hugetlb: add per node hstate attributes
[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 #ifdef CONFIG_HUGETLBFS
177 /*
178  * hugetlbfs per node attributes registration interface:
179  * When/if hugetlb[fs] subsystem initializes [sometime after this module],
180  * it will register its per node attributes for all nodes online at that
181  * time.  It will also call register_hugetlbfs_with_node(), below, to
182  * 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  */
186 static node_registration_func_t __hugetlb_register_node;
187 static node_registration_func_t __hugetlb_unregister_node;
188
189 static inline void hugetlb_register_node(struct node *node)
190 {
191         if (__hugetlb_register_node)
192                 __hugetlb_register_node(node);
193 }
194
195 static inline void hugetlb_unregister_node(struct node *node)
196 {
197         if (__hugetlb_unregister_node)
198                 __hugetlb_unregister_node(node);
199 }
200
201 void register_hugetlbfs_with_node(node_registration_func_t doregister,
202                                   node_registration_func_t unregister)
203 {
204         __hugetlb_register_node   = doregister;
205         __hugetlb_unregister_node = unregister;
206 }
207 #else
208 static inline void hugetlb_register_node(struct node *node) {}
209
210 static inline void hugetlb_unregister_node(struct node *node) {}
211 #endif
212
213
214 /*
215  * register_node - Setup a sysfs device for a node.
216  * @num - Node number to use when creating the device.
217  *
218  * Initialize and register the node device.
219  */
220 int register_node(struct node *node, int num, struct node *parent)
221 {
222         int error;
223
224         node->sysdev.id = num;
225         node->sysdev.cls = &node_class;
226         error = sysdev_register(&node->sysdev);
227
228         if (!error){
229                 sysdev_create_file(&node->sysdev, &attr_cpumap);
230                 sysdev_create_file(&node->sysdev, &attr_cpulist);
231                 sysdev_create_file(&node->sysdev, &attr_meminfo);
232                 sysdev_create_file(&node->sysdev, &attr_numastat);
233                 sysdev_create_file(&node->sysdev, &attr_distance);
234
235                 scan_unevictable_register_node(node);
236                 hugetlb_register_node(node);
237         }
238         return error;
239 }
240
241 /**
242  * unregister_node - unregister a node device
243  * @node: node going away
244  *
245  * Unregisters a node device @node.  All the devices on the node must be
246  * unregistered before calling this function.
247  */
248 void unregister_node(struct node *node)
249 {
250         sysdev_remove_file(&node->sysdev, &attr_cpumap);
251         sysdev_remove_file(&node->sysdev, &attr_cpulist);
252         sysdev_remove_file(&node->sysdev, &attr_meminfo);
253         sysdev_remove_file(&node->sysdev, &attr_numastat);
254         sysdev_remove_file(&node->sysdev, &attr_distance);
255
256         scan_unevictable_unregister_node(node);
257         hugetlb_unregister_node(node);
258
259         sysdev_unregister(&node->sysdev);
260 }
261
262 struct node node_devices[MAX_NUMNODES];
263
264 /*
265  * register cpu under node
266  */
267 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
268 {
269         if (node_online(nid)) {
270                 struct sys_device *obj = get_cpu_sysdev(cpu);
271                 if (!obj)
272                         return 0;
273                 return sysfs_create_link(&node_devices[nid].sysdev.kobj,
274                                          &obj->kobj,
275                                          kobject_name(&obj->kobj));
276          }
277
278         return 0;
279 }
280
281 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
282 {
283         if (node_online(nid)) {
284                 struct sys_device *obj = get_cpu_sysdev(cpu);
285                 if (obj)
286                         sysfs_remove_link(&node_devices[nid].sysdev.kobj,
287                                          kobject_name(&obj->kobj));
288         }
289         return 0;
290 }
291
292 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
293 #define page_initialized(page)  (page->lru.next)
294
295 static int get_nid_for_pfn(unsigned long pfn)
296 {
297         struct page *page;
298
299         if (!pfn_valid_within(pfn))
300                 return -1;
301         page = pfn_to_page(pfn);
302         if (!page_initialized(page))
303                 return -1;
304         return pfn_to_nid(pfn);
305 }
306
307 /* register memory section under specified node if it spans that node */
308 int register_mem_sect_under_node(struct memory_block *mem_blk, int nid)
309 {
310         unsigned long pfn, sect_start_pfn, sect_end_pfn;
311
312         if (!mem_blk)
313                 return -EFAULT;
314         if (!node_online(nid))
315                 return 0;
316         sect_start_pfn = section_nr_to_pfn(mem_blk->phys_index);
317         sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
318         for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
319                 int page_nid;
320
321                 page_nid = get_nid_for_pfn(pfn);
322                 if (page_nid < 0)
323                         continue;
324                 if (page_nid != nid)
325                         continue;
326                 return sysfs_create_link_nowarn(&node_devices[nid].sysdev.kobj,
327                                         &mem_blk->sysdev.kobj,
328                                         kobject_name(&mem_blk->sysdev.kobj));
329         }
330         /* mem section does not span the specified node */
331         return 0;
332 }
333
334 /* unregister memory section under all nodes that it spans */
335 int unregister_mem_sect_under_nodes(struct memory_block *mem_blk)
336 {
337         nodemask_t unlinked_nodes;
338         unsigned long pfn, sect_start_pfn, sect_end_pfn;
339
340         if (!mem_blk)
341                 return -EFAULT;
342         nodes_clear(unlinked_nodes);
343         sect_start_pfn = section_nr_to_pfn(mem_blk->phys_index);
344         sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
345         for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
346                 int nid;
347
348                 nid = get_nid_for_pfn(pfn);
349                 if (nid < 0)
350                         continue;
351                 if (!node_online(nid))
352                         continue;
353                 if (node_test_and_set(nid, unlinked_nodes))
354                         continue;
355                 sysfs_remove_link(&node_devices[nid].sysdev.kobj,
356                          kobject_name(&mem_blk->sysdev.kobj));
357         }
358         return 0;
359 }
360
361 static int link_mem_sections(int nid)
362 {
363         unsigned long start_pfn = NODE_DATA(nid)->node_start_pfn;
364         unsigned long end_pfn = start_pfn + NODE_DATA(nid)->node_spanned_pages;
365         unsigned long pfn;
366         int err = 0;
367
368         for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
369                 unsigned long section_nr = pfn_to_section_nr(pfn);
370                 struct mem_section *mem_sect;
371                 struct memory_block *mem_blk;
372                 int ret;
373
374                 if (!present_section_nr(section_nr))
375                         continue;
376                 mem_sect = __nr_to_section(section_nr);
377                 mem_blk = find_memory_block(mem_sect);
378                 ret = register_mem_sect_under_node(mem_blk, nid);
379                 if (!err)
380                         err = ret;
381
382                 /* discard ref obtained in find_memory_block() */
383                 kobject_put(&mem_blk->sysdev.kobj);
384         }
385         return err;
386 }
387 #else
388 static int link_mem_sections(int nid) { return 0; }
389 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
390
391 int register_one_node(int nid)
392 {
393         int error = 0;
394         int cpu;
395
396         if (node_online(nid)) {
397                 int p_node = parent_node(nid);
398                 struct node *parent = NULL;
399
400                 if (p_node != nid)
401                         parent = &node_devices[p_node];
402
403                 error = register_node(&node_devices[nid], nid, parent);
404
405                 /* link cpu under this node */
406                 for_each_present_cpu(cpu) {
407                         if (cpu_to_node(cpu) == nid)
408                                 register_cpu_under_node(cpu, nid);
409                 }
410
411                 /* link memory sections under this node */
412                 error = link_mem_sections(nid);
413         }
414
415         return error;
416
417 }
418
419 void unregister_one_node(int nid)
420 {
421         unregister_node(&node_devices[nid]);
422 }
423
424 /*
425  * node states attributes
426  */
427
428 static ssize_t print_nodes_state(enum node_states state, char *buf)
429 {
430         int n;
431
432         n = nodelist_scnprintf(buf, PAGE_SIZE, node_states[state]);
433         if (n > 0 && PAGE_SIZE > n + 1) {
434                 *(buf + n++) = '\n';
435                 *(buf + n++) = '\0';
436         }
437         return n;
438 }
439
440 static ssize_t print_nodes_possible(struct sysdev_class *class, char *buf)
441 {
442         return print_nodes_state(N_POSSIBLE, buf);
443 }
444
445 static ssize_t print_nodes_online(struct sysdev_class *class, char *buf)
446 {
447         return print_nodes_state(N_ONLINE, buf);
448 }
449
450 static ssize_t print_nodes_has_normal_memory(struct sysdev_class *class,
451                                                 char *buf)
452 {
453         return print_nodes_state(N_NORMAL_MEMORY, buf);
454 }
455
456 static ssize_t print_nodes_has_cpu(struct sysdev_class *class, char *buf)
457 {
458         return print_nodes_state(N_CPU, buf);
459 }
460
461 static SYSDEV_CLASS_ATTR(possible, 0444, print_nodes_possible, NULL);
462 static SYSDEV_CLASS_ATTR(online, 0444, print_nodes_online, NULL);
463 static SYSDEV_CLASS_ATTR(has_normal_memory, 0444, print_nodes_has_normal_memory,
464                                                                         NULL);
465 static SYSDEV_CLASS_ATTR(has_cpu, 0444, print_nodes_has_cpu, NULL);
466
467 #ifdef CONFIG_HIGHMEM
468 static ssize_t print_nodes_has_high_memory(struct sysdev_class *class,
469                                                  char *buf)
470 {
471         return print_nodes_state(N_HIGH_MEMORY, buf);
472 }
473
474 static SYSDEV_CLASS_ATTR(has_high_memory, 0444, print_nodes_has_high_memory,
475                                                                          NULL);
476 #endif
477
478 struct sysdev_class_attribute *node_state_attr[] = {
479         &attr_possible,
480         &attr_online,
481         &attr_has_normal_memory,
482 #ifdef CONFIG_HIGHMEM
483         &attr_has_high_memory,
484 #endif
485         &attr_has_cpu,
486 };
487
488 static int node_states_init(void)
489 {
490         int i;
491         int err = 0;
492
493         for (i = 0;  i < NR_NODE_STATES; i++) {
494                 int ret;
495                 ret = sysdev_class_create_file(&node_class, node_state_attr[i]);
496                 if (!err)
497                         err = ret;
498         }
499         return err;
500 }
501
502 static int __init register_node_type(void)
503 {
504         int ret;
505
506         ret = sysdev_class_register(&node_class);
507         if (!ret)
508                 ret = node_states_init();
509
510         /*
511          * Note:  we're not going to unregister the node class if we fail
512          * to register the node state class attribute files.
513          */
514         return ret;
515 }
516 postcore_initcall(register_node_type);