Merge branch 'master' into next
[linux-2.6] / 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         node_to_cpumask_ptr(mask, 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 #ifdef CONFIG_UNEVICTABLE_LRU
76                        "Node %d Unevictable:    %8lu kB\n"
77                        "Node %d Mlocked:        %8lu kB\n"
78 #endif
79 #ifdef CONFIG_HIGHMEM
80                        "Node %d HighTotal:      %8lu kB\n"
81                        "Node %d HighFree:       %8lu kB\n"
82                        "Node %d LowTotal:       %8lu kB\n"
83                        "Node %d LowFree:        %8lu kB\n"
84 #endif
85                        "Node %d Dirty:          %8lu kB\n"
86                        "Node %d Writeback:      %8lu kB\n"
87                        "Node %d FilePages:      %8lu kB\n"
88                        "Node %d Mapped:         %8lu kB\n"
89                        "Node %d AnonPages:      %8lu kB\n"
90                        "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",
97                        nid, K(i.totalram),
98                        nid, K(i.freeram),
99                        nid, K(i.totalram - i.freeram),
100                        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)),
108 #ifdef CONFIG_UNEVICTABLE_LRU
109                        nid, K(node_page_state(nid, NR_UNEVICTABLE)),
110                        nid, K(node_page_state(nid, NR_MLOCK)),
111 #endif
112 #ifdef CONFIG_HIGHMEM
113                        nid, K(i.totalhigh),
114                        nid, K(i.freehigh),
115                        nid, K(i.totalram - i.totalhigh),
116                        nid, K(i.freeram - i.freehigh),
117 #endif
118                        nid, K(node_page_state(nid, NR_FILE_DIRTY)),
119                        nid, K(node_page_state(nid, NR_WRITEBACK)),
120                        nid, K(node_page_state(nid, NR_FILE_PAGES)),
121                        nid, K(node_page_state(nid, NR_FILE_MAPPED)),
122                        nid, K(node_page_state(nid, NR_ANON_PAGES)),
123                        nid, K(node_page_state(nid, NR_PAGETABLE)),
124                        nid, K(node_page_state(nid, NR_UNSTABLE_NFS)),
125                        nid, K(node_page_state(nid, NR_BOUNCE)),
126                        nid, K(node_page_state(nid, NR_WRITEBACK_TEMP)),
127                        nid, K(node_page_state(nid, NR_SLAB_RECLAIMABLE) +
128                                 node_page_state(nid, NR_SLAB_UNRECLAIMABLE)),
129                        nid, K(node_page_state(nid, NR_SLAB_RECLAIMABLE)),
130                        nid, K(node_page_state(nid, NR_SLAB_UNRECLAIMABLE)));
131         n += hugetlb_report_node_meminfo(nid, buf + n);
132         return n;
133 }
134
135 #undef K
136 static SYSDEV_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
137
138 static ssize_t node_read_numastat(struct sys_device * dev,
139                                 struct sysdev_attribute *attr, char * buf)
140 {
141         return sprintf(buf,
142                        "numa_hit %lu\n"
143                        "numa_miss %lu\n"
144                        "numa_foreign %lu\n"
145                        "interleave_hit %lu\n"
146                        "local_node %lu\n"
147                        "other_node %lu\n",
148                        node_page_state(dev->id, NUMA_HIT),
149                        node_page_state(dev->id, NUMA_MISS),
150                        node_page_state(dev->id, NUMA_FOREIGN),
151                        node_page_state(dev->id, NUMA_INTERLEAVE_HIT),
152                        node_page_state(dev->id, NUMA_LOCAL),
153                        node_page_state(dev->id, NUMA_OTHER));
154 }
155 static SYSDEV_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
156
157 static ssize_t node_read_distance(struct sys_device * dev,
158                         struct sysdev_attribute *attr, char * buf)
159 {
160         int nid = dev->id;
161         int len = 0;
162         int i;
163
164         /* buf currently PAGE_SIZE, need ~4 chars per node */
165         BUILD_BUG_ON(MAX_NUMNODES*4 > PAGE_SIZE/2);
166
167         for_each_online_node(i)
168                 len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i));
169
170         len += sprintf(buf + len, "\n");
171         return len;
172 }
173 static SYSDEV_ATTR(distance, S_IRUGO, node_read_distance, NULL);
174
175
176 /*
177  * register_node - Setup a sysfs device for a node.
178  * @num - Node number to use when creating the device.
179  *
180  * Initialize and register the node device.
181  */
182 int register_node(struct node *node, int num, struct node *parent)
183 {
184         int error;
185
186         node->sysdev.id = num;
187         node->sysdev.cls = &node_class;
188         error = sysdev_register(&node->sysdev);
189
190         if (!error){
191                 sysdev_create_file(&node->sysdev, &attr_cpumap);
192                 sysdev_create_file(&node->sysdev, &attr_cpulist);
193                 sysdev_create_file(&node->sysdev, &attr_meminfo);
194                 sysdev_create_file(&node->sysdev, &attr_numastat);
195                 sysdev_create_file(&node->sysdev, &attr_distance);
196
197                 scan_unevictable_register_node(node);
198         }
199         return error;
200 }
201
202 /**
203  * unregister_node - unregister a node device
204  * @node: node going away
205  *
206  * Unregisters a node device @node.  All the devices on the node must be
207  * unregistered before calling this function.
208  */
209 void unregister_node(struct node *node)
210 {
211         sysdev_remove_file(&node->sysdev, &attr_cpumap);
212         sysdev_remove_file(&node->sysdev, &attr_cpulist);
213         sysdev_remove_file(&node->sysdev, &attr_meminfo);
214         sysdev_remove_file(&node->sysdev, &attr_numastat);
215         sysdev_remove_file(&node->sysdev, &attr_distance);
216
217         scan_unevictable_unregister_node(node);
218
219         sysdev_unregister(&node->sysdev);
220 }
221
222 struct node node_devices[MAX_NUMNODES];
223
224 /*
225  * register cpu under node
226  */
227 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
228 {
229         if (node_online(nid)) {
230                 struct sys_device *obj = get_cpu_sysdev(cpu);
231                 if (!obj)
232                         return 0;
233                 return sysfs_create_link(&node_devices[nid].sysdev.kobj,
234                                          &obj->kobj,
235                                          kobject_name(&obj->kobj));
236          }
237
238         return 0;
239 }
240
241 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
242 {
243         if (node_online(nid)) {
244                 struct sys_device *obj = get_cpu_sysdev(cpu);
245                 if (obj)
246                         sysfs_remove_link(&node_devices[nid].sysdev.kobj,
247                                          kobject_name(&obj->kobj));
248         }
249         return 0;
250 }
251
252 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
253 #define page_initialized(page)  (page->lru.next)
254
255 static int get_nid_for_pfn(unsigned long pfn)
256 {
257         struct page *page;
258
259         if (!pfn_valid_within(pfn))
260                 return -1;
261         page = pfn_to_page(pfn);
262         if (!page_initialized(page))
263                 return -1;
264         return pfn_to_nid(pfn);
265 }
266
267 /* register memory section under specified node if it spans that node */
268 int register_mem_sect_under_node(struct memory_block *mem_blk, int nid)
269 {
270         unsigned long pfn, sect_start_pfn, sect_end_pfn;
271
272         if (!mem_blk)
273                 return -EFAULT;
274         if (!node_online(nid))
275                 return 0;
276         sect_start_pfn = section_nr_to_pfn(mem_blk->phys_index);
277         sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
278         for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
279                 int page_nid;
280
281                 page_nid = get_nid_for_pfn(pfn);
282                 if (page_nid < 0)
283                         continue;
284                 if (page_nid != nid)
285                         continue;
286                 return sysfs_create_link_nowarn(&node_devices[nid].sysdev.kobj,
287                                         &mem_blk->sysdev.kobj,
288                                         kobject_name(&mem_blk->sysdev.kobj));
289         }
290         /* mem section does not span the specified node */
291         return 0;
292 }
293
294 /* unregister memory section under all nodes that it spans */
295 int unregister_mem_sect_under_nodes(struct memory_block *mem_blk)
296 {
297         nodemask_t unlinked_nodes;
298         unsigned long pfn, sect_start_pfn, sect_end_pfn;
299
300         if (!mem_blk)
301                 return -EFAULT;
302         nodes_clear(unlinked_nodes);
303         sect_start_pfn = section_nr_to_pfn(mem_blk->phys_index);
304         sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
305         for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
306                 unsigned int nid;
307
308                 nid = get_nid_for_pfn(pfn);
309                 if (nid < 0)
310                         continue;
311                 if (!node_online(nid))
312                         continue;
313                 if (node_test_and_set(nid, unlinked_nodes))
314                         continue;
315                 sysfs_remove_link(&node_devices[nid].sysdev.kobj,
316                          kobject_name(&mem_blk->sysdev.kobj));
317         }
318         return 0;
319 }
320
321 static int link_mem_sections(int nid)
322 {
323         unsigned long start_pfn = NODE_DATA(nid)->node_start_pfn;
324         unsigned long end_pfn = start_pfn + NODE_DATA(nid)->node_spanned_pages;
325         unsigned long pfn;
326         int err = 0;
327
328         for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
329                 unsigned long section_nr = pfn_to_section_nr(pfn);
330                 struct mem_section *mem_sect;
331                 struct memory_block *mem_blk;
332                 int ret;
333
334                 if (!present_section_nr(section_nr))
335                         continue;
336                 mem_sect = __nr_to_section(section_nr);
337                 mem_blk = find_memory_block(mem_sect);
338                 ret = register_mem_sect_under_node(mem_blk, nid);
339                 if (!err)
340                         err = ret;
341
342                 /* discard ref obtained in find_memory_block() */
343                 kobject_put(&mem_blk->sysdev.kobj);
344         }
345         return err;
346 }
347 #else
348 static int link_mem_sections(int nid) { return 0; }
349 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
350
351 int register_one_node(int nid)
352 {
353         int error = 0;
354         int cpu;
355
356         if (node_online(nid)) {
357                 int p_node = parent_node(nid);
358                 struct node *parent = NULL;
359
360                 if (p_node != nid)
361                         parent = &node_devices[p_node];
362
363                 error = register_node(&node_devices[nid], nid, parent);
364
365                 /* link cpu under this node */
366                 for_each_present_cpu(cpu) {
367                         if (cpu_to_node(cpu) == nid)
368                                 register_cpu_under_node(cpu, nid);
369                 }
370
371                 /* link memory sections under this node */
372                 error = link_mem_sections(nid);
373         }
374
375         return error;
376
377 }
378
379 void unregister_one_node(int nid)
380 {
381         unregister_node(&node_devices[nid]);
382 }
383
384 /*
385  * node states attributes
386  */
387
388 static ssize_t print_nodes_state(enum node_states state, char *buf)
389 {
390         int n;
391
392         n = nodelist_scnprintf(buf, PAGE_SIZE, node_states[state]);
393         if (n > 0 && PAGE_SIZE > n + 1) {
394                 *(buf + n++) = '\n';
395                 *(buf + n++) = '\0';
396         }
397         return n;
398 }
399
400 static ssize_t print_nodes_possible(struct sysdev_class *class, char *buf)
401 {
402         return print_nodes_state(N_POSSIBLE, buf);
403 }
404
405 static ssize_t print_nodes_online(struct sysdev_class *class, char *buf)
406 {
407         return print_nodes_state(N_ONLINE, buf);
408 }
409
410 static ssize_t print_nodes_has_normal_memory(struct sysdev_class *class,
411                                                 char *buf)
412 {
413         return print_nodes_state(N_NORMAL_MEMORY, buf);
414 }
415
416 static ssize_t print_nodes_has_cpu(struct sysdev_class *class, char *buf)
417 {
418         return print_nodes_state(N_CPU, buf);
419 }
420
421 static SYSDEV_CLASS_ATTR(possible, 0444, print_nodes_possible, NULL);
422 static SYSDEV_CLASS_ATTR(online, 0444, print_nodes_online, NULL);
423 static SYSDEV_CLASS_ATTR(has_normal_memory, 0444, print_nodes_has_normal_memory,
424                                                                         NULL);
425 static SYSDEV_CLASS_ATTR(has_cpu, 0444, print_nodes_has_cpu, NULL);
426
427 #ifdef CONFIG_HIGHMEM
428 static ssize_t print_nodes_has_high_memory(struct sysdev_class *class,
429                                                  char *buf)
430 {
431         return print_nodes_state(N_HIGH_MEMORY, buf);
432 }
433
434 static SYSDEV_CLASS_ATTR(has_high_memory, 0444, print_nodes_has_high_memory,
435                                                                          NULL);
436 #endif
437
438 struct sysdev_class_attribute *node_state_attr[] = {
439         &attr_possible,
440         &attr_online,
441         &attr_has_normal_memory,
442 #ifdef CONFIG_HIGHMEM
443         &attr_has_high_memory,
444 #endif
445         &attr_has_cpu,
446 };
447
448 static int node_states_init(void)
449 {
450         int i;
451         int err = 0;
452
453         for (i = 0;  i < NR_NODE_STATES; i++) {
454                 int ret;
455                 ret = sysdev_class_create_file(&node_class, node_state_attr[i]);
456                 if (!err)
457                         err = ret;
458         }
459         return err;
460 }
461
462 static int __init register_node_type(void)
463 {
464         int ret;
465
466         ret = sysdev_class_register(&node_class);
467         if (!ret)
468                 ret = node_states_init();
469
470         /*
471          * Note:  we're not going to unregister the node class if we fail
472          * to register the node state class attribute files.
473          */
474         return ret;
475 }
476 postcore_initcall(register_node_type);