blob: 70122791683d5259ed4afbcc145aaad73aa59cd3 [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>
12#include <linux/cpumask.h>
13#include <linux/topology.h>
14#include <linux/nodemask.h>
KAMEZAWA Hiroyuki76b67ed2006-06-27 02:53:41 -070015#include <linux/cpu.h>
Lee Schermerhornbde631a2007-10-16 01:26:27 -070016#include <linux/device.h>
Lee Schermerhornaf936a12008-10-18 20:26:53 -070017#include <linux/swap.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018
19static struct sysdev_class node_class = {
Kay Sieversaf5ca3f2007-12-20 02:09:39 +010020 .name = "node",
Linus Torvalds1da177e2005-04-16 15:20:36 -070021};
22
23
Mike Travis39106dc2008-04-08 11:43:03 -070024static ssize_t node_read_cpumap(struct sys_device *dev, int type, char *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -070025{
26 struct node *node_dev = to_node(dev);
Rusty Russella70f7302009-03-13 14:49:46 +103027 const struct cpumask *mask = cpumask_of_node(node_dev->sysdev.id);
Linus Torvalds1da177e2005-04-16 15:20:36 -070028 int len;
29
Mike Travis39106dc2008-04-08 11:43:03 -070030 /* 2008/04/07: buf currently PAGE_SIZE, need 9 chars per 32 bits. */
31 BUILD_BUG_ON((NR_CPUS/32 * 9) > (PAGE_SIZE-1));
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
Mike Travis39106dc2008-04-08 11:43:03 -070033 len = type?
Rusty Russell29c01772008-12-13 21:20:25 +103034 cpulist_scnprintf(buf, PAGE_SIZE-2, mask) :
35 cpumask_scnprintf(buf, PAGE_SIZE-2, mask);
Mike Travisc5f59f02008-04-04 18:11:10 -070036 buf[len++] = '\n';
37 buf[len] = '\0';
Linus Torvalds1da177e2005-04-16 15:20:36 -070038 return len;
39}
40
Andi Kleen4a0b2b42008-07-01 18:48:41 +020041static inline ssize_t node_read_cpumask(struct sys_device *dev,
42 struct sysdev_attribute *attr, char *buf)
Mike Travis39106dc2008-04-08 11:43:03 -070043{
44 return node_read_cpumap(dev, 0, buf);
45}
Andi Kleen4a0b2b42008-07-01 18:48:41 +020046static inline ssize_t node_read_cpulist(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, 1, buf);
50}
51
52static SYSDEV_ATTR(cpumap, S_IRUGO, node_read_cpumask, NULL);
53static SYSDEV_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
55#define K(x) ((x) << (PAGE_SHIFT - 10))
Andi Kleen4a0b2b42008-07-01 18:48:41 +020056static ssize_t node_read_meminfo(struct sys_device * dev,
57 struct sysdev_attribute *attr, char * buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -070058{
59 int n;
60 int nid = dev->id;
61 struct sysinfo i;
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
63 si_meminfo_node(&i, nid);
Martin Hicksc07e02d2005-09-03 15:55:11 -070064
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 n = sprintf(buf, "\n"
Rik van Riel4f98a2f2008-10-18 20:26:32 -070066 "Node %d MemTotal: %8lu kB\n"
67 "Node %d MemFree: %8lu kB\n"
68 "Node %d MemUsed: %8lu kB\n"
69 "Node %d Active: %8lu kB\n"
70 "Node %d Inactive: %8lu kB\n"
71 "Node %d Active(anon): %8lu kB\n"
72 "Node %d Inactive(anon): %8lu kB\n"
73 "Node %d Active(file): %8lu kB\n"
74 "Node %d Inactive(file): %8lu kB\n"
Nick Piggin5344b7e2008-10-18 20:26:51 -070075 "Node %d Unevictable: %8lu kB\n"
76 "Node %d Mlocked: %8lu kB\n"
Christoph Lameter182e8e22006-09-25 23:31:10 -070077#ifdef CONFIG_HIGHMEM
Rik van Riel4f98a2f2008-10-18 20:26:32 -070078 "Node %d HighTotal: %8lu kB\n"
79 "Node %d HighFree: %8lu kB\n"
80 "Node %d LowTotal: %8lu kB\n"
81 "Node %d LowFree: %8lu kB\n"
Christoph Lameter182e8e22006-09-25 23:31:10 -070082#endif
Rik van Riel4f98a2f2008-10-18 20:26:32 -070083 "Node %d Dirty: %8lu kB\n"
84 "Node %d Writeback: %8lu kB\n"
85 "Node %d FilePages: %8lu kB\n"
86 "Node %d Mapped: %8lu kB\n"
87 "Node %d AnonPages: %8lu kB\n"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -070088 "Node %d Shmem: %8lu kB\n"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -070089 "Node %d KernelStack: %8lu kB\n"
Rik van Riel4f98a2f2008-10-18 20:26:32 -070090 "Node %d PageTables: %8lu kB\n"
91 "Node %d NFS_Unstable: %8lu kB\n"
92 "Node %d Bounce: %8lu kB\n"
93 "Node %d WritebackTmp: %8lu kB\n"
94 "Node %d Slab: %8lu kB\n"
95 "Node %d SReclaimable: %8lu kB\n"
96 "Node %d SUnreclaim: %8lu kB\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 nid, K(i.totalram),
98 nid, K(i.freeram),
99 nid, K(i.totalram - i.freeram),
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700100 nid, K(node_page_state(nid, NR_ACTIVE_ANON) +
101 node_page_state(nid, NR_ACTIVE_FILE)),
102 nid, K(node_page_state(nid, NR_INACTIVE_ANON) +
103 node_page_state(nid, NR_INACTIVE_FILE)),
104 nid, K(node_page_state(nid, NR_ACTIVE_ANON)),
105 nid, K(node_page_state(nid, NR_INACTIVE_ANON)),
106 nid, K(node_page_state(nid, NR_ACTIVE_FILE)),
107 nid, K(node_page_state(nid, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -0700108 nid, K(node_page_state(nid, NR_UNEVICTABLE)),
Nick Piggin5344b7e2008-10-18 20:26:51 -0700109 nid, K(node_page_state(nid, NR_MLOCK)),
Christoph Lameter182e8e22006-09-25 23:31:10 -0700110#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 nid, K(i.totalhigh),
112 nid, K(i.freehigh),
113 nid, K(i.totalram - i.totalhigh),
Martin Hicksc07e02d2005-09-03 15:55:11 -0700114 nid, K(i.freeram - i.freehigh),
Christoph Lameter182e8e22006-09-25 23:31:10 -0700115#endif
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -0700116 nid, K(node_page_state(nid, NR_FILE_DIRTY)),
Christoph Lameterce866b32006-06-30 01:55:40 -0700117 nid, K(node_page_state(nid, NR_WRITEBACK)),
Christoph Lameter347ce432006-06-30 01:55:35 -0700118 nid, K(node_page_state(nid, NR_FILE_PAGES)),
Christoph Lameter65ba55f2006-06-30 01:55:34 -0700119 nid, K(node_page_state(nid, NR_FILE_MAPPED)),
Christoph Lameterf3dbd342006-06-30 01:55:36 -0700120 nid, K(node_page_state(nid, NR_ANON_PAGES)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700121 nid, K(node_page_state(nid, NR_SHMEM)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -0700122 nid, node_page_state(nid, NR_KERNEL_STACK) *
123 THREAD_SIZE / 1024,
Christoph Lameterdf849a12006-06-30 01:55:38 -0700124 nid, K(node_page_state(nid, NR_PAGETABLE)),
Christoph Lameterfd39fc82006-06-30 01:55:40 -0700125 nid, K(node_page_state(nid, NR_UNSTABLE_NFS)),
Christoph Lameterd2c5e302006-06-30 01:55:41 -0700126 nid, K(node_page_state(nid, NR_BOUNCE)),
Miklos Szeredifc3ba692008-04-30 00:54:38 -0700127 nid, K(node_page_state(nid, NR_WRITEBACK_TEMP)),
Christoph Lameter972d1a72006-09-25 23:31:51 -0700128 nid, K(node_page_state(nid, NR_SLAB_RECLAIMABLE) +
129 node_page_state(nid, NR_SLAB_UNRECLAIMABLE)),
130 nid, K(node_page_state(nid, NR_SLAB_RECLAIMABLE)),
131 nid, K(node_page_state(nid, NR_SLAB_UNRECLAIMABLE)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 n += hugetlb_report_node_meminfo(nid, buf + n);
133 return n;
134}
135
136#undef K
137static SYSDEV_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
138
Andi Kleen4a0b2b42008-07-01 18:48:41 +0200139static ssize_t node_read_numastat(struct sys_device * dev,
140 struct sysdev_attribute *attr, char * buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 return sprintf(buf,
143 "numa_hit %lu\n"
144 "numa_miss %lu\n"
145 "numa_foreign %lu\n"
146 "interleave_hit %lu\n"
147 "local_node %lu\n"
148 "other_node %lu\n",
Christoph Lameterca889e62006-06-30 01:55:44 -0700149 node_page_state(dev->id, NUMA_HIT),
150 node_page_state(dev->id, NUMA_MISS),
151 node_page_state(dev->id, NUMA_FOREIGN),
152 node_page_state(dev->id, NUMA_INTERLEAVE_HIT),
153 node_page_state(dev->id, NUMA_LOCAL),
154 node_page_state(dev->id, NUMA_OTHER));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155}
156static SYSDEV_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
157
Andi Kleen4a0b2b42008-07-01 18:48:41 +0200158static ssize_t node_read_distance(struct sys_device * dev,
159 struct sysdev_attribute *attr, char * buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160{
161 int nid = dev->id;
162 int len = 0;
163 int i;
164
165 /* buf currently PAGE_SIZE, need ~4 chars per node */
166 BUILD_BUG_ON(MAX_NUMNODES*4 > PAGE_SIZE/2);
167
168 for_each_online_node(i)
169 len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i));
170
171 len += sprintf(buf + len, "\n");
172 return len;
173}
174static SYSDEV_ATTR(distance, S_IRUGO, node_read_distance, NULL);
175
Lee Schermerhorn9a305232009-12-14 17:58:25 -0800176#ifdef CONFIG_HUGETLBFS
177/*
178 * hugetlbfs per node attributes registration interface:
179 * When/if hugetlb[fs] subsystem initializes [sometime after this module],
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800180 * it will register its per node attributes for all online nodes with
181 * memory. It will also call register_hugetlbfs_with_node(), below, to
Lee Schermerhorn9a305232009-12-14 17:58:25 -0800182 * register its attribute registration functions with this node driver.
183 * Once these hooks have been initialized, the node driver will call into
184 * the hugetlb module to [un]register attributes for hot-plugged nodes.
185 */
186static node_registration_func_t __hugetlb_register_node;
187static node_registration_func_t __hugetlb_unregister_node;
188
Lee Schermerhorn39da08c2009-12-14 17:58:36 -0800189static inline bool hugetlb_register_node(struct node *node)
Lee Schermerhorn9a305232009-12-14 17:58:25 -0800190{
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800191 if (__hugetlb_register_node &&
Lee Schermerhorn39da08c2009-12-14 17:58:36 -0800192 node_state(node->sysdev.id, N_HIGH_MEMORY)) {
Lee Schermerhorn9a305232009-12-14 17:58:25 -0800193 __hugetlb_register_node(node);
Lee Schermerhorn39da08c2009-12-14 17:58:36 -0800194 return true;
195 }
196 return false;
Lee Schermerhorn9a305232009-12-14 17:58:25 -0800197}
198
199static inline void hugetlb_unregister_node(struct node *node)
200{
201 if (__hugetlb_unregister_node)
202 __hugetlb_unregister_node(node);
203}
204
205void register_hugetlbfs_with_node(node_registration_func_t doregister,
206 node_registration_func_t unregister)
207{
208 __hugetlb_register_node = doregister;
209 __hugetlb_unregister_node = unregister;
210}
211#else
212static inline void hugetlb_register_node(struct node *node) {}
213
214static inline void hugetlb_unregister_node(struct node *node) {}
215#endif
216
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217
218/*
Robert P. J. Day405ae7d2007-02-17 19:13:42 +0100219 * register_node - Setup a sysfs device for a node.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 * @num - Node number to use when creating the device.
221 *
222 * Initialize and register the node device.
223 */
Keiichiro Tokunaga4b450992005-05-08 21:28:53 +0900224int register_node(struct node *node, int num, struct node *parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225{
226 int error;
227
228 node->sysdev.id = num;
229 node->sysdev.cls = &node_class;
230 error = sysdev_register(&node->sysdev);
231
232 if (!error){
233 sysdev_create_file(&node->sysdev, &attr_cpumap);
Mike Travis39106dc2008-04-08 11:43:03 -0700234 sysdev_create_file(&node->sysdev, &attr_cpulist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 sysdev_create_file(&node->sysdev, &attr_meminfo);
236 sysdev_create_file(&node->sysdev, &attr_numastat);
237 sysdev_create_file(&node->sysdev, &attr_distance);
Lee Schermerhornaf936a12008-10-18 20:26:53 -0700238
239 scan_unevictable_register_node(node);
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800240
Lee Schermerhorn9a305232009-12-14 17:58:25 -0800241 hugetlb_register_node(node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 }
243 return error;
244}
245
Keiichiro Tokunaga4b450992005-05-08 21:28:53 +0900246/**
247 * unregister_node - unregister a node device
248 * @node: node going away
249 *
250 * Unregisters a node device @node. All the devices on the node must be
251 * unregistered before calling this function.
252 */
253void unregister_node(struct node *node)
254{
255 sysdev_remove_file(&node->sysdev, &attr_cpumap);
Mike Travis39106dc2008-04-08 11:43:03 -0700256 sysdev_remove_file(&node->sysdev, &attr_cpulist);
Keiichiro Tokunaga4b450992005-05-08 21:28:53 +0900257 sysdev_remove_file(&node->sysdev, &attr_meminfo);
258 sysdev_remove_file(&node->sysdev, &attr_numastat);
259 sysdev_remove_file(&node->sysdev, &attr_distance);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
Lee Schermerhornaf936a12008-10-18 20:26:53 -0700261 scan_unevictable_unregister_node(node);
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800262 hugetlb_unregister_node(node); /* no-op, if memoryless node */
Lee Schermerhornaf936a12008-10-18 20:26:53 -0700263
Keiichiro Tokunaga4b450992005-05-08 21:28:53 +0900264 sysdev_unregister(&node->sysdev);
265}
266
Yasunori Goto0fc44152006-06-27 02:53:38 -0700267struct node node_devices[MAX_NUMNODES];
268
KAMEZAWA Hiroyuki76b67ed2006-06-27 02:53:41 -0700269/*
270 * register cpu under node
271 */
272int register_cpu_under_node(unsigned int cpu, unsigned int nid)
273{
Alex Chiang18307942009-12-14 17:59:08 -0800274 int ret;
Alex Chiangf8246f32009-12-14 17:59:06 -0800275 struct sys_device *obj;
KAMEZAWA Hiroyuki76b67ed2006-06-27 02:53:41 -0700276
Alex Chiangf8246f32009-12-14 17:59:06 -0800277 if (!node_online(nid))
278 return 0;
279
280 obj = get_cpu_sysdev(cpu);
281 if (!obj)
282 return 0;
283
Alex Chiang18307942009-12-14 17:59:08 -0800284 ret = sysfs_create_link(&node_devices[nid].sysdev.kobj,
Alex Chiangf8246f32009-12-14 17:59:06 -0800285 &obj->kobj,
286 kobject_name(&obj->kobj));
Alex Chiang18307942009-12-14 17:59:08 -0800287 if (ret)
288 return ret;
289
290 return sysfs_create_link(&obj->kobj,
291 &node_devices[nid].sysdev.kobj,
292 kobject_name(&node_devices[nid].sysdev.kobj));
KAMEZAWA Hiroyuki76b67ed2006-06-27 02:53:41 -0700293}
294
295int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
296{
Alex Chiangb9d52da2009-12-14 17:59:07 -0800297 struct sys_device *obj;
298
299 if (!node_online(nid))
300 return 0;
301
302 obj = get_cpu_sysdev(cpu);
303 if (!obj)
304 return 0;
305
306 sysfs_remove_link(&node_devices[nid].sysdev.kobj,
307 kobject_name(&obj->kobj));
Alex Chiang18307942009-12-14 17:59:08 -0800308 sysfs_remove_link(&obj->kobj,
309 kobject_name(&node_devices[nid].sysdev.kobj));
Alex Chiangb9d52da2009-12-14 17:59:07 -0800310
KAMEZAWA Hiroyuki76b67ed2006-06-27 02:53:41 -0700311 return 0;
312}
313
Gary Hadec04fc582009-01-06 14:39:14 -0800314#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
315#define page_initialized(page) (page->lru.next)
316
317static int get_nid_for_pfn(unsigned long pfn)
318{
319 struct page *page;
320
321 if (!pfn_valid_within(pfn))
322 return -1;
323 page = pfn_to_page(pfn);
324 if (!page_initialized(page))
325 return -1;
326 return pfn_to_nid(pfn);
327}
328
329/* register memory section under specified node if it spans that node */
330int register_mem_sect_under_node(struct memory_block *mem_blk, int nid)
331{
Alex Chiangdee5d0d2009-12-14 17:59:05 -0800332 int ret;
Gary Hadec04fc582009-01-06 14:39:14 -0800333 unsigned long pfn, sect_start_pfn, sect_end_pfn;
334
335 if (!mem_blk)
336 return -EFAULT;
337 if (!node_online(nid))
338 return 0;
339 sect_start_pfn = section_nr_to_pfn(mem_blk->phys_index);
340 sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
341 for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
342 int page_nid;
343
344 page_nid = get_nid_for_pfn(pfn);
345 if (page_nid < 0)
346 continue;
347 if (page_nid != nid)
348 continue;
Alex Chiangdee5d0d2009-12-14 17:59:05 -0800349 ret = sysfs_create_link_nowarn(&node_devices[nid].sysdev.kobj,
Gary Hadec04fc582009-01-06 14:39:14 -0800350 &mem_blk->sysdev.kobj,
351 kobject_name(&mem_blk->sysdev.kobj));
Alex Chiangdee5d0d2009-12-14 17:59:05 -0800352 if (ret)
353 return ret;
354
355 return sysfs_create_link_nowarn(&mem_blk->sysdev.kobj,
356 &node_devices[nid].sysdev.kobj,
357 kobject_name(&node_devices[nid].sysdev.kobj));
Gary Hadec04fc582009-01-06 14:39:14 -0800358 }
359 /* mem section does not span the specified node */
360 return 0;
361}
362
363/* unregister memory section under all nodes that it spans */
364int unregister_mem_sect_under_nodes(struct memory_block *mem_blk)
365{
David Rientjes9ae49fa2009-12-14 17:59:46 -0800366 NODEMASK_ALLOC(nodemask_t, unlinked_nodes, GFP_KERNEL);
Gary Hadec04fc582009-01-06 14:39:14 -0800367 unsigned long pfn, sect_start_pfn, sect_end_pfn;
368
David Rientjes9ae49fa2009-12-14 17:59:46 -0800369 if (!mem_blk) {
370 NODEMASK_FREE(unlinked_nodes);
Gary Hadec04fc582009-01-06 14:39:14 -0800371 return -EFAULT;
David Rientjes9ae49fa2009-12-14 17:59:46 -0800372 }
373 if (!unlinked_nodes)
374 return -ENOMEM;
375 nodes_clear(*unlinked_nodes);
Gary Hadec04fc582009-01-06 14:39:14 -0800376 sect_start_pfn = section_nr_to_pfn(mem_blk->phys_index);
377 sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
378 for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
Roel Kluin47504982009-03-10 12:55:45 -0700379 int nid;
Gary Hadec04fc582009-01-06 14:39:14 -0800380
381 nid = get_nid_for_pfn(pfn);
382 if (nid < 0)
383 continue;
384 if (!node_online(nid))
385 continue;
David Rientjes9ae49fa2009-12-14 17:59:46 -0800386 if (node_test_and_set(nid, *unlinked_nodes))
Gary Hadec04fc582009-01-06 14:39:14 -0800387 continue;
388 sysfs_remove_link(&node_devices[nid].sysdev.kobj,
389 kobject_name(&mem_blk->sysdev.kobj));
Alex Chiangdee5d0d2009-12-14 17:59:05 -0800390 sysfs_remove_link(&mem_blk->sysdev.kobj,
391 kobject_name(&node_devices[nid].sysdev.kobj));
Gary Hadec04fc582009-01-06 14:39:14 -0800392 }
David Rientjes9ae49fa2009-12-14 17:59:46 -0800393 NODEMASK_FREE(unlinked_nodes);
Gary Hadec04fc582009-01-06 14:39:14 -0800394 return 0;
395}
396
397static int link_mem_sections(int nid)
398{
399 unsigned long start_pfn = NODE_DATA(nid)->node_start_pfn;
400 unsigned long end_pfn = start_pfn + NODE_DATA(nid)->node_spanned_pages;
401 unsigned long pfn;
402 int err = 0;
403
404 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
405 unsigned long section_nr = pfn_to_section_nr(pfn);
406 struct mem_section *mem_sect;
407 struct memory_block *mem_blk;
408 int ret;
409
410 if (!present_section_nr(section_nr))
411 continue;
412 mem_sect = __nr_to_section(section_nr);
413 mem_blk = find_memory_block(mem_sect);
414 ret = register_mem_sect_under_node(mem_blk, nid);
415 if (!err)
416 err = ret;
417
418 /* discard ref obtained in find_memory_block() */
419 kobject_put(&mem_blk->sysdev.kobj);
420 }
421 return err;
422}
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800423
Lee Schermerhorn39da08c2009-12-14 17:58:36 -0800424#ifdef CONFIG_HUGETLBFS
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800425/*
426 * Handle per node hstate attribute [un]registration on transistions
427 * to/from memoryless state.
428 */
Lee Schermerhorn39da08c2009-12-14 17:58:36 -0800429static void node_hugetlb_work(struct work_struct *work)
430{
431 struct node *node = container_of(work, struct node, node_work);
432
433 /*
434 * We only get here when a node transitions to/from memoryless state.
435 * We can detect which transition occurred by examining whether the
436 * node has memory now. hugetlb_register_node() already check this
437 * so we try to register the attributes. If that fails, then the
438 * node has transitioned to memoryless, try to unregister the
439 * attributes.
440 */
441 if (!hugetlb_register_node(node))
442 hugetlb_unregister_node(node);
443}
444
445static void init_node_hugetlb_work(int nid)
446{
447 INIT_WORK(&node_devices[nid].node_work, node_hugetlb_work);
448}
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800449
450static int node_memory_callback(struct notifier_block *self,
451 unsigned long action, void *arg)
452{
453 struct memory_notify *mnb = arg;
454 int nid = mnb->status_change_nid;
455
456 switch (action) {
Lee Schermerhorn39da08c2009-12-14 17:58:36 -0800457 case MEM_ONLINE:
458 case MEM_OFFLINE:
459 /*
460 * offload per node hstate [un]registration to a work thread
461 * when transitioning to/from memoryless state.
462 */
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800463 if (nid != NUMA_NO_NODE)
Lee Schermerhorn39da08c2009-12-14 17:58:36 -0800464 schedule_work(&node_devices[nid].node_work);
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800465 break;
Lee Schermerhorn39da08c2009-12-14 17:58:36 -0800466
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800467 case MEM_GOING_ONLINE:
468 case MEM_GOING_OFFLINE:
469 case MEM_CANCEL_ONLINE:
470 case MEM_CANCEL_OFFLINE:
471 default:
472 break;
473 }
474
475 return NOTIFY_OK;
476}
Lee Schermerhorn39da08c2009-12-14 17:58:36 -0800477#endif /* CONFIG_HUGETLBFS */
478#else /* !CONFIG_MEMORY_HOTPLUG_SPARSE */
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800479
Lee Schermerhorn39da08c2009-12-14 17:58:36 -0800480static int link_mem_sections(int nid) { return 0; }
481#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
482
483#if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
484 !defined(CONFIG_HUGETLBFS)
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800485static inline int node_memory_callback(struct notifier_block *self,
486 unsigned long action, void *arg)
487{
488 return NOTIFY_OK;
489}
Lee Schermerhorn39da08c2009-12-14 17:58:36 -0800490
491static void init_node_hugetlb_work(int nid) { }
492
493#endif
Gary Hadec04fc582009-01-06 14:39:14 -0800494
Yasunori Goto0fc44152006-06-27 02:53:38 -0700495int register_one_node(int nid)
496{
497 int error = 0;
KAMEZAWA Hiroyuki76b67ed2006-06-27 02:53:41 -0700498 int cpu;
Yasunori Goto0fc44152006-06-27 02:53:38 -0700499
500 if (node_online(nid)) {
501 int p_node = parent_node(nid);
502 struct node *parent = NULL;
503
504 if (p_node != nid)
505 parent = &node_devices[p_node];
506
507 error = register_node(&node_devices[nid], nid, parent);
KAMEZAWA Hiroyuki76b67ed2006-06-27 02:53:41 -0700508
509 /* link cpu under this node */
510 for_each_present_cpu(cpu) {
511 if (cpu_to_node(cpu) == nid)
512 register_cpu_under_node(cpu, nid);
513 }
Gary Hadec04fc582009-01-06 14:39:14 -0800514
515 /* link memory sections under this node */
516 error = link_mem_sections(nid);
Lee Schermerhorn39da08c2009-12-14 17:58:36 -0800517
518 /* initialize work queue for memory hot plug */
519 init_node_hugetlb_work(nid);
Yasunori Goto0fc44152006-06-27 02:53:38 -0700520 }
521
522 return error;
523
524}
525
526void unregister_one_node(int nid)
527{
528 unregister_node(&node_devices[nid]);
529}
530
Lee Schermerhornbde631a2007-10-16 01:26:27 -0700531/*
532 * node states attributes
533 */
534
535static ssize_t print_nodes_state(enum node_states state, char *buf)
536{
537 int n;
538
539 n = nodelist_scnprintf(buf, PAGE_SIZE, node_states[state]);
540 if (n > 0 && PAGE_SIZE > n + 1) {
541 *(buf + n++) = '\n';
542 *(buf + n++) = '\0';
543 }
544 return n;
545}
546
547static ssize_t print_nodes_possible(struct sysdev_class *class, char *buf)
548{
549 return print_nodes_state(N_POSSIBLE, buf);
550}
551
552static ssize_t print_nodes_online(struct sysdev_class *class, char *buf)
553{
554 return print_nodes_state(N_ONLINE, buf);
555}
556
557static ssize_t print_nodes_has_normal_memory(struct sysdev_class *class,
558 char *buf)
559{
560 return print_nodes_state(N_NORMAL_MEMORY, buf);
561}
562
563static ssize_t print_nodes_has_cpu(struct sysdev_class *class, char *buf)
564{
565 return print_nodes_state(N_CPU, buf);
566}
567
568static SYSDEV_CLASS_ATTR(possible, 0444, print_nodes_possible, NULL);
569static SYSDEV_CLASS_ATTR(online, 0444, print_nodes_online, NULL);
570static SYSDEV_CLASS_ATTR(has_normal_memory, 0444, print_nodes_has_normal_memory,
571 NULL);
572static SYSDEV_CLASS_ATTR(has_cpu, 0444, print_nodes_has_cpu, NULL);
573
574#ifdef CONFIG_HIGHMEM
575static ssize_t print_nodes_has_high_memory(struct sysdev_class *class,
576 char *buf)
577{
578 return print_nodes_state(N_HIGH_MEMORY, buf);
579}
580
581static SYSDEV_CLASS_ATTR(has_high_memory, 0444, print_nodes_has_high_memory,
582 NULL);
583#endif
584
585struct sysdev_class_attribute *node_state_attr[] = {
586 &attr_possible,
587 &attr_online,
588 &attr_has_normal_memory,
589#ifdef CONFIG_HIGHMEM
590 &attr_has_high_memory,
591#endif
592 &attr_has_cpu,
593};
594
595static int node_states_init(void)
596{
597 int i;
598 int err = 0;
599
600 for (i = 0; i < NR_NODE_STATES; i++) {
601 int ret;
602 ret = sysdev_class_create_file(&node_class, node_state_attr[i]);
603 if (!err)
604 err = ret;
605 }
606 return err;
607}
608
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
614 ret = sysdev_class_register(&node_class);
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800615 if (!ret) {
Lee Schermerhornbde631a2007-10-16 01:26:27 -0700616 ret = node_states_init();
Lee Schermerhorn4faf8d92009-12-14 17:58:35 -0800617 hotplug_memory_notifier(node_memory_callback,
618 NODE_CALLBACK_PRI);
619 }
Lee Schermerhornbde631a2007-10-16 01:26:27 -0700620
621 /*
622 * Note: we're not going to unregister the node class if we fail
623 * to register the node state class attribute files.
624 */
625 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626}
627postcore_initcall(register_node_type);