]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 | 2 | /* |
10fbcf4c | 3 | * Basic Node interface support |
1da177e4 LT |
4 | */ |
5 | ||
1da177e4 LT |
6 | #include <linux/module.h> |
7 | #include <linux/init.h> | |
8 | #include <linux/mm.h> | |
c04fc586 | 9 | #include <linux/memory.h> |
fa25c503 | 10 | #include <linux/vmstat.h> |
6e259e7d | 11 | #include <linux/notifier.h> |
1da177e4 LT |
12 | #include <linux/node.h> |
13 | #include <linux/hugetlb.h> | |
ed4a6d7f | 14 | #include <linux/compaction.h> |
1da177e4 LT |
15 | #include <linux/cpumask.h> |
16 | #include <linux/topology.h> | |
17 | #include <linux/nodemask.h> | |
76b67ed9 | 18 | #include <linux/cpu.h> |
bde631a5 | 19 | #include <linux/device.h> |
af936a16 | 20 | #include <linux/swap.h> |
18e5b539 | 21 | #include <linux/slab.h> |
1da177e4 | 22 | |
10fbcf4c | 23 | static struct bus_type node_subsys = { |
af5ca3f4 | 24 | .name = "node", |
10fbcf4c | 25 | .dev_name = "node", |
1da177e4 LT |
26 | }; |
27 | ||
28 | ||
5aaba363 | 29 | static ssize_t node_read_cpumap(struct device *dev, bool list, char *buf) |
1da177e4 | 30 | { |
064f0e93 ZL |
31 | ssize_t n; |
32 | cpumask_var_t mask; | |
1da177e4 | 33 | struct node *node_dev = to_node(dev); |
1da177e4 | 34 | |
39106dcf MT |
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)); | |
1da177e4 | 37 | |
064f0e93 ZL |
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; | |
1da177e4 LT |
46 | } |
47 | ||
10fbcf4c KS |
48 | static inline ssize_t node_read_cpumask(struct device *dev, |
49 | struct device_attribute *attr, char *buf) | |
39106dcf | 50 | { |
5aaba363 | 51 | return node_read_cpumap(dev, false, buf); |
39106dcf | 52 | } |
10fbcf4c KS |
53 | static inline ssize_t node_read_cpulist(struct device *dev, |
54 | struct device_attribute *attr, char *buf) | |
39106dcf | 55 | { |
5aaba363 | 56 | return node_read_cpumap(dev, true, buf); |
39106dcf MT |
57 | } |
58 | ||
10fbcf4c KS |
59 | static DEVICE_ATTR(cpumap, S_IRUGO, node_read_cpumask, NULL); |
60 | static DEVICE_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL); | |
1da177e4 LT |
61 | |
62 | #define K(x) ((x) << (PAGE_SHIFT - 10)) | |
10fbcf4c KS |
63 | static ssize_t node_read_meminfo(struct device *dev, |
64 | struct device_attribute *attr, char *buf) | |
1da177e4 LT |
65 | { |
66 | int n; | |
67 | int nid = dev->id; | |
599d0c95 | 68 | struct pglist_data *pgdat = NODE_DATA(nid); |
1da177e4 | 69 | struct sysinfo i; |
61f94e18 | 70 | unsigned long sreclaimable, sunreclaimable; |
1da177e4 LT |
71 | |
72 | si_meminfo_node(&i, nid); | |
61f94e18 VB |
73 | sreclaimable = node_page_state(pgdat, NR_SLAB_RECLAIMABLE); |
74 | sunreclaimable = node_page_state(pgdat, NR_SLAB_UNRECLAIMABLE); | |
7ee92255 | 75 | n = sprintf(buf, |
4f98a2fe RR |
76 | "Node %d MemTotal: %8lu kB\n" |
77 | "Node %d MemFree: %8lu kB\n" | |
78 | "Node %d MemUsed: %8lu kB\n" | |
79 | "Node %d Active: %8lu kB\n" | |
80 | "Node %d Inactive: %8lu kB\n" | |
81 | "Node %d Active(anon): %8lu kB\n" | |
82 | "Node %d Inactive(anon): %8lu kB\n" | |
83 | "Node %d Active(file): %8lu kB\n" | |
84 | "Node %d Inactive(file): %8lu kB\n" | |
5344b7e6 | 85 | "Node %d Unevictable: %8lu kB\n" |
7ee92255 KM |
86 | "Node %d Mlocked: %8lu kB\n", |
87 | nid, K(i.totalram), | |
88 | nid, K(i.freeram), | |
89 | nid, K(i.totalram - i.freeram), | |
599d0c95 MG |
90 | nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) + |
91 | node_page_state(pgdat, NR_ACTIVE_FILE)), | |
92 | nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) + | |
93 | node_page_state(pgdat, NR_INACTIVE_FILE)), | |
94 | nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)), | |
95 | nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)), | |
96 | nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)), | |
97 | nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)), | |
98 | nid, K(node_page_state(pgdat, NR_UNEVICTABLE)), | |
75ef7184 | 99 | nid, K(sum_zone_node_page_state(nid, NR_MLOCK))); |
7ee92255 | 100 | |
182e8e23 | 101 | #ifdef CONFIG_HIGHMEM |
7ee92255 | 102 | n += sprintf(buf + n, |
4f98a2fe RR |
103 | "Node %d HighTotal: %8lu kB\n" |
104 | "Node %d HighFree: %8lu kB\n" | |
105 | "Node %d LowTotal: %8lu kB\n" | |
7ee92255 KM |
106 | "Node %d LowFree: %8lu kB\n", |
107 | nid, K(i.totalhigh), | |
108 | nid, K(i.freehigh), | |
109 | nid, K(i.totalram - i.totalhigh), | |
110 | nid, K(i.freeram - i.freehigh)); | |
182e8e23 | 111 | #endif |
7ee92255 | 112 | n += sprintf(buf + n, |
4f98a2fe RR |
113 | "Node %d Dirty: %8lu kB\n" |
114 | "Node %d Writeback: %8lu kB\n" | |
115 | "Node %d FilePages: %8lu kB\n" | |
116 | "Node %d Mapped: %8lu kB\n" | |
117 | "Node %d AnonPages: %8lu kB\n" | |
4b02108a | 118 | "Node %d Shmem: %8lu kB\n" |
c6a7f572 | 119 | "Node %d KernelStack: %8lu kB\n" |
4f98a2fe RR |
120 | "Node %d PageTables: %8lu kB\n" |
121 | "Node %d NFS_Unstable: %8lu kB\n" | |
122 | "Node %d Bounce: %8lu kB\n" | |
123 | "Node %d WritebackTmp: %8lu kB\n" | |
61f94e18 | 124 | "Node %d KReclaimable: %8lu kB\n" |
4f98a2fe RR |
125 | "Node %d Slab: %8lu kB\n" |
126 | "Node %d SReclaimable: %8lu kB\n" | |
05b258e9 DR |
127 | "Node %d SUnreclaim: %8lu kB\n" |
128 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
129 | "Node %d AnonHugePages: %8lu kB\n" | |
65c45377 KS |
130 | "Node %d ShmemHugePages: %8lu kB\n" |
131 | "Node %d ShmemPmdMapped: %8lu kB\n" | |
05b258e9 DR |
132 | #endif |
133 | , | |
11fb9989 MG |
134 | nid, K(node_page_state(pgdat, NR_FILE_DIRTY)), |
135 | nid, K(node_page_state(pgdat, NR_WRITEBACK)), | |
136 | nid, K(node_page_state(pgdat, NR_FILE_PAGES)), | |
50658e2e | 137 | nid, K(node_page_state(pgdat, NR_FILE_MAPPED)), |
4b9d0fab | 138 | nid, K(node_page_state(pgdat, NR_ANON_MAPPED)), |
cc7452b6 | 139 | nid, K(i.sharedram), |
d30dd8be | 140 | nid, sum_zone_node_page_state(nid, NR_KERNEL_STACK_KB), |
75ef7184 | 141 | nid, K(sum_zone_node_page_state(nid, NR_PAGETABLE)), |
11fb9989 | 142 | nid, K(node_page_state(pgdat, NR_UNSTABLE_NFS)), |
75ef7184 | 143 | nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)), |
11fb9989 | 144 | nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)), |
61f94e18 VB |
145 | nid, K(sreclaimable + |
146 | node_page_state(pgdat, NR_KERNEL_MISC_RECLAIMABLE)), | |
147 | nid, K(sreclaimable + sunreclaimable), | |
148 | nid, K(sreclaimable), | |
149 | nid, K(sunreclaimable) | |
05b258e9 | 150 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
61f94e18 | 151 | , |
11fb9989 | 152 | nid, K(node_page_state(pgdat, NR_ANON_THPS) * |
65c45377 | 153 | HPAGE_PMD_NR), |
11fb9989 | 154 | nid, K(node_page_state(pgdat, NR_SHMEM_THPS) * |
65c45377 | 155 | HPAGE_PMD_NR), |
11fb9989 | 156 | nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) * |
61f94e18 | 157 | HPAGE_PMD_NR) |
05b258e9 | 158 | #endif |
61f94e18 | 159 | ); |
1da177e4 LT |
160 | n += hugetlb_report_node_meminfo(nid, buf + n); |
161 | return n; | |
162 | } | |
163 | ||
164 | #undef K | |
10fbcf4c | 165 | static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL); |
1da177e4 | 166 | |
10fbcf4c KS |
167 | static ssize_t node_read_numastat(struct device *dev, |
168 | struct device_attribute *attr, char *buf) | |
1da177e4 | 169 | { |
1da177e4 LT |
170 | return sprintf(buf, |
171 | "numa_hit %lu\n" | |
172 | "numa_miss %lu\n" | |
173 | "numa_foreign %lu\n" | |
174 | "interleave_hit %lu\n" | |
175 | "local_node %lu\n" | |
176 | "other_node %lu\n", | |
3a321d2a KW |
177 | sum_zone_numa_state(dev->id, NUMA_HIT), |
178 | sum_zone_numa_state(dev->id, NUMA_MISS), | |
179 | sum_zone_numa_state(dev->id, NUMA_FOREIGN), | |
180 | sum_zone_numa_state(dev->id, NUMA_INTERLEAVE_HIT), | |
181 | sum_zone_numa_state(dev->id, NUMA_LOCAL), | |
182 | sum_zone_numa_state(dev->id, NUMA_OTHER)); | |
1da177e4 | 183 | } |
10fbcf4c | 184 | static DEVICE_ATTR(numastat, S_IRUGO, node_read_numastat, NULL); |
1da177e4 | 185 | |
10fbcf4c KS |
186 | static ssize_t node_read_vmstat(struct device *dev, |
187 | struct device_attribute *attr, char *buf) | |
2ac39037 MR |
188 | { |
189 | int nid = dev->id; | |
75ef7184 | 190 | struct pglist_data *pgdat = NODE_DATA(nid); |
fa25c503 KM |
191 | int i; |
192 | int n = 0; | |
193 | ||
194 | for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) | |
195 | n += sprintf(buf+n, "%s %lu\n", vmstat_text[i], | |
75ef7184 MG |
196 | sum_zone_node_page_state(nid, i)); |
197 | ||
3a321d2a KW |
198 | #ifdef CONFIG_NUMA |
199 | for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++) | |
75ef7184 MG |
200 | n += sprintf(buf+n, "%s %lu\n", |
201 | vmstat_text[i + NR_VM_ZONE_STAT_ITEMS], | |
3a321d2a KW |
202 | sum_zone_numa_state(nid, i)); |
203 | #endif | |
204 | ||
205 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) | |
206 | n += sprintf(buf+n, "%s %lu\n", | |
207 | vmstat_text[i + NR_VM_ZONE_STAT_ITEMS + | |
208 | NR_VM_NUMA_STAT_ITEMS], | |
75ef7184 | 209 | node_page_state(pgdat, i)); |
fa25c503 KM |
210 | |
211 | return n; | |
2ac39037 | 212 | } |
10fbcf4c | 213 | static DEVICE_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL); |
2ac39037 | 214 | |
10fbcf4c | 215 | static ssize_t node_read_distance(struct device *dev, |
518d3f38 | 216 | struct device_attribute *attr, char *buf) |
1da177e4 LT |
217 | { |
218 | int nid = dev->id; | |
219 | int len = 0; | |
220 | int i; | |
221 | ||
12ee3c0a DR |
222 | /* |
223 | * buf is currently PAGE_SIZE in length and each node needs 4 chars | |
224 | * at the most (distance + space or newline). | |
225 | */ | |
226 | BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE); | |
1da177e4 LT |
227 | |
228 | for_each_online_node(i) | |
229 | len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i)); | |
230 | ||
231 | len += sprintf(buf + len, "\n"); | |
232 | return len; | |
233 | } | |
10fbcf4c | 234 | static DEVICE_ATTR(distance, S_IRUGO, node_read_distance, NULL); |
1da177e4 | 235 | |
3c9b8aaf TI |
236 | static struct attribute *node_dev_attrs[] = { |
237 | &dev_attr_cpumap.attr, | |
238 | &dev_attr_cpulist.attr, | |
239 | &dev_attr_meminfo.attr, | |
240 | &dev_attr_numastat.attr, | |
241 | &dev_attr_distance.attr, | |
242 | &dev_attr_vmstat.attr, | |
243 | NULL | |
244 | }; | |
7ca7ec40 | 245 | ATTRIBUTE_GROUPS(node_dev); |
3c9b8aaf | 246 | |
9a305230 LS |
247 | #ifdef CONFIG_HUGETLBFS |
248 | /* | |
249 | * hugetlbfs per node attributes registration interface: | |
250 | * When/if hugetlb[fs] subsystem initializes [sometime after this module], | |
4faf8d95 LS |
251 | * it will register its per node attributes for all online nodes with |
252 | * memory. It will also call register_hugetlbfs_with_node(), below, to | |
9a305230 LS |
253 | * register its attribute registration functions with this node driver. |
254 | * Once these hooks have been initialized, the node driver will call into | |
255 | * the hugetlb module to [un]register attributes for hot-plugged nodes. | |
256 | */ | |
257 | static node_registration_func_t __hugetlb_register_node; | |
258 | static node_registration_func_t __hugetlb_unregister_node; | |
259 | ||
39da08cb | 260 | static inline bool hugetlb_register_node(struct node *node) |
9a305230 | 261 | { |
4faf8d95 | 262 | if (__hugetlb_register_node && |
8cebfcd0 | 263 | node_state(node->dev.id, N_MEMORY)) { |
9a305230 | 264 | __hugetlb_register_node(node); |
39da08cb LS |
265 | return true; |
266 | } | |
267 | return false; | |
9a305230 LS |
268 | } |
269 | ||
270 | static inline void hugetlb_unregister_node(struct node *node) | |
271 | { | |
272 | if (__hugetlb_unregister_node) | |
273 | __hugetlb_unregister_node(node); | |
274 | } | |
275 | ||
276 | void register_hugetlbfs_with_node(node_registration_func_t doregister, | |
277 | node_registration_func_t unregister) | |
278 | { | |
279 | __hugetlb_register_node = doregister; | |
280 | __hugetlb_unregister_node = unregister; | |
281 | } | |
282 | #else | |
283 | static inline void hugetlb_register_node(struct node *node) {} | |
284 | ||
285 | static inline void hugetlb_unregister_node(struct node *node) {} | |
286 | #endif | |
287 | ||
8c7b5b4e YI |
288 | static void node_device_release(struct device *dev) |
289 | { | |
290 | struct node *node = to_node(dev); | |
291 | ||
292 | #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS) | |
293 | /* | |
294 | * We schedule the work only when a memory section is | |
295 | * onlined/offlined on this node. When we come here, | |
296 | * all the memory on this node has been offlined, | |
297 | * so we won't enqueue new work to this work. | |
298 | * | |
299 | * The work is using node->node_work, so we should | |
300 | * flush work before freeing the memory. | |
301 | */ | |
302 | flush_work(&node->node_work); | |
303 | #endif | |
304 | kfree(node); | |
305 | } | |
1da177e4 LT |
306 | |
307 | /* | |
405ae7d3 | 308 | * register_node - Setup a sysfs device for a node. |
1da177e4 LT |
309 | * @num - Node number to use when creating the device. |
310 | * | |
311 | * Initialize and register the node device. | |
312 | */ | |
a7be6e5a | 313 | static int register_node(struct node *node, int num) |
1da177e4 LT |
314 | { |
315 | int error; | |
316 | ||
10fbcf4c KS |
317 | node->dev.id = num; |
318 | node->dev.bus = &node_subsys; | |
8c7b5b4e | 319 | node->dev.release = node_device_release; |
7ca7ec40 | 320 | node->dev.groups = node_dev_groups; |
10fbcf4c | 321 | error = device_register(&node->dev); |
1da177e4 | 322 | |
c1cc0d51 AY |
323 | if (error) |
324 | put_device(&node->dev); | |
325 | else { | |
9a305230 | 326 | hugetlb_register_node(node); |
ed4a6d7f MG |
327 | |
328 | compaction_register_node(node); | |
1da177e4 LT |
329 | } |
330 | return error; | |
331 | } | |
332 | ||
4b45099b KT |
333 | /** |
334 | * unregister_node - unregister a node device | |
335 | * @node: node going away | |
336 | * | |
337 | * Unregisters a node device @node. All the devices on the node must be | |
338 | * unregistered before calling this function. | |
339 | */ | |
340 | void unregister_node(struct node *node) | |
341 | { | |
4faf8d95 | 342 | hugetlb_unregister_node(node); /* no-op, if memoryless node */ |
af936a16 | 343 | |
10fbcf4c | 344 | device_unregister(&node->dev); |
4b45099b | 345 | } |
1da177e4 | 346 | |
8732794b | 347 | struct node *node_devices[MAX_NUMNODES]; |
0fc44159 | 348 | |
76b67ed9 KH |
349 | /* |
350 | * register cpu under node | |
351 | */ | |
352 | int register_cpu_under_node(unsigned int cpu, unsigned int nid) | |
353 | { | |
1830794a | 354 | int ret; |
8a25a2fd | 355 | struct device *obj; |
76b67ed9 | 356 | |
f8246f31 AC |
357 | if (!node_online(nid)) |
358 | return 0; | |
359 | ||
8a25a2fd | 360 | obj = get_cpu_device(cpu); |
f8246f31 AC |
361 | if (!obj) |
362 | return 0; | |
363 | ||
8732794b | 364 | ret = sysfs_create_link(&node_devices[nid]->dev.kobj, |
f8246f31 AC |
365 | &obj->kobj, |
366 | kobject_name(&obj->kobj)); | |
1830794a AC |
367 | if (ret) |
368 | return ret; | |
369 | ||
370 | return sysfs_create_link(&obj->kobj, | |
8732794b WC |
371 | &node_devices[nid]->dev.kobj, |
372 | kobject_name(&node_devices[nid]->dev.kobj)); | |
76b67ed9 KH |
373 | } |
374 | ||
375 | int unregister_cpu_under_node(unsigned int cpu, unsigned int nid) | |
376 | { | |
8a25a2fd | 377 | struct device *obj; |
b9d52dad AC |
378 | |
379 | if (!node_online(nid)) | |
380 | return 0; | |
381 | ||
8a25a2fd | 382 | obj = get_cpu_device(cpu); |
b9d52dad AC |
383 | if (!obj) |
384 | return 0; | |
385 | ||
8732794b | 386 | sysfs_remove_link(&node_devices[nid]->dev.kobj, |
b9d52dad | 387 | kobject_name(&obj->kobj)); |
1830794a | 388 | sysfs_remove_link(&obj->kobj, |
8732794b | 389 | kobject_name(&node_devices[nid]->dev.kobj)); |
b9d52dad | 390 | |
76b67ed9 KH |
391 | return 0; |
392 | } | |
393 | ||
c04fc586 | 394 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
bd721ea7 | 395 | static int __ref get_nid_for_pfn(unsigned long pfn) |
c04fc586 | 396 | { |
c04fc586 GH |
397 | if (!pfn_valid_within(pfn)) |
398 | return -1; | |
3a80a7fa | 399 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
8cdde385 | 400 | if (system_state < SYSTEM_RUNNING) |
3a80a7fa MG |
401 | return early_pfn_to_nid(pfn); |
402 | #endif | |
c04fc586 GH |
403 | return pfn_to_nid(pfn); |
404 | } | |
405 | ||
406 | /* register memory section under specified node if it spans that node */ | |
4fbce633 | 407 | int register_mem_sect_under_node(struct memory_block *mem_blk, void *arg) |
c04fc586 | 408 | { |
4fbce633 | 409 | int ret, nid = *(int *)arg; |
c04fc586 GH |
410 | unsigned long pfn, sect_start_pfn, sect_end_pfn; |
411 | ||
d0dc12e8 | 412 | mem_blk->nid = nid; |
d3360164 NF |
413 | |
414 | sect_start_pfn = section_nr_to_pfn(mem_blk->start_section_nr); | |
415 | sect_end_pfn = section_nr_to_pfn(mem_blk->end_section_nr); | |
416 | sect_end_pfn += PAGES_PER_SECTION - 1; | |
c04fc586 GH |
417 | for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) { |
418 | int page_nid; | |
419 | ||
04697858 YL |
420 | /* |
421 | * memory block could have several absent sections from start. | |
422 | * skip pfn range from absent section | |
423 | */ | |
424 | if (!pfn_present(pfn)) { | |
425 | pfn = round_down(pfn + PAGES_PER_SECTION, | |
426 | PAGES_PER_SECTION) - 1; | |
427 | continue; | |
428 | } | |
429 | ||
fc44f7f9 PT |
430 | /* |
431 | * We need to check if page belongs to nid only for the boot | |
432 | * case, during hotplug we know that all pages in the memory | |
433 | * block belong to the same node. | |
434 | */ | |
4fbce633 | 435 | if (system_state == SYSTEM_BOOTING) { |
fc44f7f9 PT |
436 | page_nid = get_nid_for_pfn(pfn); |
437 | if (page_nid < 0) | |
438 | continue; | |
439 | if (page_nid != nid) | |
440 | continue; | |
441 | } | |
8732794b | 442 | ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj, |
10fbcf4c KS |
443 | &mem_blk->dev.kobj, |
444 | kobject_name(&mem_blk->dev.kobj)); | |
dee5d0d5 AC |
445 | if (ret) |
446 | return ret; | |
447 | ||
10fbcf4c | 448 | return sysfs_create_link_nowarn(&mem_blk->dev.kobj, |
8732794b WC |
449 | &node_devices[nid]->dev.kobj, |
450 | kobject_name(&node_devices[nid]->dev.kobj)); | |
c04fc586 GH |
451 | } |
452 | /* mem section does not span the specified node */ | |
453 | return 0; | |
454 | } | |
455 | ||
456 | /* unregister memory section under all nodes that it spans */ | |
d3360164 NF |
457 | int unregister_mem_sect_under_nodes(struct memory_block *mem_blk, |
458 | unsigned long phys_index) | |
c04fc586 | 459 | { |
9ae49fab | 460 | NODEMASK_ALLOC(nodemask_t, unlinked_nodes, GFP_KERNEL); |
c04fc586 GH |
461 | unsigned long pfn, sect_start_pfn, sect_end_pfn; |
462 | ||
9ae49fab DR |
463 | if (!mem_blk) { |
464 | NODEMASK_FREE(unlinked_nodes); | |
c04fc586 | 465 | return -EFAULT; |
9ae49fab DR |
466 | } |
467 | if (!unlinked_nodes) | |
468 | return -ENOMEM; | |
469 | nodes_clear(*unlinked_nodes); | |
d3360164 NF |
470 | |
471 | sect_start_pfn = section_nr_to_pfn(phys_index); | |
c04fc586 GH |
472 | sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1; |
473 | for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) { | |
47504980 | 474 | int nid; |
c04fc586 GH |
475 | |
476 | nid = get_nid_for_pfn(pfn); | |
477 | if (nid < 0) | |
478 | continue; | |
479 | if (!node_online(nid)) | |
480 | continue; | |
9ae49fab | 481 | if (node_test_and_set(nid, *unlinked_nodes)) |
c04fc586 | 482 | continue; |
8732794b | 483 | sysfs_remove_link(&node_devices[nid]->dev.kobj, |
10fbcf4c KS |
484 | kobject_name(&mem_blk->dev.kobj)); |
485 | sysfs_remove_link(&mem_blk->dev.kobj, | |
8732794b | 486 | kobject_name(&node_devices[nid]->dev.kobj)); |
c04fc586 | 487 | } |
9ae49fab | 488 | NODEMASK_FREE(unlinked_nodes); |
c04fc586 GH |
489 | return 0; |
490 | } | |
491 | ||
4fbce633 | 492 | int link_mem_sections(int nid, unsigned long start_pfn, unsigned long end_pfn) |
c04fc586 | 493 | { |
4fbce633 OS |
494 | return walk_memory_range(start_pfn, end_pfn, (void *)&nid, |
495 | register_mem_sect_under_node); | |
c04fc586 | 496 | } |
4faf8d95 | 497 | |
39da08cb | 498 | #ifdef CONFIG_HUGETLBFS |
4faf8d95 LS |
499 | /* |
500 | * Handle per node hstate attribute [un]registration on transistions | |
501 | * to/from memoryless state. | |
502 | */ | |
39da08cb LS |
503 | static void node_hugetlb_work(struct work_struct *work) |
504 | { | |
505 | struct node *node = container_of(work, struct node, node_work); | |
506 | ||
507 | /* | |
508 | * We only get here when a node transitions to/from memoryless state. | |
509 | * We can detect which transition occurred by examining whether the | |
510 | * node has memory now. hugetlb_register_node() already check this | |
511 | * so we try to register the attributes. If that fails, then the | |
512 | * node has transitioned to memoryless, try to unregister the | |
513 | * attributes. | |
514 | */ | |
515 | if (!hugetlb_register_node(node)) | |
516 | hugetlb_unregister_node(node); | |
517 | } | |
518 | ||
519 | static void init_node_hugetlb_work(int nid) | |
520 | { | |
8732794b | 521 | INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work); |
39da08cb | 522 | } |
4faf8d95 LS |
523 | |
524 | static int node_memory_callback(struct notifier_block *self, | |
525 | unsigned long action, void *arg) | |
526 | { | |
527 | struct memory_notify *mnb = arg; | |
528 | int nid = mnb->status_change_nid; | |
529 | ||
530 | switch (action) { | |
39da08cb LS |
531 | case MEM_ONLINE: |
532 | case MEM_OFFLINE: | |
533 | /* | |
534 | * offload per node hstate [un]registration to a work thread | |
535 | * when transitioning to/from memoryless state. | |
536 | */ | |
4faf8d95 | 537 | if (nid != NUMA_NO_NODE) |
8732794b | 538 | schedule_work(&node_devices[nid]->node_work); |
4faf8d95 | 539 | break; |
39da08cb | 540 | |
4faf8d95 LS |
541 | case MEM_GOING_ONLINE: |
542 | case MEM_GOING_OFFLINE: | |
543 | case MEM_CANCEL_ONLINE: | |
544 | case MEM_CANCEL_OFFLINE: | |
545 | default: | |
546 | break; | |
547 | } | |
548 | ||
549 | return NOTIFY_OK; | |
550 | } | |
39da08cb | 551 | #endif /* CONFIG_HUGETLBFS */ |
9037a993 | 552 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
4faf8d95 | 553 | |
39da08cb LS |
554 | #if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \ |
555 | !defined(CONFIG_HUGETLBFS) | |
4faf8d95 LS |
556 | static inline int node_memory_callback(struct notifier_block *self, |
557 | unsigned long action, void *arg) | |
558 | { | |
559 | return NOTIFY_OK; | |
560 | } | |
39da08cb LS |
561 | |
562 | static void init_node_hugetlb_work(int nid) { } | |
563 | ||
564 | #endif | |
c04fc586 | 565 | |
9037a993 | 566 | int __register_one_node(int nid) |
0fc44159 | 567 | { |
9037a993 | 568 | int error; |
76b67ed9 | 569 | int cpu; |
0fc44159 | 570 | |
9037a993 MH |
571 | node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL); |
572 | if (!node_devices[nid]) | |
573 | return -ENOMEM; | |
c04fc586 | 574 | |
a7be6e5a | 575 | error = register_node(node_devices[nid], nid); |
39da08cb | 576 | |
9037a993 MH |
577 | /* link cpu under this node */ |
578 | for_each_present_cpu(cpu) { | |
579 | if (cpu_to_node(cpu) == nid) | |
580 | register_cpu_under_node(cpu, nid); | |
0fc44159 YG |
581 | } |
582 | ||
9037a993 MH |
583 | /* initialize work queue for memory hot plug */ |
584 | init_node_hugetlb_work(nid); | |
0fc44159 | 585 | |
9037a993 | 586 | return error; |
0fc44159 YG |
587 | } |
588 | ||
589 | void unregister_one_node(int nid) | |
590 | { | |
92d585ef XQ |
591 | if (!node_devices[nid]) |
592 | return; | |
593 | ||
8732794b | 594 | unregister_node(node_devices[nid]); |
8732794b | 595 | node_devices[nid] = NULL; |
0fc44159 YG |
596 | } |
597 | ||
bde631a5 LS |
598 | /* |
599 | * node states attributes | |
600 | */ | |
601 | ||
602 | static ssize_t print_nodes_state(enum node_states state, char *buf) | |
603 | { | |
604 | int n; | |
605 | ||
f799b1a7 TH |
606 | n = scnprintf(buf, PAGE_SIZE - 1, "%*pbl", |
607 | nodemask_pr_args(&node_states[state])); | |
f6238818 RO |
608 | buf[n++] = '\n'; |
609 | buf[n] = '\0'; | |
bde631a5 LS |
610 | return n; |
611 | } | |
612 | ||
b15f562f | 613 | struct node_attr { |
10fbcf4c | 614 | struct device_attribute attr; |
b15f562f AK |
615 | enum node_states state; |
616 | }; | |
bde631a5 | 617 | |
10fbcf4c KS |
618 | static ssize_t show_node_state(struct device *dev, |
619 | struct device_attribute *attr, char *buf) | |
bde631a5 | 620 | { |
b15f562f AK |
621 | struct node_attr *na = container_of(attr, struct node_attr, attr); |
622 | return print_nodes_state(na->state, buf); | |
bde631a5 LS |
623 | } |
624 | ||
b15f562f | 625 | #define _NODE_ATTR(name, state) \ |
10fbcf4c | 626 | { __ATTR(name, 0444, show_node_state, NULL), state } |
bde631a5 | 627 | |
b15f562f | 628 | static struct node_attr node_state_attr[] = { |
fcf07d22 LJ |
629 | [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE), |
630 | [N_ONLINE] = _NODE_ATTR(online, N_ONLINE), | |
631 | [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY), | |
bde631a5 | 632 | #ifdef CONFIG_HIGHMEM |
fcf07d22 | 633 | [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY), |
20b2f52b | 634 | #endif |
20b2f52b | 635 | [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY), |
fcf07d22 | 636 | [N_CPU] = _NODE_ATTR(has_cpu, N_CPU), |
bde631a5 LS |
637 | }; |
638 | ||
10fbcf4c | 639 | static struct attribute *node_state_attrs[] = { |
fcf07d22 LJ |
640 | &node_state_attr[N_POSSIBLE].attr.attr, |
641 | &node_state_attr[N_ONLINE].attr.attr, | |
642 | &node_state_attr[N_NORMAL_MEMORY].attr.attr, | |
3701cde6 | 643 | #ifdef CONFIG_HIGHMEM |
fcf07d22 | 644 | &node_state_attr[N_HIGH_MEMORY].attr.attr, |
20b2f52b | 645 | #endif |
20b2f52b | 646 | &node_state_attr[N_MEMORY].attr.attr, |
fcf07d22 | 647 | &node_state_attr[N_CPU].attr.attr, |
3701cde6 AK |
648 | NULL |
649 | }; | |
bde631a5 | 650 | |
10fbcf4c KS |
651 | static struct attribute_group memory_root_attr_group = { |
652 | .attrs = node_state_attrs, | |
653 | }; | |
654 | ||
655 | static const struct attribute_group *cpu_root_attr_groups[] = { | |
656 | &memory_root_attr_group, | |
657 | NULL, | |
658 | }; | |
659 | ||
4faf8d95 | 660 | #define NODE_CALLBACK_PRI 2 /* lower than SLAB */ |
4b45099b | 661 | static int __init register_node_type(void) |
1da177e4 | 662 | { |
bde631a5 LS |
663 | int ret; |
664 | ||
3701cde6 AK |
665 | BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES); |
666 | BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES); | |
667 | ||
10fbcf4c | 668 | ret = subsys_system_register(&node_subsys, cpu_root_attr_groups); |
4faf8d95 | 669 | if (!ret) { |
6e259e7d AM |
670 | static struct notifier_block node_memory_callback_nb = { |
671 | .notifier_call = node_memory_callback, | |
672 | .priority = NODE_CALLBACK_PRI, | |
673 | }; | |
674 | register_hotmemory_notifier(&node_memory_callback_nb); | |
4faf8d95 | 675 | } |
bde631a5 LS |
676 | |
677 | /* | |
678 | * Note: we're not going to unregister the node class if we fail | |
679 | * to register the node state class attribute files. | |
680 | */ | |
681 | return ret; | |
1da177e4 LT |
682 | } |
683 | postcore_initcall(register_node_type); |