNNING) |
3a80a7fa MG |
377 | return early_pfn_to_nid(pfn); |
378 | #endif | |
c04fc586 GH |
379 | return pfn_to_nid(pfn); |
380 | } | |
381 | ||
382 | /* register memory section under specified node if it spans that node */ | |
383 | int register_mem_sect_under_node(struct memory_block *mem_blk, int nid) | |
384 | { | |
dee5d0d5 | 385 | int ret; |
c04fc586 GH |
386 | unsigned long pfn, sect_start_pfn, sect_end_pfn; |
387 | ||
388 | if (!mem_blk) | |
389 | return -EFAULT; | |
390 | if (!node_online(nid)) | |
391 | return 0; | |
d3360164 NF |
392 | |
393 | sect_start_pfn = section_nr_to_pfn(mem_blk->start_section_nr); | |
394 | sect_end_pfn = section_nr_to_pfn(mem_blk->end_section_nr); | |
395 | sect_end_pfn += PAGES_PER_SECTION - 1; | |
c04fc586 GH |
396 | for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) { |
397 | int page_nid; | |
398 | ||
04697858 YL |
399 | /* |
400 | * memory block could have several absent sections from start. | |
401 | * skip pfn range from absent section | |
402 | */ | |
403 | if (!pfn_present(pfn)) { | |
404 | pfn = round_down(pfn + PAGES_PER_SECTION, | |
405 | PAGES_PER_SECTION) - 1; | |
406 | continue; | |
407 | } | |
408 | ||
c04fc586 GH |
409 | page_nid = get_nid_for_pfn(pfn); |
410 | if (page_nid < 0) | |
411 | continue; | |
412 | if (page_nid != nid) | |
413 | continue; | |
8732794b | 414 | ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj, |
10fbcf4c KS |
415 | &mem_blk->dev.kobj, |
416 | kobject_name(&mem_blk->dev.kobj)); | |
dee5d0d5 AC |
417 | if (ret) |
418 | return ret; | |
419 | ||
10fbcf4c | 420 | return sysfs_create_link_nowarn(&mem_blk->dev.kobj, |
8732794b WC |
421 | &node_devices[nid]->dev.kobj, |
422 | kobject_name(&node_devices[nid]->dev.kobj)); | |
c04fc586 GH |
423 | } |
424 | /* mem section does not span the specified node */ | |
425 | return 0; | |
426 | } | |
427 | ||
428 | /* unregister memory section under all nodes that it spans */ | |
d3360164 NF |
429 | int unregister_mem_sect_under_nodes(struct memory_block *mem_blk, |
430 | unsigned long phys_index) | |
c04fc586 | 431 | { |
9ae49fab | 432 | NODEMASK_ALLOC(nodemask_t, unlinked_nodes, GFP_KERNEL); |
c04fc586 GH |
433 | unsigned long pfn, sect_start_pfn, sect_end_pfn; |
434 | ||
9ae49fab DR |
435 | if (!mem_blk) { |
436 | NODEMASK_FREE(unlinked_nodes); | |
c04fc586 | 437 | return -EFAULT; |
9ae49fab DR |
438 | } |
439 | if (!unlinked_nodes) | |
440 | return -ENOMEM; | |
441 | nodes_clear(*unlinked_nodes); | |
d3360164 NF |
442 | |
443 | sect_start_pfn = section_nr_to_pfn(phys_index); | |
c04fc586 GH |
444 | sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1; |
445 | for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) { | |
47504980 | 446 | int nid; |
c04fc586 GH |
447 | |
448 | nid = get_nid_for_pfn(pfn); | |
449 | if (nid < 0) | |
450 | continue; | |
451 | if (!node_online(nid)) | |
452 | continue; | |
9ae49fab | 453 | if (node_test_and_set(nid, *unlinked_nodes)) |
c04fc586 | 454 | continue; |
8732794b | 455 | sysfs_remove_link(&node_devices[nid]->dev.kobj, |
10fbcf4c KS |
456 | kobject_name(&mem_blk->dev.kobj)); |
457 | sysfs_remove_link(&mem_blk->dev.kobj, | |
8732794b | 458 | kobject_name(&node_devices[nid]->dev.kobj)); |
c04fc586 | 459 | } |
9ae49fab | 460 | NODEMASK_FREE(unlinked_nodes); |
c04fc586 GH |
461 | return 0; |
462 | } | |
463 | ||
464 | static int link_mem_sections(int nid) | |
465 | { | |
466 | unsigned long start_pfn = NODE_DATA(nid)->node_start_pfn; | |
467 | unsigned long end_pfn = start_pfn + NODE_DATA(nid)->node_spanned_pages; | |
468 | unsigned long pfn; | |
63d027a6 | 469 | struct memory_block *mem_blk = NULL; |
c04fc586 GH |
470 | int err = 0; |
471 | ||
472 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
473 | unsigned long section_nr = pfn_to_section_nr(pfn); | |
474 | struct mem_section *mem_sect; | |
c04fc586 GH |
475 | int ret; |
476 | ||
477 | if (!present_section_nr(section_nr)) | |
478 | continue; | |
479 | mem_sect = __nr_to_section(section_nr); | |
321bf4ed YL |
480 | |
481 | /* same memblock ? */ | |
482 | if (mem_blk) | |
483 | if ((section_nr >= mem_blk->start_section_nr) && | |
484 | (section_nr <= mem_blk->end_section_nr)) | |
485 | continue; | |
486 | ||
63d027a6 | 487 | mem_blk = find_memory_block_hinted(mem_sect, mem_blk); |
321bf4ed | 488 | |
c04fc586 GH |
489 | ret = register_mem_sect_under_node(mem_blk, nid); |
490 | if (!err) | |
491 | err = ret; | |
492 | ||
493 | /* discard ref obtained in find_memory_block() */ | |
c04fc586 | 494 | } |
63d027a6 RH |
495 | |
496 | if (mem_blk) | |
10fbcf4c | 497 | kobject_put(&mem_blk->dev.kobj); |
c04fc586 GH |
498 | return err; |
499 | } | |
4faf8d95 | 500 | |
39da08cb | 501 | #ifdef CONFIG_HUGETLBFS |
4faf8d95 LS |
502 | /* |
503 | * Handle per node hstate attribute [un]registration on transistions | |
504 | * to/from memoryless state. | |
505 | */ | |
39da08cb LS |
506 | static void node_hugetlb_work(struct work_struct *work) |
507 | { | |
508 | struct node *node = container_of(work, struct node, node_work); | |
509 | ||
510 | /* | |
511 | * We only get here when a node transitions to/from memoryless state. | |
512 | * We can detect which transition occurred by examining whether the | |
513 | * node has memory now. hugetlb_register_node() already check this | |
514 | * so we try to register the attributes. If that fails, then the | |
515 | * node has transitioned to memoryless, try to unregister the | |
516 | * attributes. | |
517 | */ | |
518 | if (!hugetlb_register_node(node)) | |
519 | hugetlb_unregister_node(node); | |
520 | } | |
521 | ||
522 | static void init_node_hugetlb_work(int nid) | |
523 | { | |
8732794b | 524 | INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work); |
39da08cb | 525 | } |
4faf8d95 LS |
526 | |
527 | static int node_memory_callback(struct notifier_block *self, | |
528 | unsigned long action, void *arg) | |
529 | { | |
530 | struct memory_notify *mnb = arg; | |
531 | int nid = mnb->status_change_nid; | |
532 | ||
533 | switch (action) { | |
39da08cb LS |
534 | case MEM_ONLINE: |
535 | case MEM_OFFLINE: | |
536 | /* | |
537 | * offload per node hstate [un]registration to a work thread | |
538 | * when transitioning to/from memoryless state. | |
539 | */ | |
4faf8d95 | 540 | if (nid != NUMA_NO_NODE) |
8732794b | 541 | schedule_work(&node_devices[nid]->node_work); |
4faf8d95 | 542 | break; |
39da08cb | 543 | |
4faf8d95 LS |
544 | case MEM_GOING_ONLINE: |
545 | case MEM_GOING_OFFLINE: | |
546 | case MEM_CANCEL_ONLINE: | |
547 | case MEM_CANCEL_OFFLINE: | |
548 | default: | |
549 | break; | |
550 | } | |
551 | ||
552 | return NOTIFY_OK; | |
553 | } | |
39da08cb LS |
554 | #endif /* CONFIG_HUGETLBFS */ |
555 | #else /* !CONFIG_MEMORY_HOTPLUG_SPARSE */ | |
556 | ||
c04fc586 | 557 | static int link_mem_sections(int nid) { return 0; } |
39da08cb | 558 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
4faf8d95 | 559 | |
39da08cb LS |
560 | #if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \ |
561 | !defined(CONFIG_HUGETLBFS) | |
4faf8d95 LS |
562 | static inline int node_memory_callback(struct notifier_block *self, |
563 | unsigned long action, void *arg) | |
564 | { | |
565 | return NOTIFY_OK; | |
566 | } | |
39da08cb LS |
567 | |
568 | static void init_node_hugetlb_work(int nid) { } | |
569 | ||
570 | #endif | |
c04fc586 | 571 | |
0fc44159 YG |
572 | int register_one_node(int nid) |
573 | { | |
574 | int error = 0; | |
76b67ed9 | 575 | int cpu; |
0fc44159 YG |
576 | |
577 | if (node_online(nid)) { | |
578 | int p_node = parent_node(nid); | |
579 | struct node *parent = NULL; | |
580 | ||
581 | if (p_node != nid) | |
8732794b | 582 | parent = node_devices[p_node]; |
0fc44159 | 583 | |
8732794b WC |
584 | node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL); |
585 | if (!node_devices[nid]) | |
586 | return -ENOMEM; | |
587 | ||
588 | error = register_node(node_devices[nid], nid, parent); | |
76b67ed9 KH |
589 | |
590 | /* link cpu under this node */ | |
591 | for_each_present_cpu(cpu) { | |
592 | if (cpu_to_node(cpu) == nid) | |
593 | register_cpu_under_node(cpu, nid); | |
594 | } | |
c04fc586 GH |
595 | |
596 | /* link memory sections under this node */ | |
597 | error = link_mem_sections(nid); | |
39da08cb LS |
598 | |
599 | /* initialize work queue for memory hot plug */ | |
600 | init_node_hugetlb_work(nid); | |
0fc44159 YG |
601 | } |
602 | ||
603 | return error; | |
604 | ||
605 | } | |
606 | ||
607 | void unregister_one_node(int nid) | |
608 | { | |
92d585ef XQ |
609 | if (!node_devices[nid]) |
610 | return; | |
611 | ||
8732794b | 612 | unregister_node(node_devices[nid]); |
8732794b | 613 | node_devices[nid] = NULL; |
0fc44159 YG |
614 | } |
615 | ||
bde631a5 LS |
616 | /* |
617 | * node states attributes | |
618 | */ | |
619 | ||
620 | static ssize_t print_nodes_state(enum node_states state, char *buf) | |
621 | { | |
622 | int n; | |
623 | ||
f799b1a7 TH |
624 | n = scnprintf(buf, PAGE_SIZE - 1, "%*pbl", |
625 | nodemask_pr_args(&node_states[state])); | |
f6238818 RO |
626 | buf[n++] = '\n'; |
627 | buf[n] = '\0'; | |
bde631a5 LS |
628 | return n; |
629 | } | |
630 | ||
b15f562f | 631 | struct node_attr { |
10fbcf4c | 632 | struct device_attribute attr; |
b15f562f AK |
633 | enum node_states state; |
634 | }; | |
bde631a5 | 635 | |
10fbcf4c KS |
636 | static ssize_t show_node_state(struct device *dev, |
637 | struct device_attribute *attr, char *buf) | |
bde631a5 | 638 | { |
b15f562f AK |
639 | struct node_attr *na = container_of(attr, struct node_attr, attr); |
640 | return print_nodes_state(na->state, buf); | |
bde631a5 LS |
641 | } |
642 | ||
b15f562f | 643 | #define _NODE_ATTR(name, state) \ |
10fbcf4c | 644 | { __ATTR(name, 0444, show_node_state, NULL), state } |
bde631a5 | 645 | |
b15f562f | 646 | static struct node_attr node_state_attr[] = { |
fcf07d22 LJ |
647 | [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE), |
648 | [N_ONLINE] = _NODE_ATTR(online, N_ONLINE), | |
649 | [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY), | |
bde631a5 | 650 | #ifdef CONFIG_HIGHMEM |
fcf07d22 | 651 | [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY), |
20b2f52b LJ |
652 | #endif |
653 | #ifdef CONFIG_MOVABLE_NODE | |
654 | [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY), | |
bde631a5 | 655 | #endif |
fcf07d22 | 656 | [N_CPU] = _NODE_ATTR(has_cpu, N_CPU), |
bde631a5 LS |
657 | }; |
658 | ||
10fbcf4c | 659 | static struct attribute *node_state_attrs[] = { |
fcf07d22 LJ |
660 | &node_state_attr[N_POSSIBLE].attr.attr, |
661 | &node_state_attr[N_ONLINE].attr.attr, | |
662 | &node_state_attr[N_NORMAL_MEMORY].attr.attr, | |
3701cde6 | 663 | #ifdef CONFIG_HIGHMEM |
fcf07d22 | 664 | &node_state_attr[N_HIGH_MEMORY].attr.attr, |
20b2f52b LJ |
665 | #endif |
666 | #ifdef CONFIG_MOVABLE_NODE | |
667 | &node_state_attr[N_MEMORY].attr.attr, | |
3701cde6 | 668 | #endif |
fcf07d22 | 669 | &node_state_attr[N_CPU].attr.attr, |
3701cde6 AK |
670 | NULL |
671 | }; | |
bde631a5 | 672 | |
10fbcf4c KS |
673 | static struct attribute_group memory_root_attr_group = { |
674 | .attrs = node_state_attrs, | |
675 | }; | |
676 | ||
677 | static const struct attribute_group *cpu_root_attr_groups[] = { | |
678 | &memory_root_attr_group, | |
679 | NULL, | |
680 | }; | |
681 | ||
4faf8d95 | 682 | #define NODE_CALLBACK_PRI 2 /* lower than SLAB */ |
4b45099b | 683 | static int __init register_node_type(void) |
1da177e4 | 684 | { |
bde631a5 LS |
685 | int ret; |
686 | ||
3701cde6 AK |
687 | BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES); |
688 | BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES); | |
689 | ||
10fbcf4c | 690 | ret = subsys_system_register(&node_subsys, cpu_root_attr_groups); |
4faf8d95 | 691 | if (!ret) { |
6e259e7d AM |
692 | static struct notifier_block node_memory_callback_nb = { |
693 | .notifier_call = node_memory_callback, | |
694 | .priority = NODE_CALLBACK_PRI, | |
695 | }; | |
696 | register_hotmemory_notifier(&node_memory_callback_nb); | |
4faf8d95 | 697 | } |
bde631a5 LS |
698 | |
699 | /* | |
700 | * Note: we're not going to unregister the node class if we fail | |
701 | * to register the node state class attribute files. | |
702 | */ | |
703 | return ret; | |
1da177e4 LT |
704 | } |
705 | postcore_initcall(register_node_type); |