blob: ce012a9c6201ace3fcf143fcf02f424ce0d5c29b [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * drivers/base/node.c - basic Node class support
3 */
4
5#include <linux/sysdev.h>
6#include <linux/module.h>
7#include <linux/init.h>
8#include <linux/mm.h>
Gary Hadec04fc582009-01-06 14:39:14 -08009#include <linux/memory.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070010#include <linux/node.h>
11#include <linux/hugetlb.h>
Mel Gormaned4a6d72010-05-24 14:32:29 -070012#include <linux/compaction.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/cpumask.h>
14#include <linux/topology.h>
15#include <linux/nodemask.h>
KAMEZAWA Hiroyuki76b67ed92006-06-27 02:53:41 -070016#include <linux/cpu.h>
Lee Schermerhornbde631a2007-10-16 01:26:27 -070017#include <linux/device.h>
Lee Schermerhornaf936a12008-10-18 20:26:53 -070018#include <linux/swap.h>
Tejun Heo18e5b532010-04-06 19:23:33 +090019#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020
Andi Kleen3701cde2010-01-05 12:48:04 +010021static struct sysdev_class_attribute *node_state_attrs[];
22
Linus Torvalds1da177e2005-04-16 15:20:36 -070023static struct sysdev_class node_class = {
Kay Sieversaf5ca3f2007-12-20 02:09:39 +010024 .name = "node",
Andi Kleen3701cde2010-01-05 12:48:04 +010025 .attrs = node_state_attrs,
Linus Torvalds1da177e2005-04-16 15:20:36 -070026};
27
28
Mike Travis39106dc2008-04-08 11:43:03 -070029static ssize_t node_read_cpumap(struct sys_device *dev, int type, char *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -070030{
31 struct node *node_dev = to_node(dev);
Rusty Russella70f7302009-03-13 14:49:46 +103032 const struct cpumask *mask = cpumask_of_node(node_dev->sysdev.id);
Linus Torvalds1da177e2005-04-16 15:20:36 -070033 int len;
34
Mike Travis39106dc2008-04-08 11:43:03 -070035 /* 2008/04/07: buf currently PAGE_SIZE, need 9 chars per 32 bits. */
36 BUILD_BUG_ON((NR_CPUS/32 * 9) > (PAGE_SIZE-1));
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
Mike Travis39106dc2008-04-08 11:43:03 -070038 len = type?
Rusty Russell29c01772008-12-13 21:20:25 +103039 cpulist_scnprintf(buf, PAGE_SIZE-2, mask) :
40 cpumask_scnprintf(buf, PAGE_SIZE-2, mask);
Mike Travisc5f59f02008-04-04 18:11:10 -070041 buf[len++] = '\n';
42 buf[len] = '\0';
Linus Torvalds1da177e2005-04-16 15:20:36 -070043 return len;
44}
45
Andi Kleen4a0b2b42008-07-01 18:48:41 +020046static inline ssize_t node_read_cpumask(struct sys_device *dev,
47 struct sysdev_attribute *attr, char *buf)
Mike Travis39106dc2008-04-08 11:43:03 -070048{
49 return node_read_cpumap(dev, 0, buf);
50}
Andi Kleen4a0b2b42008-07-01 18:48:41 +020051static inline ssize_t node_read_cpulist(struct sys_device *dev,
52 struct sysdev_attribute *attr, char *buf)
Mike Travis39106dc2008-04-08 11:43:03 -070053{
54 return node_read_cpumap(dev, 1, buf);
55}
56
57static SYSDEV_ATTR(cpumap, S_IRUGO, node_read_cpumask, NULL);
58static SYSDEV_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
60#define K(x) ((x) << (PAGE_SHIFT - 10))
Andi Kleen4a0b2b42008-07-01 18:48:41 +020061static ssize_t node_read_meminfo(struct sys_device * dev,
62 struct sysdev_attribute *attr, char * buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -070063{
64 int n;
65 int nid = dev->id;
66 struct sysinfo i;
Linus Torvalds1da177e2005-04-16 15:20:36 -070067
68 si_meminfo_node(&i, nid);
KOSAKI Motohiro7ee92252010-08-09 17:19:50 -070069 n = sprintf(buf,
Rik van Riel4f98a2f2008-10-18 20:26:32 -070070 "Node %d MemTotal: %8lu kB\n"
71 "Node %d MemFree: %8lu kB\n"
72 "Node %d MemUsed: %8lu kB\n"
73 "Node %d Active: %8lu kB\n"
74 "Node %d Inactive: %8lu kB\n"
75 "Node %d Active(anon): %8lu kB\n"
76 "Node %d Inactive(anon): %8lu kB\n"
77 "Node %d Active(file): %8lu kB\n"
78 "Node %d Inactive(file): %8lu kB\n"
Nick Piggin5344b7e2008-10-18 20:26:51 -070079 "Node %d Unevictable: %8lu kB\n"
KOSAKI Motohiro7ee92252010-08-09 17:19:50 -070080 "Node %d Mlocked: %8lu kB\n",
81 nid, K(i.totalram),
82 nid, K(i.freeram),
83 nid, K(i.totalram - i.freeram),
84 nid, K(node_page_state(nid, NR_ACTIVE_ANON) +
85 node_page_state(nid, NR_ACTIVE_FILE)),
86 nid, K(node_page_state(nid, NR_INACTIVE_ANON) +
87 node_page_state(nid, NR_INACTIVE_FILE)),
88 nid, K(node_page_state(nid, NR_ACTIVE_ANON)),
89 nid, K(node_page_state(nid, NR_INACTIVE_ANON)),
90 nid, K(node_page_state(nid, NR_ACTIVE_FILE)),
91 nid, K(node_page_state(nid, NR_INACTIVE_FILE)),
92 nid, K(node_page_state(nid, NR_UNEVICTABLE)),
93 nid, K(node_page_state(nid, NR_MLOCK)));
94
Christoph Lameter182e8e22006-09-25 23:31:10 -070095#ifdef CONFIG_HIGHMEM
KOSAKI Motohiro7ee92252010-08-09 17:19:50 -070096 n += sprintf(buf + n,
Rik van Riel4f98a2f2008-10-18 20:26:32 -070097 "Node %d HighTotal: %8lu kB\n"
98 "Node %d HighFree: %8lu kB\n"
99 "Node %d LowTotal: %8lu kB\n"
KOSAKI Motohiro7ee92252010-08-09 17:19:50 -0700100 "Node %d LowFree: %8lu kB\n",
101 nid, K(i.totalhigh),
102 nid, K(i.freehigh),
103 nid, K(i.totalram - i.totalhigh),
104 nid, K(i.freeram - i.freehigh));
Christoph Lameter182e8e22006-09-25 23:31:10 -0700105#endif
KOSAKI Motohiro7ee92252010-08-09 17:19:50 -0700106 n += sprintf(buf + n,
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700107 "Node %d Dirty: %8lu kB\n"
108 "Node %d Writeback: %8lu kB\n"
109 "Node %d FilePages: %8lu kB\n"
110 "Node %d Mapped: %8lu kB\n"
111 "Node %d AnonPages: %8lu kB\n"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700112 "Node %d Shmem: %8lu kB\n"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -0700113 "Node %d KernelStack: %8lu kB\n"
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700114 "Node %d PageTables: %8lu kB\n"
115 "Node %d NFS_Unstable: %8lu kB\n"
116 "Node %d Bounce: %8lu kB\n"
117 "Node %d WritebackTmp: %8lu kB\n"
118 "Node %d Slab: %8lu kB\n"
119 "Node %d SReclaimable: %8lu kB\n"
120 "Node %d SUnreclaim: %8lu kB\n",
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -0700121 nid, K(node_page_state(nid, NR_FILE_DIRTY)),
Christoph Lameterce866b32006-06-30 01:55:40 -0700122 nid, K(node_page_state(nid, NR_WRITEBACK)),
Christoph Lameter347ce432006-06-30 01:55:35 -0700123 nid, K(node_page_state(nid, NR_FILE_PAGES)),
Christoph Lameter65ba55f2006-06-30 01:55:34 -0700124 nid, K(node_page_state(nid, NR_FILE_MAPPED)),
Christoph Lameterf3dbd342006-06-30 01:55:36 -0700125 nid, K(node_page_state(nid, NR_ANON_PAGES)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700126 nid, K(node_page_state(nid, NR_SHMEM)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -0700127 nid, node_page_state(nid, NR_KERNEL_STACK) *
128 THREAD_SIZE / 1024,
Christoph Lameterdf849a12006-06-30 01:55:38 -0700129 nid, K(node_page_state(nid, NR_PAGETABLE)),
Christoph Lameterfd39fc82006-06-30 01:55:40 -0700130 nid, K(node_page_state(nid, NR_UNSTABLE_NFS)),
Christoph Lameterd2c5e302006-06-30 01:55:41 -0700131 nid, K(node_page_state(nid, NR_BOUNCE)),
Miklos Szeredifc3ba692008-04-30 00:54:38 -0700132 nid, K(node_page_state(nid, NR_WRITEBACK_TEMP)),
Christoph Lameter972d1a72006-09-25 23:31:51 -0700133 nid, K(node_page_state(nid, NR_SLAB_RECLAIMABLE) +
134 node_page_state(nid, NR_SLAB_UNRECLAIMABLE)),
135 nid, K(node_page_state(nid, NR_SLAB_RECLAIMABLE)),
136 nid, K(node_page_state(nid, NR_SLAB_UNRECLAIMABLE)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 n += hugetlb_report_node_meminfo(nid, buf + n);
138 return n;
139}
140
141#undef K
142static SYSDEV_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
143
Andi Kleen4a0b2b42008-07-01 18:48:41 +0200144static ssize_t node_read_numastat(struct sys_device * dev,
145 struct sysdev_attribute *attr, char * buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 return sprintf(buf,
148 "numa_hit %lu\n"
149 "numa_miss %lu\n"
150 "numa_foreign %lu\n"
151 "interleave_hit %lu\n"
152 "local_node %lu\n"
153 "other_node %lu\n",
Christoph Lameterca889e62006-06-30 01:55:44 -0700154 node_page_state(dev->id, NUMA_HIT),
155 node_page_state(dev->id, NUMA_MISS),
156 node_page_state(dev->id, NUMA_FOREIGN),
157 node_page_state(dev->id, NUMA_INTERLEAVE_HIT),
158 node_page_state(dev->id, NUMA_LOCAL),
159 node_page_state(dev->id, NUMA_OTHER));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160}
161static SYSDEV_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
162
Michael Rubin2ac39032010-10-26 14:21:35 -0700163static ssize_t node_read_vmstat(struct sys_device *dev,
164 struct sysdev_attribute *attr, char *buf)
165{
166 int nid = dev->id;
167 return sprintf(buf,
168 "nr_written %lu\n"
169 "nr_dirtied %lu\n",
170 node_page_state(nid, NR_WRITTEN),
171 node_page_state(nid, NR_DIRTIED));
172}
173static SYSDEV_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL);
174
Andi Kleen4a0b2b42008-07-01 18:48:41 +0200175static ssize_t node_read_distance(struct sys_device * dev,
176 struct sysdev_attribute *attr, char * buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177{
178 int nid = dev->id;
179 int len = 0;
180 int i;
181
David Rientjes12ee3c02010-03-10 14:50:21 -0800182 /*
183 * buf is currently PAGE_SIZE in length and each node needs 4 chars
184 * at the most (distance + space or newline).
185 */
186 BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187
188 for_each_online_node(i)
189 len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i));
190
191 len += sprintf(buf + len, "\n");
192 return len;
193}
194static SYSDEV_ATTR(distance, S_IRUGO, node_read_distance, NULL);
195
Lee Schermerhorn9a305232009-12-14 17:58:25 -0800196#ifdef CONFIG_HUGETLBFS
197/*
198 * hugetlbfs per node attributes registration interface:
199 * When/if hugetlb[fs] subsystem initializes [sometime after this module],
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800200 * it will register its per node attributes for all online nodes with
201 * memory. It will also call register_hugetlbfs_with_node(), below, to
Lee Schermerhorn9a305232009-12-14 17:58:25 -0800202 * register its attribute registration functions with this node driver.
203 * Once these hooks have been initialized, the node driver will call into
204 * the hugetlb module to [un]register attributes for hot-plugged nodes.
205 */
206static node_registration_func_t __hugetlb_register_node;
207static node_registration_func_t __hugetlb_unregister_node;
208
Lee Schermerhorn39da08c2009-12-14 17:58:36 -0800209static inline bool hugetlb_register_node(struct node *node)
Lee Schermerhorn9a305232009-12-14 17:58:25 -0800210{
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800211 if (__hugetlb_register_node &&
Lee Schermerhorn39da08c2009-12-14 17:58:36 -0800212 node_state(node->sysdev.id, N_HIGH_MEMORY)) {
Lee Schermerhorn9a305232009-12-14 17:58:25 -0800213 __hugetlb_register_node(node);
Lee Schermerhorn39da08c2009-12-14 17:58:36 -0800214 return true;
215 }
216 return false;
Lee Schermerhorn9a305232009-12-14 17:58:25 -0800217}
218
219static inline void hugetlb_unregister_node(struct node *node)
220{
221 if (__hugetlb_unregister_node)
222 __hugetlb_unregister_node(node);
223}
224
225void register_hugetlbfs_with_node(node_registration_func_t doregister,
226 node_registration_func_t unregister)
227{
228 __hugetlb_register_node = doregister;
229 __hugetlb_unregister_node = unregister;
230}
231#else
232static inline void hugetlb_register_node(struct node *node) {}
233
234static inline void hugetlb_unregister_node(struct node *node) {}
235#endif
236
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237
238/*
Robert P. J. Day405ae7d2007-02-17 19:13:42 +0100239 * register_node - Setup a sysfs device for a node.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 * @num - Node number to use when creating the device.
241 *
242 * Initialize and register the node device.
243 */
Keiichiro Tokunaga4b450992005-05-08 21:28:53 +0900244int register_node(struct node *node, int num, struct node *parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245{
246 int error;
247
248 node->sysdev.id = num;
249 node->sysdev.cls = &node_class;
250 error = sysdev_register(&node->sysdev);
251
252 if (!error){
253 sysdev_create_file(&node->sysdev, &attr_cpumap);
Mike Travis39106dc2008-04-08 11:43:03 -0700254 sysdev_create_file(&node->sysdev, &attr_cpulist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 sysdev_create_file(&node->sysdev, &attr_meminfo);
256 sysdev_create_file(&node->sysdev, &attr_numastat);
257 sysdev_create_file(&node->sysdev, &attr_distance);
Michael Rubin2ac39032010-10-26 14:21:35 -0700258 sysdev_create_file(&node->sysdev, &attr_vmstat);
Lee Schermerhornaf936a12008-10-18 20:26:53 -0700259
260 scan_unevictable_register_node(node);
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800261
Lee Schermerhorn9a305232009-12-14 17:58:25 -0800262 hugetlb_register_node(node);
Mel Gormaned4a6d72010-05-24 14:32:29 -0700263
264 compaction_register_node(node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 }
266 return error;
267}
268
Keiichiro Tokunaga4b450992005-05-08 21:28:53 +0900269/**
270 * unregister_node - unregister a node device
271 * @node: node going away
272 *
273 * Unregisters a node device @node. All the devices on the node must be
274 * unregistered before calling this function.
275 */
276void unregister_node(struct node *node)
277{
278 sysdev_remove_file(&node->sysdev, &attr_cpumap);
Mike Travis39106dc2008-04-08 11:43:03 -0700279 sysdev_remove_file(&node->sysdev, &attr_cpulist);
Keiichiro Tokunaga4b450992005-05-08 21:28:53 +0900280 sysdev_remove_file(&node->sysdev, &attr_meminfo);
281 sysdev_remove_file(&node->sysdev, &attr_numastat);
282 sysdev_remove_file(&node->sysdev, &attr_distance);
Michael Rubin2ac39032010-10-26 14:21:35 -0700283 sysdev_remove_file(&node->sysdev, &attr_vmstat);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284
Lee Schermerhornaf936a12008-10-18 20:26:53 -0700285 scan_unevictable_unregister_node(node);
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800286 hugetlb_unregister_node(node); /* no-op, if memoryless node */
Lee Schermerhornaf936a12008-10-18 20:26:53 -0700287
Keiichiro Tokunaga4b450992005-05-08 21:28:53 +0900288 sysdev_unregister(&node->sysdev);
289}
290
Yasunori Goto0fc44152006-06-27 02:53:38 -0700291struct node node_devices[MAX_NUMNODES];
292
KAMEZAWA Hiroyuki76b67ed92006-06-27 02:53:41 -0700293/*
294 * register cpu under node
295 */
296int register_cpu_under_node(unsigned int cpu, unsigned int nid)
297{
Alex Chiang18307942009-12-14 17:59:08 -0800298 int ret;
Alex Chiangf8246f32009-12-14 17:59:06 -0800299 struct sys_device *obj;
KAMEZAWA Hiroyuki76b67ed92006-06-27 02:53:41 -0700300
Alex Chiangf8246f32009-12-14 17:59:06 -0800301 if (!node_online(nid))
302 return 0;
303
304 obj = get_cpu_sysdev(cpu);
305 if (!obj)
306 return 0;
307
Alex Chiang18307942009-12-14 17:59:08 -0800308 ret = sysfs_create_link(&node_devices[nid].sysdev.kobj,
Alex Chiangf8246f32009-12-14 17:59:06 -0800309 &obj->kobj,
310 kobject_name(&obj->kobj));
Alex Chiang18307942009-12-14 17:59:08 -0800311 if (ret)
312 return ret;
313
314 return sysfs_create_link(&obj->kobj,
315 &node_devices[nid].sysdev.kobj,
316 kobject_name(&node_devices[nid].sysdev.kobj));
KAMEZAWA Hiroyuki76b67ed92006-06-27 02:53:41 -0700317}
318
319int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
320{
Alex Chiangb9d52da2009-12-14 17:59:07 -0800321 struct sys_device *obj;
322
323 if (!node_online(nid))
324 return 0;
325
326 obj = get_cpu_sysdev(cpu);
327 if (!obj)
328 return 0;
329
330 sysfs_remove_link(&node_devices[nid].sysdev.kobj,
331 kobject_name(&obj->kobj));
Alex Chiang18307942009-12-14 17:59:08 -0800332 sysfs_remove_link(&obj->kobj,
333 kobject_name(&node_devices[nid].sysdev.kobj));
Alex Chiangb9d52da2009-12-14 17:59:07 -0800334
KAMEZAWA Hiroyuki76b67ed92006-06-27 02:53:41 -0700335 return 0;
336}
337
Gary Hadec04fc582009-01-06 14:39:14 -0800338#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
339#define page_initialized(page) (page->lru.next)
340
341static int get_nid_for_pfn(unsigned long pfn)
342{
343 struct page *page;
344
345 if (!pfn_valid_within(pfn))
346 return -1;
347 page = pfn_to_page(pfn);
348 if (!page_initialized(page))
349 return -1;
350 return pfn_to_nid(pfn);
351}
352
353/* register memory section under specified node if it spans that node */
354int register_mem_sect_under_node(struct memory_block *mem_blk, int nid)
355{
Alex Chiangdee5d0d2009-12-14 17:59:05 -0800356 int ret;
Gary Hadec04fc582009-01-06 14:39:14 -0800357 unsigned long pfn, sect_start_pfn, sect_end_pfn;
358
359 if (!mem_blk)
360 return -EFAULT;
361 if (!node_online(nid))
362 return 0;
363 sect_start_pfn = section_nr_to_pfn(mem_blk->phys_index);
364 sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
365 for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
366 int page_nid;
367
368 page_nid = get_nid_for_pfn(pfn);
369 if (page_nid < 0)
370 continue;
371 if (page_nid != nid)
372 continue;
Alex Chiangdee5d0d2009-12-14 17:59:05 -0800373 ret = sysfs_create_link_nowarn(&node_devices[nid].sysdev.kobj,
Gary Hadec04fc582009-01-06 14:39:14 -0800374 &mem_blk->sysdev.kobj,
375 kobject_name(&mem_blk->sysdev.kobj));
Alex Chiangdee5d0d2009-12-14 17:59:05 -0800376 if (ret)
377 return ret;
378
379 return sysfs_create_link_nowarn(&mem_blk->sysdev.kobj,
380 &node_devices[nid].sysdev.kobj,
381 kobject_name(&node_devices[nid].sysdev.kobj));
Gary Hadec04fc582009-01-06 14:39:14 -0800382 }
383 /* mem section does not span the specified node */
384 return 0;
385}
386
387/* unregister memory section under all nodes that it spans */
388int unregister_mem_sect_under_nodes(struct memory_block *mem_blk)
389{
David Rientjes9ae49fa2009-12-14 17:59:46 -0800390 NODEMASK_ALLOC(nodemask_t, unlinked_nodes, GFP_KERNEL);
Gary Hadec04fc582009-01-06 14:39:14 -0800391 unsigned long pfn, sect_start_pfn, sect_end_pfn;
392
David Rientjes9ae49fa2009-12-14 17:59:46 -0800393 if (!mem_blk) {
394 NODEMASK_FREE(unlinked_nodes);
Gary Hadec04fc582009-01-06 14:39:14 -0800395 return -EFAULT;
David Rientjes9ae49fa2009-12-14 17:59:46 -0800396 }
397 if (!unlinked_nodes)
398 return -ENOMEM;
399 nodes_clear(*unlinked_nodes);
Gary Hadec04fc582009-01-06 14:39:14 -0800400 sect_start_pfn = section_nr_to_pfn(mem_blk->phys_index);
401 sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
402 for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
Roel Kluin47504982009-03-10 12:55:45 -0700403 int nid;
Gary Hadec04fc582009-01-06 14:39:14 -0800404
405 nid = get_nid_for_pfn(pfn);
406 if (nid < 0)
407 continue;
408 if (!node_online(nid))
409 continue;
David Rientjes9ae49fa2009-12-14 17:59:46 -0800410 if (node_test_and_set(nid, *unlinked_nodes))
Gary Hadec04fc582009-01-06 14:39:14 -0800411 continue;
412 sysfs_remove_link(&node_devices[nid].sysdev.kobj,
413 kobject_name(&mem_blk->sysdev.kobj));
Alex Chiangdee5d0d2009-12-14 17:59:05 -0800414 sysfs_remove_link(&mem_blk->sysdev.kobj,
415 kobject_name(&node_devices[nid].sysdev.kobj));
Gary Hadec04fc582009-01-06 14:39:14 -0800416 }
David Rientjes9ae49fa2009-12-14 17:59:46 -0800417 NODEMASK_FREE(unlinked_nodes);
Gary Hadec04fc582009-01-06 14:39:14 -0800418 return 0;
419}
420
421static int link_mem_sections(int nid)
422{
423 unsigned long start_pfn = NODE_DATA(nid)->node_start_pfn;
424 unsigned long end_pfn = start_pfn + NODE_DATA(nid)->node_spanned_pages;
425 unsigned long pfn;
Robin Holt63d027a2010-09-29 14:00:56 -0500426 struct memory_block *mem_blk = NULL;
Gary Hadec04fc582009-01-06 14:39:14 -0800427 int err = 0;
428
429 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
430 unsigned long section_nr = pfn_to_section_nr(pfn);
431 struct mem_section *mem_sect;
Gary Hadec04fc582009-01-06 14:39:14 -0800432 int ret;
433
434 if (!present_section_nr(section_nr))
435 continue;
436 mem_sect = __nr_to_section(section_nr);
Robin Holt63d027a2010-09-29 14:00:56 -0500437 mem_blk = find_memory_block_hinted(mem_sect, mem_blk);
Gary Hadec04fc582009-01-06 14:39:14 -0800438 ret = register_mem_sect_under_node(mem_blk, nid);
439 if (!err)
440 err = ret;
441
442 /* discard ref obtained in find_memory_block() */
Gary Hadec04fc582009-01-06 14:39:14 -0800443 }
Robin Holt63d027a2010-09-29 14:00:56 -0500444
445 if (mem_blk)
446 kobject_put(&mem_blk->sysdev.kobj);
Gary Hadec04fc582009-01-06 14:39:14 -0800447 return err;
448}
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800449
Lee Schermerhorn39da08c2009-12-14 17:58:36 -0800450#ifdef CONFIG_HUGETLBFS
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800451/*
452 * Handle per node hstate attribute [un]registration on transistions
453 * to/from memoryless state.
454 */
Lee Schermerhorn39da08c2009-12-14 17:58:36 -0800455static void node_hugetlb_work(struct work_struct *work)
456{
457 struct node *node = container_of(work, struct node, node_work);
458
459 /*
460 * We only get here when a node transitions to/from memoryless state.
461 * We can detect which transition occurred by examining whether the
462 * node has memory now. hugetlb_register_node() already check this
463 * so we try to register the attributes. If that fails, then the
464 * node has transitioned to memoryless, try to unregister the
465 * attributes.
466 */
467 if (!hugetlb_register_node(node))
468 hugetlb_unregister_node(node);
469}
470
471static void init_node_hugetlb_work(int nid)
472{
473 INIT_WORK(&node_devices[nid].node_work, node_hugetlb_work);
474}
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800475
476static int node_memory_callback(struct notifier_block *self,
477 unsigned long action, void *arg)
478{
479 struct memory_notify *mnb = arg;
480 int nid = mnb->status_change_nid;
481
482 switch (action) {
Lee Schermerhorn39da08c2009-12-14 17:58:36 -0800483 case MEM_ONLINE:
484 case MEM_OFFLINE:
485 /*
486 * offload per node hstate [un]registration to a work thread
487 * when transitioning to/from memoryless state.
488 */
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800489 if (nid != NUMA_NO_NODE)
Lee Schermerhorn39da08c2009-12-14 17:58:36 -0800490 schedule_work(&node_devices[nid].node_work);
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800491 break;
Lee Schermerhorn39da08c2009-12-14 17:58:36 -0800492
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800493 case MEM_GOING_ONLINE:
494 case MEM_GOING_OFFLINE:
495 case MEM_CANCEL_ONLINE:
496 case MEM_CANCEL_OFFLINE:
497 default:
498 break;
499 }
500
501 return NOTIFY_OK;
502}
Lee Schermerhorn39da08c2009-12-14 17:58:36 -0800503#endif /* CONFIG_HUGETLBFS */
504#else /* !CONFIG_MEMORY_HOTPLUG_SPARSE */
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800505
Lee Schermerhorn39da08c2009-12-14 17:58:36 -0800506static int link_mem_sections(int nid) { return 0; }
507#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
508
509#if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
510 !defined(CONFIG_HUGETLBFS)
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800511static inline int node_memory_callback(struct notifier_block *self,
512 unsigned long action, void *arg)
513{
514 return NOTIFY_OK;
515}
Lee Schermerhorn39da08c2009-12-14 17:58:36 -0800516
517static void init_node_hugetlb_work(int nid) { }
518
519#endif
Gary Hadec04fc582009-01-06 14:39:14 -0800520
Yasunori Goto0fc44152006-06-27 02:53:38 -0700521int register_one_node(int nid)
522{
523 int error = 0;
KAMEZAWA Hiroyuki76b67ed92006-06-27 02:53:41 -0700524 int cpu;
Yasunori Goto0fc44152006-06-27 02:53:38 -0700525
526 if (node_online(nid)) {
527 int p_node = parent_node(nid);
528 struct node *parent = NULL;
529
530 if (p_node != nid)
531 parent = &node_devices[p_node];
532
533 error = register_node(&node_devices[nid], nid, parent);
KAMEZAWA Hiroyuki76b67ed92006-06-27 02:53:41 -0700534
535 /* link cpu under this node */
536 for_each_present_cpu(cpu) {
537 if (cpu_to_node(cpu) == nid)
538 register_cpu_under_node(cpu, nid);
539 }
Gary Hadec04fc582009-01-06 14:39:14 -0800540
541 /* link memory sections under this node */
542 error = link_mem_sections(nid);
Lee Schermerhorn39da08c2009-12-14 17:58:36 -0800543
544 /* initialize work queue for memory hot plug */
545 init_node_hugetlb_work(nid);
Yasunori Goto0fc44152006-06-27 02:53:38 -0700546 }
547
548 return error;
549
550}
551
552void unregister_one_node(int nid)
553{
554 unregister_node(&node_devices[nid]);
555}
556
Lee Schermerhornbde631a2007-10-16 01:26:27 -0700557/*
558 * node states attributes
559 */
560
561static ssize_t print_nodes_state(enum node_states state, char *buf)
562{
563 int n;
564
565 n = nodelist_scnprintf(buf, PAGE_SIZE, node_states[state]);
566 if (n > 0 && PAGE_SIZE > n + 1) {
567 *(buf + n++) = '\n';
568 *(buf + n++) = '\0';
569 }
570 return n;
571}
572
Andi Kleenb15f5622010-01-05 12:47:59 +0100573struct node_attr {
574 struct sysdev_class_attribute attr;
575 enum node_states state;
576};
577
578static ssize_t show_node_state(struct sysdev_class *class,
579 struct sysdev_class_attribute *attr, char *buf)
Lee Schermerhornbde631a2007-10-16 01:26:27 -0700580{
Andi Kleenb15f5622010-01-05 12:47:59 +0100581 struct node_attr *na = container_of(attr, struct node_attr, attr);
582 return print_nodes_state(na->state, buf);
Lee Schermerhornbde631a2007-10-16 01:26:27 -0700583}
584
Andi Kleenb15f5622010-01-05 12:47:59 +0100585#define _NODE_ATTR(name, state) \
586 { _SYSDEV_CLASS_ATTR(name, 0444, show_node_state, NULL), state }
Lee Schermerhornbde631a2007-10-16 01:26:27 -0700587
Andi Kleenb15f5622010-01-05 12:47:59 +0100588static struct node_attr node_state_attr[] = {
589 _NODE_ATTR(possible, N_POSSIBLE),
590 _NODE_ATTR(online, N_ONLINE),
591 _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
592 _NODE_ATTR(has_cpu, N_CPU),
Lee Schermerhornbde631a2007-10-16 01:26:27 -0700593#ifdef CONFIG_HIGHMEM
Andi Kleenb15f5622010-01-05 12:47:59 +0100594 _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
Lee Schermerhornbde631a2007-10-16 01:26:27 -0700595#endif
Lee Schermerhornbde631a2007-10-16 01:26:27 -0700596};
597
Andi Kleen3701cde2010-01-05 12:48:04 +0100598static struct sysdev_class_attribute *node_state_attrs[] = {
599 &node_state_attr[0].attr,
600 &node_state_attr[1].attr,
601 &node_state_attr[2].attr,
602 &node_state_attr[3].attr,
603#ifdef CONFIG_HIGHMEM
604 &node_state_attr[4].attr,
605#endif
606 NULL
607};
Lee Schermerhornbde631a2007-10-16 01:26:27 -0700608
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800609#define NODE_CALLBACK_PRI 2 /* lower than SLAB */
Keiichiro Tokunaga4b450992005-05-08 21:28:53 +0900610static int __init register_node_type(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611{
Lee Schermerhornbde631a2007-10-16 01:26:27 -0700612 int ret;
613
Andi Kleen3701cde2010-01-05 12:48:04 +0100614 BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
615 BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
616
Lee Schermerhornbde631a2007-10-16 01:26:27 -0700617 ret = sysdev_class_register(&node_class);
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800618 if (!ret) {
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800619 hotplug_memory_notifier(node_memory_callback,
620 NODE_CALLBACK_PRI);
621 }
Lee Schermerhornbde631a2007-10-16 01:26:27 -0700622
623 /*
624 * Note: we're not going to unregister the node class if we fail
625 * to register the node state class attribute files.
626 */
627 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628}
629postcore_initcall(register_node_type);