]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - drivers/base/node.c
mm: move most file-based accounting to the node
[mirror_ubuntu-jammy-kernel.git] / drivers / base / node.c
CommitLineData
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 22static struct bus_type node_subsys = {
af5ca3f4 23 .name = "node",
10fbcf4c 24 .dev_name = "node",
1da177e4
LT
25};
26
27
5aaba363 28static 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
39static 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
44static 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
50static DEVICE_ATTR(cpumap, S_IRUGO, node_read_cpumask, NULL);
51static DEVICE_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL);
1da177e4
LT
52
53#define K(x) ((x) << (PAGE_SHIFT - 10))
10fbcf4c
KS
54static 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),
75ef7184 127 nid, sum_zone_node_page_state(nid, NR_KERNEL_STACK) *
c6a7f572 128 THREAD_SIZE / 1024,
75ef7184 129 nid, K(sum_zone_node_page_state(nid, NR_PAGETABLE)),
11fb9989 130 nid, K(node_page_state(pgdat, NR_UNSTABLE_NFS)),
75ef7184 131 nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
11fb9989 132 nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
75ef7184
MG
133 nid, K(sum_zone_node_page_state(nid, NR_SLAB_RECLAIMABLE) +
134 sum_zone_node_page_state(nid, NR_SLAB_UNRECLAIMABLE)),
135 nid, K(sum_zone_node_page_state(nid, NR_SLAB_RECLAIMABLE)),
05b258e9 136#ifdef CONFIG_TRANSPARENT_HUGEPAGE
75ef7184 137 nid, K(sum_zone_node_page_state(nid, NR_SLAB_UNRECLAIMABLE)),
11fb9989 138 nid, K(node_page_state(pgdat, NR_ANON_THPS) *
65c45377 139 HPAGE_PMD_NR),
11fb9989 140 nid, K(node_page_state(pgdat, NR_SHMEM_THPS) *
65c45377 141 HPAGE_PMD_NR),
11fb9989 142 nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) *
65c45377 143 HPAGE_PMD_NR));
91a13c28 144#else
75ef7184 145 nid, K(sum_zone_node_page_state(nid, NR_SLAB_UNRECLAIMABLE)));
05b258e9 146#endif
1da177e4
LT
147 n += hugetlb_report_node_meminfo(nid, buf + n);
148 return n;
149}
150
151#undef K
10fbcf4c 152static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
1da177e4 153
10fbcf4c
KS
154static ssize_t node_read_numastat(struct device *dev,
155 struct device_attribute *attr, char *buf)
1da177e4 156{
1da177e4
LT
157 return sprintf(buf,
158 "numa_hit %lu\n"
159 "numa_miss %lu\n"
160 "numa_foreign %lu\n"
161 "interleave_hit %lu\n"
162 "local_node %lu\n"
163 "other_node %lu\n",
75ef7184
MG
164 sum_zone_node_page_state(dev->id, NUMA_HIT),
165 sum_zone_node_page_state(dev->id, NUMA_MISS),
166 sum_zone_node_page_state(dev->id, NUMA_FOREIGN),
167 sum_zone_node_page_state(dev->id, NUMA_INTERLEAVE_HIT),
168 sum_zone_node_page_state(dev->id, NUMA_LOCAL),
169 sum_zone_node_page_state(dev->id, NUMA_OTHER));
1da177e4 170}
10fbcf4c 171static DEVICE_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
1da177e4 172
10fbcf4c
KS
173static ssize_t node_read_vmstat(struct device *dev,
174 struct device_attribute *attr, char *buf)
2ac39037
MR
175{
176 int nid = dev->id;
75ef7184 177 struct pglist_data *pgdat = NODE_DATA(nid);
fa25c503
KM
178 int i;
179 int n = 0;
180
181 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
182 n += sprintf(buf+n, "%s %lu\n", vmstat_text[i],
75ef7184
MG
183 sum_zone_node_page_state(nid, i));
184
185 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
186 n += sprintf(buf+n, "%s %lu\n",
187 vmstat_text[i + NR_VM_ZONE_STAT_ITEMS],
188 node_page_state(pgdat, i));
fa25c503
KM
189
190 return n;
2ac39037 191}
10fbcf4c 192static DEVICE_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL);
2ac39037 193
10fbcf4c 194static ssize_t node_read_distance(struct device *dev,
518d3f38 195 struct device_attribute *attr, char *buf)
1da177e4
LT
196{
197 int nid = dev->id;
198 int len = 0;
199 int i;
200
12ee3c0a
DR
201 /*
202 * buf is currently PAGE_SIZE in length and each node needs 4 chars
203 * at the most (distance + space or newline).
204 */
205 BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
1da177e4
LT
206
207 for_each_online_node(i)
208 len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i));
209
210 len += sprintf(buf + len, "\n");
211 return len;
212}
10fbcf4c 213static DEVICE_ATTR(distance, S_IRUGO, node_read_distance, NULL);
1da177e4 214
3c9b8aaf
TI
215static struct attribute *node_dev_attrs[] = {
216 &dev_attr_cpumap.attr,
217 &dev_attr_cpulist.attr,
218 &dev_attr_meminfo.attr,
219 &dev_attr_numastat.attr,
220 &dev_attr_distance.attr,
221 &dev_attr_vmstat.attr,
222 NULL
223};
7ca7ec40 224ATTRIBUTE_GROUPS(node_dev);
3c9b8aaf 225
9a305230
LS
226#ifdef CONFIG_HUGETLBFS
227/*
228 * hugetlbfs per node attributes registration interface:
229 * When/if hugetlb[fs] subsystem initializes [sometime after this module],
4faf8d95
LS
230 * it will register its per node attributes for all online nodes with
231 * memory. It will also call register_hugetlbfs_with_node(), below, to
9a305230
LS
232 * register its attribute registration functions with this node driver.
233 * Once these hooks have been initialized, the node driver will call into
234 * the hugetlb module to [un]register attributes for hot-plugged nodes.
235 */
236static node_registration_func_t __hugetlb_register_node;
237static node_registration_func_t __hugetlb_unregister_node;
238
39da08cb 239static inline bool hugetlb_register_node(struct node *node)
9a305230 240{
4faf8d95 241 if (__hugetlb_register_node &&
8cebfcd0 242 node_state(node->dev.id, N_MEMORY)) {
9a305230 243 __hugetlb_register_node(node);
39da08cb
LS
244 return true;
245 }
246 return false;
9a305230
LS
247}
248
249static inline void hugetlb_unregister_node(struct node *node)
250{
251 if (__hugetlb_unregister_node)
252 __hugetlb_unregister_node(node);
253}
254
255void register_hugetlbfs_with_node(node_registration_func_t doregister,
256 node_registration_func_t unregister)
257{
258 __hugetlb_register_node = doregister;
259 __hugetlb_unregister_node = unregister;
260}
261#else
262static inline void hugetlb_register_node(struct node *node) {}
263
264static inline void hugetlb_unregister_node(struct node *node) {}
265#endif
266
8c7b5b4e
YI
267static void node_device_release(struct device *dev)
268{
269 struct node *node = to_node(dev);
270
271#if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
272 /*
273 * We schedule the work only when a memory section is
274 * onlined/offlined on this node. When we come here,
275 * all the memory on this node has been offlined,
276 * so we won't enqueue new work to this work.
277 *
278 * The work is using node->node_work, so we should
279 * flush work before freeing the memory.
280 */
281 flush_work(&node->node_work);
282#endif
283 kfree(node);
284}
1da177e4
LT
285
286/*
405ae7d3 287 * register_node - Setup a sysfs device for a node.
1da177e4
LT
288 * @num - Node number to use when creating the device.
289 *
290 * Initialize and register the node device.
291 */
fa264375 292static int register_node(struct node *node, int num, struct node *parent)
1da177e4
LT
293{
294 int error;
295
10fbcf4c
KS
296 node->dev.id = num;
297 node->dev.bus = &node_subsys;
8c7b5b4e 298 node->dev.release = node_device_release;
7ca7ec40 299 node->dev.groups = node_dev_groups;
10fbcf4c 300 error = device_register(&node->dev);
1da177e4
LT
301
302 if (!error){
9a305230 303 hugetlb_register_node(node);
ed4a6d7f
MG
304
305 compaction_register_node(node);
1da177e4
LT
306 }
307 return error;
308}
309
4b45099b
KT
310/**
311 * unregister_node - unregister a node device
312 * @node: node going away
313 *
314 * Unregisters a node device @node. All the devices on the node must be
315 * unregistered before calling this function.
316 */
317void unregister_node(struct node *node)
318{
4faf8d95 319 hugetlb_unregister_node(node); /* no-op, if memoryless node */
af936a16 320
10fbcf4c 321 device_unregister(&node->dev);
4b45099b 322}
1da177e4 323
8732794b 324struct node *node_devices[MAX_NUMNODES];
0fc44159 325
76b67ed9
KH
326/*
327 * register cpu under node
328 */
329int register_cpu_under_node(unsigned int cpu, unsigned int nid)
330{
1830794a 331 int ret;
8a25a2fd 332 struct device *obj;
76b67ed9 333
f8246f31
AC
334 if (!node_online(nid))
335 return 0;
336
8a25a2fd 337 obj = get_cpu_device(cpu);
f8246f31
AC
338 if (!obj)
339 return 0;
340
8732794b 341 ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
f8246f31
AC
342 &obj->kobj,
343 kobject_name(&obj->kobj));
1830794a
AC
344 if (ret)
345 return ret;
346
347 return sysfs_create_link(&obj->kobj,
8732794b
WC
348 &node_devices[nid]->dev.kobj,
349 kobject_name(&node_devices[nid]->dev.kobj));
76b67ed9
KH
350}
351
352int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
353{
8a25a2fd 354 struct device *obj;
b9d52dad
AC
355
356 if (!node_online(nid))
357 return 0;
358
8a25a2fd 359 obj = get_cpu_device(cpu);
b9d52dad
AC
360 if (!obj)
361 return 0;
362
8732794b 363 sysfs_remove_link(&node_devices[nid]->dev.kobj,
b9d52dad 364 kobject_name(&obj->kobj));
1830794a 365 sysfs_remove_link(&obj->kobj,
8732794b 366 kobject_name(&node_devices[nid]->dev.kobj));
b9d52dad 367
76b67ed9
KH
368 return 0;
369}
370
c04fc586
GH
371#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
372#define page_initialized(page) (page->lru.next)
373
3a80a7fa 374static int __init_refok get_nid_for_pfn(unsigned long pfn)
c04fc586
GH
375{
376 struct page *page;
377
378 if (!pfn_valid_within(pfn))
379 return -1;
3a80a7fa
MG
380#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
381 if (system_state == SYSTEM_BOOTING)
382 return early_pfn_to_nid(pfn);
383#endif
c04fc586
GH
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 */
391int 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
04697858
YL
407 /*
408 * memory block could have several absent sections from start.
409 * skip pfn range from absent section
410 */
411 if (!pfn_present(pfn)) {
412 pfn = round_down(pfn + PAGES_PER_SECTION,
413 PAGES_PER_SECTION) - 1;
414 continue;
415 }
416
c04fc586
GH
417 page_nid = get_nid_for_pfn(pfn);
418 if (page_nid < 0)
419 continue;
420 if (page_nid != nid)
421 continue;
8732794b 422 ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
10fbcf4c
KS
423 &mem_blk->dev.kobj,
424 kobject_name(&mem_blk->dev.kobj));
dee5d0d5
AC
425 if (ret)
426 return ret;
427
10fbcf4c 428 return sysfs_create_link_nowarn(&mem_blk->dev.kobj,
8732794b
WC
429 &node_devices[nid]->dev.kobj,
430 kobject_name(&node_devices[nid]->dev.kobj));
c04fc586
GH
431 }
432 /* mem section does not span the specified node */
433 return 0;
434}
435
436/* unregister memory section under all nodes that it spans */
d3360164
NF
437int unregister_mem_sect_under_nodes(struct memory_block *mem_blk,
438 unsigned long phys_index)
c04fc586 439{
9ae49fab 440 NODEMASK_ALLOC(nodemask_t, unlinked_nodes, GFP_KERNEL);
c04fc586
GH
441 unsigned long pfn, sect_start_pfn, sect_end_pfn;
442
9ae49fab
DR
443 if (!mem_blk) {
444 NODEMASK_FREE(unlinked_nodes);
c04fc586 445 return -EFAULT;
9ae49fab
DR
446 }
447 if (!unlinked_nodes)
448 return -ENOMEM;
449 nodes_clear(*unlinked_nodes);
d3360164
NF
450
451 sect_start_pfn = section_nr_to_pfn(phys_index);
c04fc586
GH
452 sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
453 for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
47504980 454 int nid;
c04fc586
GH
455
456 nid = get_nid_for_pfn(pfn);
457 if (nid < 0)
458 continue;
459 if (!node_online(nid))
460 continue;
9ae49fab 461 if (node_test_and_set(nid, *unlinked_nodes))
c04fc586 462 continue;
8732794b 463 sysfs_remove_link(&node_devices[nid]->dev.kobj,
10fbcf4c
KS
464 kobject_name(&mem_blk->dev.kobj));
465 sysfs_remove_link(&mem_blk->dev.kobj,
8732794b 466 kobject_name(&node_devices[nid]->dev.kobj));
c04fc586 467 }
9ae49fab 468 NODEMASK_FREE(unlinked_nodes);
c04fc586
GH
469 return 0;
470}
471
472static int link_mem_sections(int nid)
473{
474 unsigned long start_pfn = NODE_DATA(nid)->node_start_pfn;
475 unsigned long end_pfn = start_pfn + NODE_DATA(nid)->node_spanned_pages;
476 unsigned long pfn;
63d027a6 477 struct memory_block *mem_blk = NULL;
c04fc586
GH
478 int err = 0;
479
480 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
481 unsigned long section_nr = pfn_to_section_nr(pfn);
482 struct mem_section *mem_sect;
c04fc586
GH
483 int ret;
484
485 if (!present_section_nr(section_nr))
486 continue;
487 mem_sect = __nr_to_section(section_nr);
321bf4ed
YL
488
489 /* same memblock ? */
490 if (mem_blk)
491 if ((section_nr >= mem_blk->start_section_nr) &&
492 (section_nr <= mem_blk->end_section_nr))
493 continue;
494
63d027a6 495 mem_blk = find_memory_block_hinted(mem_sect, mem_blk);
321bf4ed 496
c04fc586
GH
497 ret = register_mem_sect_under_node(mem_blk, nid);
498 if (!err)
499 err = ret;
500
501 /* discard ref obtained in find_memory_block() */
c04fc586 502 }
63d027a6
RH
503
504 if (mem_blk)
10fbcf4c 505 kobject_put(&mem_blk->dev.kobj);
c04fc586
GH
506 return err;
507}
4faf8d95 508
39da08cb 509#ifdef CONFIG_HUGETLBFS
4faf8d95
LS
510/*
511 * Handle per node hstate attribute [un]registration on transistions
512 * to/from memoryless state.
513 */
39da08cb
LS
514static void node_hugetlb_work(struct work_struct *work)
515{
516 struct node *node = container_of(work, struct node, node_work);
517
518 /*
519 * We only get here when a node transitions to/from memoryless state.
520 * We can detect which transition occurred by examining whether the
521 * node has memory now. hugetlb_register_node() already check this
522 * so we try to register the attributes. If that fails, then the
523 * node has transitioned to memoryless, try to unregister the
524 * attributes.
525 */
526 if (!hugetlb_register_node(node))
527 hugetlb_unregister_node(node);
528}
529
530static void init_node_hugetlb_work(int nid)
531{
8732794b 532 INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
39da08cb 533}
4faf8d95
LS
534
535static int node_memory_callback(struct notifier_block *self,
536 unsigned long action, void *arg)
537{
538 struct memory_notify *mnb = arg;
539 int nid = mnb->status_change_nid;
540
541 switch (action) {
39da08cb
LS
542 case MEM_ONLINE:
543 case MEM_OFFLINE:
544 /*
545 * offload per node hstate [un]registration to a work thread
546 * when transitioning to/from memoryless state.
547 */
4faf8d95 548 if (nid != NUMA_NO_NODE)
8732794b 549 schedule_work(&node_devices[nid]->node_work);
4faf8d95 550 break;
39da08cb 551
4faf8d95
LS
552 case MEM_GOING_ONLINE:
553 case MEM_GOING_OFFLINE:
554 case MEM_CANCEL_ONLINE:
555 case MEM_CANCEL_OFFLINE:
556 default:
557 break;
558 }
559
560 return NOTIFY_OK;
561}
39da08cb
LS
562#endif /* CONFIG_HUGETLBFS */
563#else /* !CONFIG_MEMORY_HOTPLUG_SPARSE */
564
c04fc586 565static int link_mem_sections(int nid) { return 0; }
39da08cb 566#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
4faf8d95 567
39da08cb
LS
568#if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
569 !defined(CONFIG_HUGETLBFS)
4faf8d95
LS
570static inline int node_memory_callback(struct notifier_block *self,
571 unsigned long action, void *arg)
572{
573 return NOTIFY_OK;
574}
39da08cb
LS
575
576static void init_node_hugetlb_work(int nid) { }
577
578#endif
c04fc586 579
0fc44159
YG
580int register_one_node(int nid)
581{
582 int error = 0;
76b67ed9 583 int cpu;
0fc44159
YG
584
585 if (node_online(nid)) {
586 int p_node = parent_node(nid);
587 struct node *parent = NULL;
588
589 if (p_node != nid)
8732794b 590 parent = node_devices[p_node];
0fc44159 591
8732794b
WC
592 node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
593 if (!node_devices[nid])
594 return -ENOMEM;
595
596 error = register_node(node_devices[nid], nid, parent);
76b67ed9
KH
597
598 /* link cpu under this node */
599 for_each_present_cpu(cpu) {
600 if (cpu_to_node(cpu) == nid)
601 register_cpu_under_node(cpu, nid);
602 }
c04fc586
GH
603
604 /* link memory sections under this node */
605 error = link_mem_sections(nid);
39da08cb
LS
606
607 /* initialize work queue for memory hot plug */
608 init_node_hugetlb_work(nid);
0fc44159
YG
609 }
610
611 return error;
612
613}
614
615void unregister_one_node(int nid)
616{
92d585ef
XQ
617 if (!node_devices[nid])
618 return;
619
8732794b 620 unregister_node(node_devices[nid]);
8732794b 621 node_devices[nid] = NULL;
0fc44159
YG
622}
623
bde631a5
LS
624/*
625 * node states attributes
626 */
627
628static ssize_t print_nodes_state(enum node_states state, char *buf)
629{
630 int n;
631
f799b1a7
TH
632 n = scnprintf(buf, PAGE_SIZE - 1, "%*pbl",
633 nodemask_pr_args(&node_states[state]));
f6238818
RO
634 buf[n++] = '\n';
635 buf[n] = '\0';
bde631a5
LS
636 return n;
637}
638
b15f562f 639struct node_attr {
10fbcf4c 640 struct device_attribute attr;
b15f562f
AK
641 enum node_states state;
642};
bde631a5 643
10fbcf4c
KS
644static ssize_t show_node_state(struct device *dev,
645 struct device_attribute *attr, char *buf)
bde631a5 646{
b15f562f
AK
647 struct node_attr *na = container_of(attr, struct node_attr, attr);
648 return print_nodes_state(na->state, buf);
bde631a5
LS
649}
650
b15f562f 651#define _NODE_ATTR(name, state) \
10fbcf4c 652 { __ATTR(name, 0444, show_node_state, NULL), state }
bde631a5 653
b15f562f 654static struct node_attr node_state_attr[] = {
fcf07d22
LJ
655 [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
656 [N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
657 [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
bde631a5 658#ifdef CONFIG_HIGHMEM
fcf07d22 659 [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
20b2f52b
LJ
660#endif
661#ifdef CONFIG_MOVABLE_NODE
662 [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
bde631a5 663#endif
fcf07d22 664 [N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
bde631a5
LS
665};
666
10fbcf4c 667static struct attribute *node_state_attrs[] = {
fcf07d22
LJ
668 &node_state_attr[N_POSSIBLE].attr.attr,
669 &node_state_attr[N_ONLINE].attr.attr,
670 &node_state_attr[N_NORMAL_MEMORY].attr.attr,
3701cde6 671#ifdef CONFIG_HIGHMEM
fcf07d22 672 &node_state_attr[N_HIGH_MEMORY].attr.attr,
20b2f52b
LJ
673#endif
674#ifdef CONFIG_MOVABLE_NODE
675 &node_state_attr[N_MEMORY].attr.attr,
3701cde6 676#endif
fcf07d22 677 &node_state_attr[N_CPU].attr.attr,
3701cde6
AK
678 NULL
679};
bde631a5 680
10fbcf4c
KS
681static struct attribute_group memory_root_attr_group = {
682 .attrs = node_state_attrs,
683};
684
685static const struct attribute_group *cpu_root_attr_groups[] = {
686 &memory_root_attr_group,
687 NULL,
688};
689
4faf8d95 690#define NODE_CALLBACK_PRI 2 /* lower than SLAB */
4b45099b 691static int __init register_node_type(void)
1da177e4 692{
bde631a5
LS
693 int ret;
694
3701cde6
AK
695 BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
696 BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
697
10fbcf4c 698 ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
4faf8d95 699 if (!ret) {
6e259e7d
AM
700 static struct notifier_block node_memory_callback_nb = {
701 .notifier_call = node_memory_callback,
702 .priority = NODE_CALLBACK_PRI,
703 };
704 register_hotmemory_notifier(&node_memory_callback_nb);
4faf8d95 705 }
bde631a5
LS
706
707 /*
708 * Note: we're not going to unregister the node class if we fail
709 * to register the node state class attribute files.
710 */
711 return ret;
1da177e4
LT
712}
713postcore_initcall(register_node_type);