1fe5536d404f134e53b61324f9350c11080337dc
[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 Shmem:          %8lu kB\n"
89                        "Node %d KernelStack:    %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                        nid, K(node_page_state(nid, NR_UNEVICTABLE)),
109                        nid, K(node_page_state(nid, NR_MLOCK)),
110 #ifdef CONFIG_HIGHMEM
111                        nid, K(i.totalhigh),
112                        nid, K(i.freehigh),
113                        nid, K(i.totalram - i.totalhigh),
114                        nid, K(i.freeram - i.freehigh),
115 #endif
116                        nid, K(node_page_state(nid, NR_FILE_DIRTY)),
117                        nid, K(node_page_state(nid, NR_WRITEBACK)),
118                        nid, K(node_page_state(nid, NR_FILE_PAGES)),
119                        nid, K(node_page_state(nid, NR_FILE_MAPPED)),
120                        nid, K(node_page_state(nid, NR_ANON_PAGES)),
121                        nid, K(node_page_state(nid, NR_SHMEM)),
122                        nid, node_page_state(nid, NR_KERNEL_STACK) *
123                                 THREAD_SIZE / 1024,
124                        nid, K(node_page_state(nid, NR_PAGETABLE)),
125                        nid, K(node_page_state(nid, NR_UNSTABLE_NFS)),
126                        nid, K(node_page_state(nid, NR_BOUNCE)),
127                        nid, K(node_page_state(nid, NR_WRITEBACK_TEMP)),
128                        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)));
132         n += hugetlb_report_node_meminfo(nid, buf + n);
133         return n;
134 }
135
136 #undef K
137 static SYSDEV_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
138
139 static ssize_t node_read_numastat(struct sys_device * dev,
140                                 struct sysdev_attribute *attr, char * buf)
141 {
142         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",
149                        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));
155 }
156 static SYSDEV_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
157
158 static ssize_t node_read_distance(struct sys_device * dev,
159                         struct sysdev_attribute *attr, char * buf)
160 {
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 }
174 static SYSDEV_ATTR(distance, S_IRUGO, node_read_distance, NULL);
175
176
177 /*
178  * register_node - Setup a sysfs device for a node.
179  * @num - Node number to use when creating the device.
180  *
181  * Initialize and register the node device.
182  */
183 int register_node(struct node *node, int num, struct node *parent)
184 {
185         int error;
186
187         node->sysdev.id = num;
188         node->sysdev.cls = &node_class;
189         error = sysdev_register(&node->sysdev);
190
191         if (!error){
192                 sysdev_create_file(&node->sysdev, &attr_cpumap);
193                 sysdev_create_file(&node->sysdev, &attr_cpulist);
194                 sysdev_create_file(&node->sysdev, &attr_meminfo);
195                 sysdev_create_file(&node->sysdev, &attr_numastat);
196                 sysdev_create_file(&node->sysdev, &attr_distance);
197
198                 scan_unevictable_register_node(node);
199         }
200         return error;
201 }
202
203 /**
204  * unregister_node - unregister a node device
205  * @node: node going away
206  *
207  * Unregisters a node device @node.  All the devices on the node must be
208  * unregistered before calling this function.
209  */
210 void unregister_node(struct node *node)
211 {
212         sysdev_remove_file(&node->sysdev, &attr_cpumap);
213         sysdev_remove_file(&node->sysdev, &attr_cpulist);
214         sysdev_remove_file(&node->sysdev, &attr_meminfo);
215         sysdev_remove_file(&node->sysdev, &attr_numastat);
216         sysdev_remove_file(&node->sysdev, &attr_distance);
217
218         scan_unevictable_unregister_node(node);
219
220         sysdev_unregister(&node->sysdev);
221 }
222
223 struct node node_devices[MAX_NUMNODES];
224
225 /*
226  * register cpu under node
227  */
228 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
229 {
230         if (node_online(nid)) {
231                 struct sys_device *obj = get_cpu_sysdev(cpu);
232                 if (!obj)
233                         return 0;
234                 return sysfs_create_link(&node_devices[nid].sysdev.kobj,
235                                          &obj->kobj,
236                                          kobject_name(&obj->kobj));
237          }
238
239         return 0;
240 }
241
242 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
243 {
244         if (node_online(nid)) {
245                 struct sys_device *obj = get_cpu_sysdev(cpu);
246                 if (obj)
247                         sysfs_remove_link(&node_devices[nid].sysdev.kobj,
248                                          kobject_name(&obj->kobj));
249         }
250         return 0;
251 }
252
253 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
254 #define page_initialized(page)  (page->lru.next)
255
256 static int get_nid_for_pfn(unsigned long pfn)
257 {
258         struct page *page;
259
260         if (!pfn_valid_within(pfn))
261                 return -1;
262         page = pfn_to_page(pfn);
263         if (!page_initialized(page))
264                 return -1;
265         return pfn_to_nid(pfn);
266 }
267
268 /* register memory section under specified node if it spans that node */
269 int register_mem_sect_under_node(struct memory_block *mem_blk, int nid)
270 {
271         unsigned long pfn, sect_start_pfn, sect_end_pfn;
272
273         if (!mem_blk)
274                 return -EFAULT;
275         if (!node_online(nid))
276                 return 0;
277         sect_start_pfn = section_nr_to_pfn(mem_blk->phys_index);
278         sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
279         for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
280                 int page_nid;
281
282                 page_nid = get_nid_for_pfn(pfn);
283                 if (page_nid < 0)
284                         continue;
285                 if (page_nid != nid)
286                         continue;
287                 return sysfs_create_link_nowarn(&node_devices[nid].sysdev.kobj,
288                                         &mem_blk->sysdev.kobj,
289                                         kobject_name(&mem_blk->sysdev.kobj));
290         }
291         /* mem section does not span the specified node */
292         return 0;
293 }
294
295 /* unregister memory section under all nodes that it spans */
296 int unregister_mem_sect_under_nodes(struct memory_block *mem_blk)
297 {
298         nodemask_t unlinked_nodes;
299         unsigned long pfn, sect_start_pfn, sect_end_pfn;
300
301         if (!mem_blk)
302                 return -EFAULT;
303         nodes_clear(unlinked_nodes);
304         sect_start_pfn = section_nr_to_pfn(mem_blk->phys_index);
305         sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
306         for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
307                 int nid;
308
309                 nid = get_nid_for_pfn(pfn);
310                 if (nid < 0)
311                         continue;
312                 if (!node_online(nid))
313                         continue;
314                 if (node_test_and_set(nid, unlinked_nodes))
315                         continue;
316                 sysfs_remove_link(&node_devices[nid].sysdev.kobj,
317                          kobject_name(&mem_blk->sysdev.kobj));
318         }
319         return 0;
320 }
321
322 static int link_mem_sections(int nid)
323 {
324         unsigned long start_pfn = NODE_DATA(nid)->node_start_pfn;
325         unsigned long end_pfn = start_pfn + NODE_DATA(nid)->node_spanned_pages;
326         unsigned long pfn;
327         int err = 0;
328
329         for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
330                 unsigned long section_nr = pfn_to_section_nr(pfn);
331                 struct mem_section *mem_sect;
332                 struct memory_block *mem_blk;
333                 int ret;
334
335                 if (!present_section_nr(section_nr))
336                         continue;
337                 mem_sect = __nr_to_section(section_nr);
338                 mem_blk = find_memory_block(mem_sect);
339                 ret = register_mem_sect_under_node(mem_blk, nid);
340                 if (!err)
341                         err = ret;
342
343                 /* discard ref obtained in find_memory_block() */
344                 kobject_put(&mem_blk->sysdev.kobj);
345         }
346         return err;
347 }
348 #else
349 static int link_mem_sections(int nid) { return 0; }
350 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
351
352 int register_one_node(int nid)
353 {
354         int error = 0;
355         int cpu;
356
357         if (node_online(nid)) {
358                 int p_node = parent_node(nid);
359                 struct node *parent = NULL;
360
361                 if (p_node != nid)
362                         parent = &node_devices[p_node];
363
364                 error = register_node(&node_devices[nid], nid, parent);
365
366                 /* link cpu under this node */
367                 for_each_present_cpu(cpu) {
368                         if (cpu_to_node(cpu) == nid)
369                                 register_cpu_under_node(cpu, nid);
370                 }
371
372                 /* link memory sections under this node */
373                 error = link_mem_sections(nid);
374         }
375
376         return error;
377
378 }
379
380 void unregister_one_node(int nid)
381 {
382         unregister_node(&node_devices[nid]);
383 }
384
385 /*
386  * node states attributes
387  */
388
389 static ssize_t print_nodes_state(enum node_states state, char *buf)
390 {
391         int n;
392
393         n = nodelist_scnprintf(buf, PAGE_SIZE, node_states[state]);
394         if (n > 0 && PAGE_SIZE > n + 1) {
395                 *(buf + n++) = '\n';
396                 *(buf + n++) = '\0';
397         }
398         return n;
399 }
400
401 static ssize_t print_nodes_possible(struct sysdev_class *class, char *buf)
402 {
403         return print_nodes_state(N_POSSIBLE, buf);
404 }
405
406 static ssize_t print_nodes_online(struct sysdev_class *class, char *buf)
407 {
408         return print_nodes_state(N_ONLINE, buf);
409 }
410
411 static ssize_t print_nodes_has_normal_memory(struct sysdev_class *class,
412                                                 char *buf)
413 {
414         return print_nodes_state(N_NORMAL_MEMORY, buf);
415 }
416
417 static ssize_t print_nodes_has_cpu(struct sysdev_class *class, char *buf)
418 {
419         return print_nodes_state(N_CPU, buf);
420 }
421
422 static SYSDEV_CLASS_ATTR(possible, 0444, print_nodes_possible, NULL);
423 static SYSDEV_CLASS_ATTR(online, 0444, print_nodes_online, NULL);
424 static SYSDEV_CLASS_ATTR(has_normal_memory, 0444, print_nodes_has_normal_memory,
425                                                                         NULL);
426 static SYSDEV_CLASS_ATTR(has_cpu, 0444, print_nodes_has_cpu, NULL);
427
428 #ifdef CONFIG_HIGHMEM
429 static ssize_t print_nodes_has_high_memory(struct sysdev_class *class,
430                                                  char *buf)
431 {
432         return print_nodes_state(N_HIGH_MEMORY, buf);
433 }
434
435 static SYSDEV_CLASS_ATTR(has_high_memory, 0444, print_nodes_has_high_memory,
436                                                                          NULL);
437 #endif
438
439 struct sysdev_class_attribute *node_state_attr[] = {
440         &attr_possible,
441         &attr_online,
442         &attr_has_normal_memory,
443 #ifdef CONFIG_HIGHMEM
444         &attr_has_high_memory,
445 #endif
446         &attr_has_cpu,
447 };
448
449 static int node_states_init(void)
450 {
451         int i;
452         int err = 0;
453
454         for (i = 0;  i < NR_NODE_STATES; i++) {
455                 int ret;
456                 ret = sysdev_class_create_file(&node_class, node_state_attr[i]);
457                 if (!err)
458                         err = ret;
459         }
460         return err;
461 }
462
463 static int __init register_node_type(void)
464 {
465         int ret;
466
467         ret = sysdev_class_register(&node_class);
468         if (!ret)
469                 ret = node_states_init();
470
471         /*
472          * Note:  we're not going to unregister the node class if we fail
473          * to register the node state class attribute files.
474          */
475         return ret;
476 }
477 postcore_initcall(register_node_type);