91d4087b4039490a844d23e909ffe34eaa0ff0f9
[linux-2.6.git] / drivers / base / node.c
1 /*
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>
9 #include <linux/memory.h>
10 #include <linux/node.h>
11 #include <linux/hugetlb.h>
12 #include <linux/cpumask.h>
13 #include <linux/topology.h>
14 #include <linux/nodemask.h>
15 #include <linux/cpu.h>
16 #include <linux/device.h>
17 #include <linux/swap.h>
18
19 static struct sysdev_class node_class = {
20         .name = "node",
21 };
22
23
24 static ssize_t node_read_cpumap(struct sys_device *dev, int type, char *buf)
25 {
26         struct node *node_dev = to_node(dev);
27         const struct cpumask *mask = cpumask_of_node(node_dev->sysdev.id);
28         int len;
29
30         /* 2008/04/07: buf currently PAGE_SIZE, need 9 chars per 32 bits. */
31         BUILD_BUG_ON((NR_CPUS/32 * 9) > (PAGE_SIZE-1));
32
33         len = type?
34                 cpulist_scnprintf(buf, PAGE_SIZE-2, mask) :
35                 cpumask_scnprintf(buf, PAGE_SIZE-2, mask);
36         buf[len++] = '\n';
37         buf[len] = '\0';
38         return len;
39 }
40
41 static inline ssize_t node_read_cpumask(struct sys_device *dev,
42                                 struct sysdev_attribute *attr, char *buf)
43 {
44         return node_read_cpumap(dev, 0, buf);
45 }
46 static inline ssize_t node_read_cpulist(struct sys_device *dev,
47                                 struct sysdev_attribute *attr, char *buf)
48 {
49         return node_read_cpumap(dev, 1, buf);
50 }
51
52 static SYSDEV_ATTR(cpumap,  S_IRUGO, node_read_cpumask, NULL);
53 static SYSDEV_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL);
54
55 #define K(x) ((x) << (PAGE_SHIFT - 10))
56 static ssize_t node_read_meminfo(struct sys_device * dev,
57                         struct sysdev_attribute *attr, char * buf)
58 {
59         int n;
60         int nid = dev->id;
61         struct sysinfo i;
62
63         si_meminfo_node(&i, nid);
64
65         n = sprintf(buf, "\n"
66                        "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"
75                        "Node %d Unevictable:    %8lu kB\n"
76                        "Node %d Mlocked:        %8lu kB\n"
77 #ifdef CONFIG_HIGHMEM
78                        "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"
82 #endif
83                        "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"
88                        "Node %d PageTables:     %8lu kB\n"
89                        "Node %d NFS_Unstable:   %8lu kB\n"
90                        "Node %d Bounce:         %8lu kB\n"
91                        "Node %d WritebackTmp:   %8lu kB\n"
92                        "Node %d Slab:           %8lu kB\n"
93                        "Node %d SReclaimable:   %8lu kB\n"
94                        "Node %d SUnreclaim:     %8lu kB\n",
95                        nid, K(i.totalram),
96                        nid, K(i.freeram),
97                        nid, K(i.totalram - i.freeram),
98                        nid, K(node_page_state(nid, NR_ACTIVE_ANON) +
99                                 node_page_state(nid, NR_ACTIVE_FILE)),
100                        nid, K(node_page_state(nid, NR_INACTIVE_ANON) +
101                                 node_page_state(nid, NR_INACTIVE_FILE)),
102                        nid, K(node_page_state(nid, NR_ACTIVE_ANON)),
103                        nid, K(node_page_state(nid, NR_INACTIVE_ANON)),
104                        nid, K(node_page_state(nid, NR_ACTIVE_FILE)),
105                        nid, K(node_page_state(nid, NR_INACTIVE_FILE)),
106                        nid, K(node_page_state(nid, NR_UNEVICTABLE)),
107                        nid, K(node_page_state(nid, NR_MLOCK)),
108 #ifdef CONFIG_HIGHMEM
109                        nid, K(i.totalhigh),
110                        nid, K(i.freehigh),
111                        nid, K(i.totalram - i.totalhigh),
112                        nid, K(i.freeram - i.freehigh),
113 #endif
114                        nid, K(node_page_state(nid, NR_FILE_DIRTY)),
115                        nid, K(node_page_state(nid, NR_WRITEBACK)),
116                        nid, K(node_page_state(nid, NR_FILE_PAGES)),
117                        nid, K(node_page_state(nid, NR_FILE_MAPPED)),
118                        nid, K(node_page_state(nid, NR_ANON_PAGES)),
119                        nid, K(node_page_state(nid, NR_PAGETABLE)),
120                        nid, K(node_page_state(nid, NR_UNSTABLE_NFS)),
121                        nid, K(node_page_state(nid, NR_BOUNCE)),
122                        nid, K(node_page_state(nid, NR_WRITEBACK_TEMP)),
123                        nid, K(node_page_state(nid, NR_SLAB_RECLAIMABLE) +
124                                 node_page_state(nid, NR_SLAB_UNRECLAIMABLE)),
125                        nid, K(node_page_state(nid, NR_SLAB_RECLAIMABLE)),
126                        nid, K(node_page_state(nid, NR_SLAB_UNRECLAIMABLE)));
127         n += hugetlb_report_node_meminfo(nid, buf + n);
128         return n;
129 }
130
131 #undef K
132 static SYSDEV_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
133
134 static ssize_t node_read_numastat(struct sys_device * dev,
135                                 struct sysdev_attribute *attr, char * buf)
136 {
137         return sprintf(buf,
138                        "numa_hit %lu\n"
139                        "numa_miss %lu\n"
140                        "numa_foreign %lu\n"
141                        "interleave_hit %lu\n"
142                        "local_node %lu\n"
143                        "other_node %lu\n",
144                        node_page_state(dev->id, NUMA_HIT),
145                        node_page_state(dev->id, NUMA_MISS),
146                        node_page_state(dev->id, NUMA_FOREIGN),
147                        node_page_state(dev->id, NUMA_INTERLEAVE_HIT),
148                        node_page_state(dev->id, NUMA_LOCAL),
149                        node_page_state(dev->id, NUMA_OTHER));
150 }
151 static SYSDEV_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
152
153 static ssize_t node_read_distance(struct sys_device * dev,
154                         struct sysdev_attribute *attr, char * buf)
155 {
156         int nid = dev->id;
157         int len = 0;
158         int i;
159
160         /* buf currently PAGE_SIZE, need ~4 chars per node */
161         BUILD_BUG_ON(MAX_NUMNODES*4 > PAGE_SIZE/2);
162
163         for_each_online_node(i)
164                 len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i));
165
166         len += sprintf(buf + len, "\n");
167         return len;
168 }
169 static SYSDEV_ATTR(distance, S_IRUGO, node_read_distance, NULL);
170
171
172 /*
173  * register_node - Setup a sysfs device for a node.
174  * @num - Node number to use when creating the device.
175  *
176  * Initialize and register the node device.
177  */
178 int register_node(struct node *node, int num, struct node *parent)
179 {
180         int error;
181
182         node->sysdev.id = num;
183         node->sysdev.cls = &node_class;
184         error = sysdev_register(&node->sysdev);
185
186         if (!error){
187                 sysdev_create_file(&node->sysdev, &attr_cpumap);
188                 sysdev_create_file(&node->sysdev, &attr_cpulist);
189                 sysdev_create_file(&node->sysdev, &attr_meminfo);
190                 sysdev_create_file(&node->sysdev, &attr_numastat);
191                 sysdev_create_file(&node->sysdev, &attr_distance);
192
193                 scan_unevictable_register_node(node);
194         }
195         return error;
196 }
197
198 /**
199  * unregister_node - unregister a node device
200  * @node: node going away
201  *
202  * Unregisters a node device @node.  All the devices on the node must be
203  * unregistered before calling this function.
204  */
205 void unregister_node(struct node *node)
206 {
207         sysdev_remove_file(&node->sysdev, &attr_cpumap);
208         sysdev_remove_file(&node->sysdev, &attr_cpulist);
209         sysdev_remove_file(&node->sysdev, &attr_meminfo);
210         sysdev_remove_file(&node->sysdev, &attr_numastat);
211         sysdev_remove_file(&node->sysdev, &attr_distance);
212
213         scan_unevictable_unregister_node(node);
214
215         sysdev_unregister(&node->sysdev);
216 }
217
218 struct node node_devices[MAX_NUMNODES];
219
220 /*
221  * register cpu under node
222  */
223 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
224 {
225         if (node_online(nid)) {
226                 struct sys_device *obj = get_cpu_sysdev(cpu);
227                 if (!obj)
228                         return 0;
229                 return sysfs_create_link(&node_devices[nid].sysdev.kobj,
230                                          &obj->kobj,
231                                          kobject_name(&obj->kobj));
232          }
233
234         return 0;
235 }
236
237 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
238 {
239         if (node_online(nid)) {
240                 struct sys_device *obj = get_cpu_sysdev(cpu);
241                 if (obj)
242                         sysfs_remove_link(&node_devices[nid].sysdev.kobj,
243                                          kobject_name(&obj->kobj));
244         }
245         return 0;
246 }
247
248 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
249 #define page_initialized(page)  (page->lru.next)
250
251 static int get_nid_for_pfn(unsigned long pfn)
252 {
253         struct page *page;
254
255         if (!pfn_valid_within(pfn))
256                 return -1;
257         page = pfn_to_page(pfn);
258         if (!page_initialized(page))
259                 return -1;
260         return pfn_to_nid(pfn);
261 }
262
263 /* register memory section under specified node if it spans that node */
264 int register_mem_sect_under_node(struct memory_block *mem_blk, int nid)
265 {
266         unsigned long pfn, sect_start_pfn, sect_end_pfn;
267
268         if (!mem_blk)
269                 return -EFAULT;
270         if (!node_online(nid))
271                 return 0;
272         sect_start_pfn = section_nr_to_pfn(mem_blk->phys_index);
273         sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
274         for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
275                 int page_nid;
276
277                 page_nid = get_nid_for_pfn(pfn);
278                 if (page_nid < 0)
279                         continue;
280                 if (page_nid != nid)
281                         continue;
282                 return sysfs_create_link_nowarn(&node_devices[nid].sysdev.kobj,
283                                         &mem_blk->sysdev.kobj,
284                                         kobject_name(&mem_blk->sysdev.kobj));
285         }
286         /* mem section does not span the specified node */
287         return 0;
288 }
289
290 /* unregister memory section under all nodes that it spans */
291 int unregister_mem_sect_under_nodes(struct memory_block *mem_blk)
292 {
293         nodemask_t unlinked_nodes;
294         unsigned long pfn, sect_start_pfn, sect_end_pfn;
295
296         if (!mem_blk)
297                 return -EFAULT;
298         nodes_clear(unlinked_nodes);
299         sect_start_pfn = section_nr_to_pfn(mem_blk->phys_index);
300         sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
301         for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
302                 int nid;
303
304                 nid = get_nid_for_pfn(pfn);
305                 if (nid < 0)
306                         continue;
307                 if (!node_online(nid))
308                         continue;
309                 if (node_test_and_set(nid, unlinked_nodes))
310                         continue;
311                 sysfs_remove_link(&node_devices[nid].sysdev.kobj,
312                          kobject_name(&mem_blk->sysdev.kobj));
313         }
314         return 0;
315 }
316
317 static int link_mem_sections(int nid)
318 {
319         unsigned long start_pfn = NODE_DATA(nid)->node_start_pfn;
320         unsigned long end_pfn = start_pfn + NODE_DATA(nid)->node_spanned_pages;
321         unsigned long pfn;
322         int err = 0;
323
324         for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
325                 unsigned long section_nr = pfn_to_section_nr(pfn);
326                 struct mem_section *mem_sect;
327                 struct memory_block *mem_blk;
328                 int ret;
329
330                 if (!present_section_nr(section_nr))
331                         continue;
332                 mem_sect = __nr_to_section(section_nr);
333                 mem_blk = find_memory_block(mem_sect);
334                 ret = register_mem_sect_under_node(mem_blk, nid);
335                 if (!err)
336                         err = ret;
337
338                 /* discard ref obtained in find_memory_block() */
339                 kobject_put(&mem_blk->sysdev.kobj);
340         }
341         return err;
342 }
343 #else
344 static int link_mem_sections(int nid) { return 0; }
345 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
346
347 int register_one_node(int nid)
348 {
349         int error = 0;
350         int cpu;
351
352         if (node_online(nid)) {
353                 int p_node = parent_node(nid);
354                 struct node *parent = NULL;
355
356                 if (p_node != nid)
357                         parent = &node_devices[p_node];
358
359                 error = register_node(&node_devices[nid], nid, parent);
360
361                 /* link cpu under this node */
362                 for_each_present_cpu(cpu) {
363                         if (cpu_to_node(cpu) == nid)
364                                 register_cpu_under_node(cpu, nid);
365                 }
366
367                 /* link memory sections under this node */
368                 error = link_mem_sections(nid);
369         }
370
371         return error;
372
373 }
374
375 void unregister_one_node(int nid)
376 {
377         unregister_node(&node_devices[nid]);
378 }
379
380 /*
381  * node states attributes
382  */
383
384 static ssize_t print_nodes_state(enum node_states state, char *buf)
385 {
386         int n;
387
388         n = nodelist_scnprintf(buf, PAGE_SIZE, node_states[state]);
389         if (n > 0 && PAGE_SIZE > n + 1) {
390                 *(buf + n++) = '\n';
391                 *(buf + n++) = '\0';
392         }
393         return n;
394 }
395
396 static ssize_t print_nodes_possible(struct sysdev_class *class, char *buf)
397 {
398         return print_nodes_state(N_POSSIBLE, buf);
399 }
400
401 static ssize_t print_nodes_online(struct sysdev_class *class, char *buf)
402 {
403         return print_nodes_state(N_ONLINE, buf);
404 }
405
406 static ssize_t print_nodes_has_normal_memory(struct sysdev_class *class,
407                                                 char *buf)
408 {
409         return print_nodes_state(N_NORMAL_MEMORY, buf);
410 }
411
412 static ssize_t print_nodes_has_cpu(struct sysdev_class *class, char *buf)
413 {
414         return print_nodes_state(N_CPU, buf);
415 }
416
417 static SYSDEV_CLASS_ATTR(possible, 0444, print_nodes_possible, NULL);
418 static SYSDEV_CLASS_ATTR(online, 0444, print_nodes_online, NULL);
419 static SYSDEV_CLASS_ATTR(has_normal_memory, 0444, print_nodes_has_normal_memory,
420                                                                         NULL);
421 static SYSDEV_CLASS_ATTR(has_cpu, 0444, print_nodes_has_cpu, NULL);
422
423 #ifdef CONFIG_HIGHMEM
424 static ssize_t print_nodes_has_high_memory(struct sysdev_class *class,
425                                                  char *buf)
426 {
427         return print_nodes_state(N_HIGH_MEMORY, buf);
428 }
429
430 static SYSDEV_CLASS_ATTR(has_high_memory, 0444, print_nodes_has_high_memory,
431                                                                          NULL);
432 #endif
433
434 struct sysdev_class_attribute *node_state_attr[] = {
435         &attr_possible,
436         &attr_online,
437         &attr_has_normal_memory,
438 #ifdef CONFIG_HIGHMEM
439         &attr_has_high_memory,
440 #endif
441         &attr_has_cpu,
442 };
443
444 static int node_states_init(void)
445 {
446         int i;
447         int err = 0;
448
449         for (i = 0;  i < NR_NODE_STATES; i++) {
450                 int ret;
451                 ret = sysdev_class_create_file(&node_class, node_state_attr[i]);
452                 if (!err)
453                         err = ret;
454         }
455         return err;
456 }
457
458 static int __init register_node_type(void)
459 {
460         int ret;
461
462         ret = sysdev_class_register(&node_class);
463         if (!ret)
464                 ret = node_states_init();
465
466         /*
467          * Note:  we're not going to unregister the node class if we fail
468          * to register the node state class attribute files.
469          */
470         return ret;
471 }
472 postcore_initcall(register_node_type);