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> |
1da177e4 LT |
10 | #include <linux/node.h> |
11 | #include <linux/hugetlb.h> | |
ed4a6d7f | 12 | #include <linux/compaction.h> |
1da177e4 LT |
13 | #include <linux/cpumask.h> |
14 | #include <linux/topology.h> | |
15 | #include <linux/nodemask.h> | |
76b67ed9 | 16 | #include <linux/cpu.h> |
bde631a5 | 17 | #include <linux/device.h> |
af936a16 | 18 | #include <linux/swap.h> |
18e5b539 | 19 | #include <linux/slab.h> |
1da177e4 | 20 | |
10fbcf4c | 21 | static struct bus_type node_subsys = { |
af5ca3f4 | 22 | .name = "node", |
10fbcf4c | 23 | .dev_name = "node", |
1da177e4 LT |
24 | }; |
25 | ||
26 | ||
10fbcf4c | 27 | static ssize_t node_read_cpumap(struct device *dev, int type, char *buf) |
1da177e4 LT |
28 | { |
29 | struct node *node_dev = to_node(dev); | |
10fbcf4c | 30 | const struct cpumask *mask = cpumask_of_node(node_dev->dev.id); |
1da177e4 LT |
31 | int len; |
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 | |
39106dcf | 36 | len = type? |
29c0177e RR |
37 | cpulist_scnprintf(buf, PAGE_SIZE-2, mask) : |
38 | cpumask_scnprintf(buf, PAGE_SIZE-2, mask); | |
c5f59f08 MT |
39 | buf[len++] = '\n'; |
40 | buf[len] = '\0'; | |
1da177e4 LT |
41 | return len; |
42 | } | |
43 | ||
10fbcf4c KS |
44 | static inline ssize_t node_read_cpumask(struct device *dev, |
45 | struct device_attribute *attr, char *buf) | |
39106dcf MT |
46 | { |
47 | return node_read_cpumap(dev, 0, buf); | |
48 | } | |
10fbcf4c KS |
49 | static inline ssize_t node_read_cpulist(struct device *dev, |
50 | struct device_attribute *attr, char *buf) | |
39106dcf MT |
51 | { |
52 | return node_read_cpumap(dev, 1, buf); | |
53 | } | |
54 | ||
10fbcf4c KS |
55 | static DEVICE_ATTR(cpumap, S_IRUGO, node_read_cpumask, NULL); |
56 | static DEVICE_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL); | |
1da177e4 LT |
57 | |
58 | #define K(x) ((x) << (PAGE_SHIFT - 10)) | |
10fbcf4c KS |
59 | static ssize_t node_read_meminfo(struct device *dev, |
60 | struct device_attribute *attr, char *buf) | |
1da177e4 LT |
61 | { |
62 | int n; | |
63 | int nid = dev->id; | |
64 | struct sysinfo i; | |
1da177e4 LT |
65 | |
66 | si_meminfo_node(&i, nid); | |
7ee92255 | 67 | n = sprintf(buf, |
4f98a2fe RR |
68 | "Node %d MemTotal: %8lu kB\n" |
69 | "Node %d MemFree: %8lu kB\n" | |
70 | "Node %d MemUsed: %8lu kB\n" | |
71 | "Node %d Active: %8lu kB\n" | |
72 | "Node %d Inactive: %8lu kB\n" | |
73 | "Node %d Active(anon): %8lu kB\n" | |
74 | "Node %d Inactive(anon): %8lu kB\n" | |
75 | "Node %d Active(file): %8lu kB\n" | |
76 | "Node %d Inactive(file): %8lu kB\n" | |
5344b7e6 | 77 | "Node %d Unevictable: %8lu kB\n" |
7ee92255 KM |
78 | "Node %d Mlocked: %8lu kB\n", |
79 | nid, K(i.totalram), | |
80 | nid, K(i.freeram), | |
81 | nid, K(i.totalram - i.freeram), | |
82 | nid, K(node_page_state(nid, NR_ACTIVE_ANON) + | |
83 | node_page_state(nid, NR_ACTIVE_FILE)), | |
84 | nid, K(node_page_state(nid, NR_INACTIVE_ANON) + | |
85 | node_page_state(nid, NR_INACTIVE_FILE)), | |
86 | nid, K(node_page_state(nid, NR_ACTIVE_ANON)), | |
87 | nid, K(node_page_state(nid, NR_INACTIVE_ANON)), | |
88 | nid, K(node_page_state(nid, NR_ACTIVE_FILE)), | |
89 | nid, K(node_page_state(nid, NR_INACTIVE_FILE)), | |
90 | nid, K(node_page_state(nid, NR_UNEVICTABLE)), | |
91 | nid, K(node_page_state(nid, NR_MLOCK))); | |
92 | ||
182e8e23 | 93 | #ifdef CONFIG_HIGHMEM |
7ee92255 | 94 | n += sprintf(buf + n, |
4f98a2fe RR |
95 | "Node %d HighTotal: %8lu kB\n" |
96 | "Node %d HighFree: %8lu kB\n" | |
97 | "Node %d LowTotal: %8lu kB\n" | |
7ee92255 KM |
98 | "Node %d LowFree: %8lu kB\n", |
99 | nid, K(i.totalhigh), | |
100 | nid, K(i.freehigh), | |
101 | nid, K(i.totalram - i.totalhigh), | |
102 | nid, K(i.freeram - i.freehigh)); | |
182e8e23 | 103 | #endif |
7ee92255 | 104 | n += sprintf(buf + n, |
4f98a2fe RR |
105 | "Node %d Dirty: %8lu kB\n" |
106 | "Node %d Writeback: %8lu kB\n" | |
107 | "Node %d FilePages: %8lu kB\n" | |
108 | "Node %d Mapped: %8lu kB\n" | |
109 | "Node %d AnonPages: %8lu kB\n" | |
4b02108a | 110 | "Node %d Shmem: %8lu kB\n" |
c6a7f572 | 111 | "Node %d KernelStack: %8lu kB\n" |
4f98a2fe RR |
112 | "Node %d PageTables: %8lu kB\n" |
113 | "Node %d NFS_Unstable: %8lu kB\n" | |
114 | "Node %d Bounce: %8lu kB\n" | |
115 | "Node %d WritebackTmp: %8lu kB\n" | |
116 | "Node %d Slab: %8lu kB\n" | |
117 | "Node %d SReclaimable: %8lu kB\n" | |
05b258e9 DR |
118 | "Node %d SUnreclaim: %8lu kB\n" |
119 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
120 | "Node %d AnonHugePages: %8lu kB\n" | |
121 | #endif | |
122 | , | |
b1e7a8fd | 123 | nid, K(node_page_state(nid, NR_FILE_DIRTY)), |
ce866b34 | 124 | nid, K(node_page_state(nid, NR_WRITEBACK)), |
347ce434 | 125 | nid, K(node_page_state(nid, NR_FILE_PAGES)), |
65ba55f5 | 126 | nid, K(node_page_state(nid, NR_FILE_MAPPED)), |
05b258e9 | 127 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
91a13c28 | 128 | nid, K(node_page_state(nid, NR_ANON_PAGES) |
05b258e9 | 129 | + node_page_state(nid, NR_ANON_TRANSPARENT_HUGEPAGES) * |
91a13c28 CS |
130 | HPAGE_PMD_NR), |
131 | #else | |
132 | nid, K(node_page_state(nid, NR_ANON_PAGES)), | |
05b258e9 | 133 | #endif |
4b02108a | 134 | nid, K(node_page_state(nid, NR_SHMEM)), |
c6a7f572 KM |
135 | nid, node_page_state(nid, NR_KERNEL_STACK) * |
136 | THREAD_SIZE / 1024, | |
df849a15 | 137 | nid, K(node_page_state(nid, NR_PAGETABLE)), |
fd39fc85 | 138 | nid, K(node_page_state(nid, NR_UNSTABLE_NFS)), |
d2c5e30c | 139 | nid, K(node_page_state(nid, NR_BOUNCE)), |
fc3ba692 | 140 | nid, K(node_page_state(nid, NR_WRITEBACK_TEMP)), |
972d1a7b CL |
141 | nid, K(node_page_state(nid, NR_SLAB_RECLAIMABLE) + |
142 | node_page_state(nid, NR_SLAB_UNRECLAIMABLE)), | |
143 | nid, K(node_page_state(nid, NR_SLAB_RECLAIMABLE)), | |
05b258e9 | 144 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
91a13c28 | 145 | nid, K(node_page_state(nid, NR_SLAB_UNRECLAIMABLE)) |
05b258e9 DR |
146 | , nid, |
147 | K(node_page_state(nid, NR_ANON_TRANSPARENT_HUGEPAGES) * | |
91a13c28 CS |
148 | HPAGE_PMD_NR)); |
149 | #else | |
150 | nid, K(node_page_state(nid, NR_SLAB_UNRECLAIMABLE))); | |
05b258e9 | 151 | #endif |
1da177e4 LT |
152 | n += hugetlb_report_node_meminfo(nid, buf + n); |
153 | return n; | |
154 | } | |
155 | ||
156 | #undef K | |
10fbcf4c | 157 | static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL); |
1da177e4 | 158 | |
10fbcf4c KS |
159 | static ssize_t node_read_numastat(struct device *dev, |
160 | struct device_attribute *attr, char *buf) | |
1da177e4 | 161 | { |
1da177e4 LT |
162 | return sprintf(buf, |
163 | "numa_hit %lu\n" | |
164 | "numa_miss %lu\n" | |
165 | "numa_foreign %lu\n" | |
166 | "interleave_hit %lu\n" | |
167 | "local_node %lu\n" | |
168 | "other_node %lu\n", | |
ca889e6c CL |
169 | node_page_state(dev->id, NUMA_HIT), |
170 | node_page_state(dev->id, NUMA_MISS), | |
171 | node_page_state(dev->id, NUMA_FOREIGN), | |
172 | node_page_state(dev->id, NUMA_INTERLEAVE_HIT), | |
173 | node_page_state(dev->id, NUMA_LOCAL), | |
174 | node_page_state(dev->id, NUMA_OTHER)); | |
1da177e4 | 175 | } |
10fbcf4c | 176 | static DEVICE_ATTR(numastat, S_IRUGO, node_read_numastat, NULL); |
1da177e4 | 177 | |
10fbcf4c KS |
178 | static ssize_t node_read_vmstat(struct device *dev, |
179 | struct device_attribute *attr, char *buf) | |
2ac39037 MR |
180 | { |
181 | int nid = dev->id; | |
fa25c503 KM |
182 | int i; |
183 | int n = 0; | |
184 | ||
185 | for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) | |
186 | n += sprintf(buf+n, "%s %lu\n", vmstat_text[i], | |
187 | node_page_state(nid, i)); | |
188 | ||
189 | return n; | |
2ac39037 | 190 | } |
10fbcf4c | 191 | static DEVICE_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL); |
2ac39037 | 192 | |
10fbcf4c KS |
193 | static ssize_t node_read_distance(struct device *dev, |
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 | |
9a305230 LS |
214 | #ifdef CONFIG_HUGETLBFS |
215 | /* | |
216 | * hugetlbfs per node attributes registration interface: | |
217 | * When/if hugetlb[fs] subsystem initializes [sometime after this module], | |
4faf8d95 LS |
218 | * it will register its per node attributes for all online nodes with |
219 | * memory. It will also call register_hugetlbfs_with_node(), below, to | |
9a305230 LS |
220 | * register its attribute registration functions with this node driver. |
221 | * Once these hooks have been initialized, the node driver will call into | |
222 | * the hugetlb module to [un]register attributes for hot-plugged nodes. | |
223 | */ | |
224 | static node_registration_func_t __hugetlb_register_node; | |
225 | static node_registration_func_t __hugetlb_unregister_node; | |
226 | ||
39da08cb | 227 | static inline bool hugetlb_register_node(struct node *node) |
9a305230 | 228 | { |
4faf8d95 | 229 | if (__hugetlb_register_node && |
8cebfcd0 | 230 | node_state(node->dev.id, N_MEMORY)) { |
9a305230 | 231 | __hugetlb_register_node(node); |
39da08cb LS |
232 | return true; |
233 | } | |
234 | return false; | |
9a305230 LS |
235 | } |
236 | ||
237 | static inline void hugetlb_unregister_node(struct node *node) | |
238 | { | |
239 | if (__hugetlb_unregister_node) | |
240 | __hugetlb_unregister_node(node); | |
241 | } | |
242 | ||
243 | void register_hugetlbfs_with_node(node_registration_func_t doregister, | |
244 | node_registration_func_t unregister) | |
245 | { | |
246 | __hugetlb_register_node = doregister; | |
247 | __hugetlb_unregister_node = unregister; | |
248 | } | |
249 | #else | |
250 | static inline void hugetlb_register_node(struct node *node) {} | |
251 | ||
252 | static inline void hugetlb_unregister_node(struct node *node) {} | |
253 | #endif | |
254 | ||
8c7b5b4e YI |
255 | static void node_device_release(struct device *dev) |
256 | { | |
257 | struct node *node = to_node(dev); | |
258 | ||
259 | #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS) | |
260 | /* | |
261 | * We schedule the work only when a memory section is | |
262 | * onlined/offlined on this node. When we come here, | |
263 | * all the memory on this node has been offlined, | |
264 | * so we won't enqueue new work to this work. | |
265 | * | |
266 | * The work is using node->node_work, so we should | |
267 | * flush work before freeing the memory. | |
268 | */ | |
269 | flush_work(&node->node_work); | |
270 | #endif | |
271 | kfree(node); | |
272 | } | |
1da177e4 LT |
273 | |
274 | /* | |
405ae7d3 | 275 | * register_node - Setup a sysfs device for a node. |
1da177e4 LT |
276 | * @num - Node number to use when creating the device. |
277 | * | |
278 | * Initialize and register the node device. | |
279 | */ | |
fa264375 | 280 | static int register_node(struct node *node, int num, struct node *parent) |
1da177e4 LT |
281 | { |
282 | int error; | |
283 | ||
10fbcf4c KS |
284 | node->dev.id = num; |
285 | node->dev.bus = &node_subsys; | |
8c7b5b4e | 286 | node->dev.release = node_device_release; |
10fbcf4c | 287 | error = device_register(&node->dev); |
1da177e4 LT |
288 | |
289 | if (!error){ | |
10fbcf4c KS |
290 | device_create_file(&node->dev, &dev_attr_cpumap); |
291 | device_create_file(&node->dev, &dev_attr_cpulist); | |
292 | device_create_file(&node->dev, &dev_attr_meminfo); | |
293 | device_create_file(&node->dev, &dev_attr_numastat); | |
294 | device_create_file(&node->dev, &dev_attr_distance); | |
295 | device_create_file(&node->dev, &dev_attr_vmstat); | |
af936a16 LS |
296 | |
297 | scan_unevictable_register_node(node); | |
4faf8d95 | 298 | |
9a305230 | 299 | hugetlb_register_node(node); |
ed4a6d7f MG |
300 | |
301 | compaction_register_node(node); | |
1da177e4 LT |
302 | } |
303 | return error; | |
304 | } | |
305 | ||
4b45099b KT |
306 | /** |
307 | * unregister_node - unregister a node device | |
308 | * @node: node going away | |
309 | * | |
310 | * Unregisters a node device @node. All the devices on the node must be | |
311 | * unregistered before calling this function. | |
312 | */ | |
313 | void unregister_node(struct node *node) | |
314 | { | |
10fbcf4c KS |
315 | device_remove_file(&node->dev, &dev_attr_cpumap); |
316 | device_remove_file(&node->dev, &dev_attr_cpulist); | |
317 | device_remove_file(&node->dev, &dev_attr_meminfo); | |
318 | device_remove_file(&node->dev, &dev_attr_numastat); | |
319 | device_remove_file(&node->dev, &dev_attr_distance); | |
320 | device_remove_file(&node->dev, &dev_attr_vmstat); | |
4b45099b | 321 | |
af936a16 | 322 | scan_unevictable_unregister_node(node); |
4faf8d95 | 323 | hugetlb_unregister_node(node); /* no-op, if memoryless node */ |
af936a16 | 324 | |
10fbcf4c | 325 | device_unregister(&node->dev); |
4b45099b | 326 | } |
1da177e4 | 327 | |
8732794b | 328 | struct node *node_devices[MAX_NUMNODES]; |
0fc44159 | 329 | |
76b67ed9 KH |
330 | /* |
331 | * register cpu under node | |
332 | */ | |
333 | int register_cpu_under_node(unsigned int cpu, unsigned int nid) | |
334 | { | |
1830794a | 335 | int ret; |
8a25a2fd | 336 | struct device *obj; |
76b67ed9 | 337 | |
f8246f31 AC |
338 | if (!node_online(nid)) |
339 | return 0; | |
340 | ||
8a25a2fd | 341 | obj = get_cpu_device(cpu); |
f8246f31 AC |
342 | if (!obj) |
343 | return 0; | |
344 | ||
8732794b | 345 | ret = sysfs_create_link(&node_devices[nid]->dev.kobj, |
f8246f31 AC |
346 | &obj->kobj, |
347 | kobject_name(&obj->kobj)); | |
1830794a AC |
348 | if (ret) |
349 | return ret; | |
350 | ||
351 | return sysfs_create_link(&obj->kobj, | |
8732794b WC |
352 | &node_devices[nid]->dev.kobj, |
353 | kobject_name(&node_devices[nid]->dev.kobj)); | |
76b67ed9 KH |
354 | } |
355 | ||
356 | int unregister_cpu_under_node(unsigned int cpu, unsigned int nid) | |
357 | { | |
8a25a2fd | 358 | struct device *obj; |
b9d52dad AC |
359 | |
360 | if (!node_online(nid)) | |
361 | return 0; | |
362 | ||
8a25a2fd | 363 | obj = get_cpu_device(cpu); |
b9d52dad AC |
364 | if (!obj) |
365 | return 0; | |
366 | ||
8732794b | 367 | sysfs_remove_link(&node_devices[nid]->dev.kobj, |
b9d52dad | 368 | kobject_name(&obj->kobj)); |
1830794a | 369 | sysfs_remove_link(&obj->kobj, |
8732794b | 370 | kobject_name(&node_devices[nid]->dev.kobj)); |
b9d52dad | 371 | |
76b67ed9 KH |
372 | return 0; |
373 | } | |
374 | ||
c04fc586 GH |
375 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
376 | #define page_initialized(page) (page->lru.next) | |
377 | ||
378 | static int get_nid_for_pfn(unsigned long pfn) | |
379 | { | |
380 | struct page *page; | |
381 | ||
382 | if (!pfn_valid_within(pfn)) | |
383 | return -1; | |
384 | page = pfn_to_page(pfn); | |
385 | if (!page_initialized(page)) | |
386 | return -1; | |
387 | return pfn_to_nid(pfn); | |
388 | } | |
389 | ||
390 | /* register memory section under specified node if it spans that node */ | |
391 | int register_mem_sect_under_node(struct memory_block *mem_blk, int nid) | |
392 | { | |
dee5d0d5 | 393 | int ret; |
c04fc586 GH |
394 | unsigned long pfn, sect_start_pfn, sect_end_pfn; |
395 | ||
396 | if (!mem_blk) | |
397 | return -EFAULT; | |
398 | if (!node_online(nid)) | |
399 | return 0; | |
d3360164 NF |
400 | |
401 | sect_start_pfn = section_nr_to_pfn(mem_blk->start_section_nr); | |
402 | sect_end_pfn = section_nr_to_pfn(mem_blk->end_section_nr); | |
403 | sect_end_pfn += PAGES_PER_SECTION - 1; | |
c04fc586 GH |
404 | for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) { |
405 | int page_nid; | |
406 | ||
407 | page_nid = get_nid_for_pfn(pfn); | |
408 | if (page_nid < 0) | |
409 | continue; | |
410 | if (page_nid != nid) | |
411 | continue; | |
8732794b | 412 | ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj, |
10fbcf4c KS |
413 | &mem_blk->dev.kobj, |
414 | kobject_name(&mem_blk->dev.kobj)); | |
dee5d0d5 AC |
415 | if (ret) |
416 | return ret; | |
417 | ||
10fbcf4c | 418 | return sysfs_create_link_nowarn(&mem_blk->dev.kobj, |
8732794b WC |
419 | &node_devices[nid]->dev.kobj, |
420 | kobject_name(&node_devices[nid]->dev.kobj)); | |
c04fc586 GH |
421 | } |
422 | /* mem section does not span the specified node */ | |
423 | return 0; | |
424 | } | |
425 | ||
426 | /* unregister memory section under all nodes that it spans */ | |
d3360164 NF |
427 | int unregister_mem_sect_under_nodes(struct memory_block *mem_blk, |
428 | unsigned long phys_index) | |
c04fc586 | 429 | { |
9ae49fab | 430 | NODEMASK_ALLOC(nodemask_t, unlinked_nodes, GFP_KERNEL); |
c04fc586 GH |
431 | unsigned long pfn, sect_start_pfn, sect_end_pfn; |
432 | ||
9ae49fab DR |
433 | if (!mem_blk) { |
434 | NODEMASK_FREE(unlinked_nodes); | |
c04fc586 | 435 | return -EFAULT; |
9ae49fab DR |
436 | } |
437 | if (!unlinked_nodes) | |
438 | return -ENOMEM; | |
439 | nodes_clear(*unlinked_nodes); | |
d3360164 NF |
440 | |
441 | sect_start_pfn = section_nr_to_pfn(phys_index); | |
c04fc586 GH |
442 | sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1; |
443 | for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) { | |
47504980 | 444 | int nid; |
c04fc586 GH |
445 | |
446 | nid = get_nid_for_pfn(pfn); | |
447 | if (nid < 0) | |
448 | continue; | |
449 | if (!node_online(nid)) | |
450 | continue; | |
9ae49fab | 451 | if (node_test_and_set(nid, *unlinked_nodes)) |
c04fc586 | 452 | continue; |
8732794b | 453 | sysfs_remove_link(&node_devices[nid]->dev.kobj, |
10fbcf4c KS |
454 | kobject_name(&mem_blk->dev.kobj)); |
455 | sysfs_remove_link(&mem_blk->dev.kobj, | |
8732794b | 456 | kobject_name(&node_devices[nid]->dev.kobj)); |
c04fc586 | 457 | } |
9ae49fab | 458 | NODEMASK_FREE(unlinked_nodes); |
c04fc586 GH |
459 | return 0; |
460 | } | |
461 | ||
462 | static int link_mem_sections(int nid) | |
463 | { | |
464 | unsigned long start_pfn = NODE_DATA(nid)->node_start_pfn; | |
465 | unsigned long end_pfn = start_pfn + NODE_DATA(nid)->node_spanned_pages; | |
466 | unsigned long pfn; | |
63d027a6 | 467 | struct memory_block *mem_blk = NULL; |
c04fc586 GH |
468 | int err = 0; |
469 | ||
470 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
471 | unsigned long section_nr = pfn_to_section_nr(pfn); | |
472 | struct mem_section *mem_sect; | |
c04fc586 GH |
473 | int ret; |
474 | ||
475 | if (!present_section_nr(section_nr)) | |
476 | continue; | |
477 | mem_sect = __nr_to_section(section_nr); | |
321bf4ed YL |
478 | |
479 | /* same memblock ? */ | |
480 | if (mem_blk) | |
481 | if ((section_nr >= mem_blk->start_section_nr) && | |
482 | (section_nr <= mem_blk->end_section_nr)) | |
483 | continue; | |
484 | ||
63d027a6 | 485 | mem_blk = find_memory_block_hinted(mem_sect, mem_blk); |
321bf4ed | 486 | |
c04fc586 GH |
487 | ret = register_mem_sect_under_node(mem_blk, nid); |
488 | if (!err) | |
489 | err = ret; | |
490 | ||
491 | /* discard ref obtained in find_memory_block() */ | |
c04fc586 | 492 | } |
63d027a6 RH |
493 | |
494 | if (mem_blk) | |
10fbcf4c | 495 | kobject_put(&mem_blk->dev.kobj); |
c04fc586 GH |
496 | return err; |
497 | } | |
4faf8d95 | 498 | |
39da08cb | 499 | #ifdef CONFIG_HUGETLBFS |
4faf8d95 LS |
500 | /* |
501 | * Handle per node hstate attribute [un]registration on transistions | |
502 | * to/from memoryless state. | |
503 | */ | |
39da08cb LS |
504 | static void node_hugetlb_work(struct work_struct *work) |
505 | { | |
506 | struct node *node = container_of(work, struct node, node_work); | |
507 | ||
508 | /* | |
509 | * We only get here when a node transitions to/from memoryless state. | |
510 | * We can detect which transition occurred by examining whether the | |
511 | * node has memory now. hugetlb_register_node() already check this | |
512 | * so we try to register the attributes. If that fails, then the | |
513 | * node has transitioned to memoryless, try to unregister the | |
514 | * attributes. | |
515 | */ | |
516 | if (!hugetlb_register_node(node)) | |
517 | hugetlb_unregister_node(node); | |
518 | } | |
519 | ||
520 | static void init_node_hugetlb_work(int nid) | |
521 | { | |
8732794b | 522 | INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work); |
39da08cb | 523 | } |
4faf8d95 LS |
524 | |
525 | static int node_memory_callback(struct notifier_block *self, | |
526 | unsigned long action, void *arg) | |
527 | { | |
528 | struct memory_notify *mnb = arg; | |
529 | int nid = mnb->status_change_nid; | |
530 | ||
531 | switch (action) { | |
39da08cb LS |
532 | case MEM_ONLINE: |
533 | case MEM_OFFLINE: | |
534 | /* | |
535 | * offload per node hstate [un]registration to a work thread | |
536 | * when transitioning to/from memoryless state. | |
537 | */ | |
4faf8d95 | 538 | if (nid != NUMA_NO_NODE) |
8732794b | 539 | schedule_work(&node_devices[nid]->node_work); |
4faf8d95 | 540 | break; |
39da08cb | 541 | |
4faf8d95 LS |
542 | case MEM_GOING_ONLINE: |
543 | case MEM_GOING_OFFLINE: | |
544 | case MEM_CANCEL_ONLINE: | |
545 | case MEM_CANCEL_OFFLINE: | |
546 | default: | |
547 | break; | |
548 | } | |
549 | ||
550 | return NOTIFY_OK; | |
551 | } | |
39da08cb LS |
552 | #endif /* CONFIG_HUGETLBFS */ |
553 | #else /* !CONFIG_MEMORY_HOTPLUG_SPARSE */ | |
554 | ||
c04fc586 | 555 | static int link_mem_sections(int nid) { return 0; } |
39da08cb | 556 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
4faf8d95 | 557 | |
39da08cb LS |
558 | #if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \ |
559 | !defined(CONFIG_HUGETLBFS) | |
4faf8d95 LS |
560 | static inline int node_memory_callback(struct notifier_block *self, |
561 | unsigned long action, void *arg) | |
562 | { | |
563 | return NOTIFY_OK; | |
564 | } | |
39da08cb LS |
565 | |
566 | static void init_node_hugetlb_work(int nid) { } | |
567 | ||
568 | #endif | |
c04fc586 | 569 | |
0fc44159 YG |
570 | int register_one_node(int nid) |
571 | { | |
572 | int error = 0; | |
76b67ed9 | 573 | int cpu; |
0fc44159 YG |
574 | |
575 | if (node_online(nid)) { | |
576 | int p_node = parent_node(nid); | |
577 | struct node *parent = NULL; | |
578 | ||
579 | if (p_node != nid) | |
8732794b | 580 | parent = node_devices[p_node]; |
0fc44159 | 581 | |
8732794b WC |
582 | node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL); |
583 | if (!node_devices[nid]) | |
584 | return -ENOMEM; | |
585 | ||
586 | error = register_node(node_devices[nid], nid, parent); | |
76b67ed9 KH |
587 | |
588 | /* link cpu under this node */ | |
589 | for_each_present_cpu(cpu) { | |
590 | if (cpu_to_node(cpu) == nid) | |
591 | register_cpu_under_node(cpu, nid); | |
592 | } | |
c04fc586 GH |
593 | |
594 | /* link memory sections under this node */ | |
595 | error = link_mem_sections(nid); | |
39da08cb LS |
596 | |
597 | /* initialize work queue for memory hot plug */ | |
598 | init_node_hugetlb_work(nid); | |
0fc44159 YG |
599 | } |
600 | ||
601 | return error; | |
602 | ||
603 | } | |
604 | ||
605 | void unregister_one_node(int nid) | |
606 | { | |
8732794b | 607 | unregister_node(node_devices[nid]); |
8732794b | 608 | node_devices[nid] = NULL; |
0fc44159 YG |
609 | } |
610 | ||
bde631a5 LS |
611 | /* |
612 | * node states attributes | |
613 | */ | |
614 | ||
615 | static ssize_t print_nodes_state(enum node_states state, char *buf) | |
616 | { | |
617 | int n; | |
618 | ||
f6238818 RO |
619 | n = nodelist_scnprintf(buf, PAGE_SIZE-2, node_states[state]); |
620 | buf[n++] = '\n'; | |
621 | buf[n] = '\0'; | |
bde631a5 LS |
622 | return n; |
623 | } | |
624 | ||
b15f562f | 625 | struct node_attr { |
10fbcf4c | 626 | struct device_attribute attr; |
b15f562f AK |
627 | enum node_states state; |
628 | }; | |
bde631a5 | 629 | |
10fbcf4c KS |
630 | static ssize_t show_node_state(struct device *dev, |
631 | struct device_attribute *attr, char *buf) | |
bde631a5 | 632 | { |
b15f562f AK |
633 | struct node_attr *na = container_of(attr, struct node_attr, attr); |
634 | return print_nodes_state(na->state, buf); | |
bde631a5 LS |
635 | } |
636 | ||
b15f562f | 637 | #define _NODE_ATTR(name, state) \ |
10fbcf4c | 638 | { __ATTR(name, 0444, show_node_state, NULL), state } |
bde631a5 | 639 | |
b15f562f | 640 | static struct node_attr node_state_attr[] = { |
fcf07d22 LJ |
641 | [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE), |
642 | [N_ONLINE] = _NODE_ATTR(online, N_ONLINE), | |
643 | [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY), | |
bde631a5 | 644 | #ifdef CONFIG_HIGHMEM |
fcf07d22 | 645 | [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY), |
bde631a5 | 646 | #endif |
fcf07d22 | 647 | [N_CPU] = _NODE_ATTR(has_cpu, N_CPU), |
bde631a5 LS |
648 | }; |
649 | ||
10fbcf4c | 650 | static struct attribute *node_state_attrs[] = { |
fcf07d22 LJ |
651 | &node_state_attr[N_POSSIBLE].attr.attr, |
652 | &node_state_attr[N_ONLINE].attr.attr, | |
653 | &node_state_attr[N_NORMAL_MEMORY].attr.attr, | |
3701cde6 | 654 | #ifdef CONFIG_HIGHMEM |
fcf07d22 | 655 | &node_state_attr[N_HIGH_MEMORY].attr.attr, |
3701cde6 | 656 | #endif |
fcf07d22 | 657 | &node_state_attr[N_CPU].attr.attr, |
3701cde6 AK |
658 | NULL |
659 | }; | |
bde631a5 | 660 | |
10fbcf4c KS |
661 | static struct attribute_group memory_root_attr_group = { |
662 | .attrs = node_state_attrs, | |
663 | }; | |
664 | ||
665 | static const struct attribute_group *cpu_root_attr_groups[] = { | |
666 | &memory_root_attr_group, | |
667 | NULL, | |
668 | }; | |
669 | ||
4faf8d95 | 670 | #define NODE_CALLBACK_PRI 2 /* lower than SLAB */ |
4b45099b | 671 | static int __init register_node_type(void) |
1da177e4 | 672 | { |
bde631a5 LS |
673 | int ret; |
674 | ||
3701cde6 AK |
675 | BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES); |
676 | BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES); | |
677 | ||
10fbcf4c | 678 | ret = subsys_system_register(&node_subsys, cpu_root_attr_groups); |
4faf8d95 | 679 | if (!ret) { |
4faf8d95 LS |
680 | hotplug_memory_notifier(node_memory_callback, |
681 | NODE_CALLBACK_PRI); | |
682 | } | |
bde631a5 LS |
683 | |
684 | /* | |
685 | * Note: we're not going to unregister the node class if we fail | |
686 | * to register the node state class attribute files. | |
687 | */ | |
688 | return ret; | |
1da177e4 LT |
689 | } |
690 | postcore_initcall(register_node_type); |