hugetlb: handle memory hot-plug events
[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 online nodes with
181  * memory.  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                         node_state(node->sysdev.id, N_HIGH_MEMORY))
193                 __hugetlb_register_node(node);
194 }
195
196 static inline void hugetlb_unregister_node(struct node *node)
197 {
198         if (__hugetlb_unregister_node)
199                 __hugetlb_unregister_node(node);
200 }
201
202 void register_hugetlbfs_with_node(node_registration_func_t doregister,
203                                   node_registration_func_t unregister)
204 {
205         __hugetlb_register_node   = doregister;
206         __hugetlb_unregister_node = unregister;
207 }
208 #else
209 static inline void hugetlb_register_node(struct node *node) {}
210
211 static inline void hugetlb_unregister_node(struct node *node) {}
212 #endif
213
214
215 /*
216  * register_node - Setup a sysfs device for a node.
217  * @num - Node number to use when creating the device.
218  *
219  * Initialize and register the node device.
220  */
221 int register_node(struct node *node, int num, struct node *parent)
222 {
223         int error;
224
225         node->sysdev.id = num;
226         node->sysdev.cls = &node_class;
227         error = sysdev_register(&node->sysdev);
228
229         if (!error){
230                 sysdev_create_file(&node->sysdev, &attr_cpumap);
231                 sysdev_create_file(&node->sysdev, &attr_cpulist);
232                 sysdev_create_file(&node->sysdev, &attr_meminfo);
233                 sysdev_create_file(&node->sysdev, &attr_numastat);
234                 sysdev_create_file(&node->sysdev, &attr_distance);
235
236                 scan_unevictable_register_node(node);
237
238                 hugetlb_register_node(node);
239         }
240         return error;
241 }
242
243 /**
244  * unregister_node - unregister a node device
245  * @node: node going away
246  *
247  * Unregisters a node device @node.  All the devices on the node must be
248  * unregistered before calling this function.
249  */
250 void unregister_node(struct node *node)
251 {
252         sysdev_remove_file(&node->sysdev, &attr_cpumap);
253         sysdev_remove_file(&node->sysdev, &attr_cpulist);
254         sysdev_remove_file(&node->sysdev, &attr_meminfo);
255         sysdev_remove_file(&node->sysdev, &attr_numastat);
256         sysdev_remove_file(&node->sysdev, &attr_distance);
257
258         scan_unevictable_unregister_node(node);
259         hugetlb_unregister_node(node);          /* no-op, if memoryless node */
260
261         sysdev_unregister(&node->sysdev);
262 }
263
264 struct node node_devices[MAX_NUMNODES];
265
266 /*
267  * register cpu under node
268  */
269 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
270 {
271         if (node_online(nid)) {
272                 struct sys_device *obj = get_cpu_sysdev(cpu);
273                 if (!obj)
274                         return 0;
275                 return sysfs_create_link(&node_devices[nid].sysdev.kobj,
276                                          &obj->kobj,
277                                          kobject_name(&obj->kobj));
278          }
279
280         return 0;
281 }
282
283 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
284 {
285         if (node_online(nid)) {
286                 struct sys_device *obj = get_cpu_sysdev(cpu);
287                 if (obj)
288                         sysfs_remove_link(&node_devices[nid].sysdev.kobj,
289                                          kobject_name(&obj->kobj));
290         }
291         return 0;
292 }
293
294 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
295 #define page_initialized(page)  (page->lru.next)
296
297 static int get_nid_for_pfn(unsigned long pfn)
298 {
299         struct page *page;
300
301         if (!pfn_valid_within(pfn))
302                 return -1;
303         page = pfn_to_page(pfn);
304         if (!page_initialized(page))
305                 return -1;
306         return pfn_to_nid(pfn);
307 }
308
309 /* register memory section under specified node if it spans that node */
310 int register_mem_sect_under_node(struct memory_block *mem_blk, int nid)
311 {
312         unsigned long pfn, sect_start_pfn, sect_end_pfn;
313
314         if (!mem_blk)
315                 return -EFAULT;
316         if (!node_online(nid))
317                 return 0;
318         sect_start_pfn = section_nr_to_pfn(mem_blk->phys_index);
319         sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
320         for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
321                 int page_nid;
322
323                 page_nid = get_nid_for_pfn(pfn);
324                 if (page_nid < 0)
325                         continue;
326                 if (page_nid != nid)
327                         continue;
328                 return sysfs_create_link_nowarn(&node_devices[nid].sysdev.kobj,
329                                         &mem_blk->sysdev.kobj,
330                                         kobject_name(&mem_blk->sysdev.kobj));
331         }
332         /* mem section does not span the specified node */
333         return 0;
334 }
335
336 /* unregister memory section under all nodes that it spans */
337 int unregister_mem_sect_under_nodes(struct memory_block *mem_blk)
338 {
339         nodemask_t unlinked_nodes;
340         unsigned long pfn, sect_start_pfn, sect_end_pfn;
341
342         if (!mem_blk)
343                 return -EFAULT;
344         nodes_clear(unlinked_nodes);
345         sect_start_pfn = section_nr_to_pfn(mem_blk->phys_index);
346         sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
347         for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
348                 int nid;
349
350                 nid = get_nid_for_pfn(pfn);
351                 if (nid < 0)
352                         continue;
353                 if (!node_online(nid))
354                         continue;
355                 if (node_test_and_set(nid, unlinked_nodes))
356                         continue;
357                 sysfs_remove_link(&node_devices[nid].sysdev.kobj,
358                          kobject_name(&mem_blk->sysdev.kobj));
359         }
360         return 0;
361 }
362
363 static int link_mem_sections(int nid)
364 {
365         unsigned long start_pfn = NODE_DATA(nid)->node_start_pfn;
366         unsigned long end_pfn = start_pfn + NODE_DATA(nid)->node_spanned_pages;
367         unsigned long pfn;
368         int err = 0;
369
370         for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
371                 unsigned long section_nr = pfn_to_section_nr(pfn);
372                 struct mem_section *mem_sect;
373                 struct memory_block *mem_blk;
374                 int ret;
375
376                 if (!present_section_nr(section_nr))
377                         continue;
378                 mem_sect = __nr_to_section(section_nr);
379                 mem_blk = find_memory_block(mem_sect);
380                 ret = register_mem_sect_under_node(mem_blk, nid);
381                 if (!err)
382                         err = ret;
383
384                 /* discard ref obtained in find_memory_block() */
385                 kobject_put(&mem_blk->sysdev.kobj);
386         }
387         return err;
388 }
389
390 /*
391  * Handle per node hstate attribute [un]registration on transistions
392  * to/from memoryless state.
393  */
394
395 static int node_memory_callback(struct notifier_block *self,
396                                 unsigned long action, void *arg)
397 {
398         struct memory_notify *mnb = arg;
399         int nid = mnb->status_change_nid;
400
401         switch (action) {
402         case MEM_ONLINE:    /* memory successfully brought online */
403                 if (nid != NUMA_NO_NODE)
404                         hugetlb_register_node(&node_devices[nid]);
405                 break;
406         case MEM_OFFLINE:   /* or offline */
407                 if (nid != NUMA_NO_NODE)
408                         hugetlb_unregister_node(&node_devices[nid]);
409                 break;
410         case MEM_GOING_ONLINE:
411         case MEM_GOING_OFFLINE:
412         case MEM_CANCEL_ONLINE:
413         case MEM_CANCEL_OFFLINE:
414         default:
415                 break;
416         }
417
418         return NOTIFY_OK;
419 }
420 #else
421 static int link_mem_sections(int nid) { return 0; }
422
423 static inline int node_memory_callback(struct notifier_block *self,
424                                 unsigned long action, void *arg)
425 {
426         return NOTIFY_OK;
427 }
428 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
429
430 int register_one_node(int nid)
431 {
432         int error = 0;
433         int cpu;
434
435         if (node_online(nid)) {
436                 int p_node = parent_node(nid);
437                 struct node *parent = NULL;
438
439                 if (p_node != nid)
440                         parent = &node_devices[p_node];
441
442                 error = register_node(&node_devices[nid], nid, parent);
443
444                 /* link cpu under this node */
445                 for_each_present_cpu(cpu) {
446                         if (cpu_to_node(cpu) == nid)
447                                 register_cpu_under_node(cpu, nid);
448                 }
449
450                 /* link memory sections under this node */
451                 error = link_mem_sections(nid);
452         }
453
454         return error;
455
456 }
457
458 void unregister_one_node(int nid)
459 {
460         unregister_node(&node_devices[nid]);
461 }
462
463 /*
464  * node states attributes
465  */
466
467 static ssize_t print_nodes_state(enum node_states state, char *buf)
468 {
469         int n;
470
471         n = nodelist_scnprintf(buf, PAGE_SIZE, node_states[state]);
472         if (n > 0 && PAGE_SIZE > n + 1) {
473                 *(buf + n++) = '\n';
474                 *(buf + n++) = '\0';
475         }
476         return n;
477 }
478
479 static ssize_t print_nodes_possible(struct sysdev_class *class, char *buf)
480 {
481         return print_nodes_state(N_POSSIBLE, buf);
482 }
483
484 static ssize_t print_nodes_online(struct sysdev_class *class, char *buf)
485 {
486         return print_nodes_state(N_ONLINE, buf);
487 }
488
489 static ssize_t print_nodes_has_normal_memory(struct sysdev_class *class,
490                                                 char *buf)
491 {
492         return print_nodes_state(N_NORMAL_MEMORY, buf);
493 }
494
495 static ssize_t print_nodes_has_cpu(struct sysdev_class *class, char *buf)
496 {
497         return print_nodes_state(N_CPU, buf);
498 }
499
500 static SYSDEV_CLASS_ATTR(possible, 0444, print_nodes_possible, NULL);
501 static SYSDEV_CLASS_ATTR(online, 0444, print_nodes_online, NULL);
502 static SYSDEV_CLASS_ATTR(has_normal_memory, 0444, print_nodes_has_normal_memory,
503                                                                         NULL);
504 static SYSDEV_CLASS_ATTR(has_cpu, 0444, print_nodes_has_cpu, NULL);
505
506 #ifdef CONFIG_HIGHMEM
507 static ssize_t print_nodes_has_high_memory(struct sysdev_class *class,
508                                                  char *buf)
509 {
510         return print_nodes_state(N_HIGH_MEMORY, buf);
511 }
512
513 static SYSDEV_CLASS_ATTR(has_high_memory, 0444, print_nodes_has_high_memory,
514                                                                          NULL);
515 #endif
516
517 struct sysdev_class_attribute *node_state_attr[] = {
518         &attr_possible,
519         &attr_online,
520         &attr_has_normal_memory,
521 #ifdef CONFIG_HIGHMEM
522         &attr_has_high_memory,
523 #endif
524         &attr_has_cpu,
525 };
526
527 static int node_states_init(void)
528 {
529         int i;
530         int err = 0;
531
532         for (i = 0;  i < NR_NODE_STATES; i++) {
533                 int ret;
534                 ret = sysdev_class_create_file(&node_class, node_state_attr[i]);
535                 if (!err)
536                         err = ret;
537         }
538         return err;
539 }
540
541 #define NODE_CALLBACK_PRI       2       /* lower than SLAB */
542 static int __init register_node_type(void)
543 {
544         int ret;
545
546         ret = sysdev_class_register(&node_class);
547         if (!ret) {
548                 ret = node_states_init();
549                 hotplug_memory_notifier(node_memory_callback,
550                                         NODE_CALLBACK_PRI);
551         }
552
553         /*
554          * Note:  we're not going to unregister the node class if we fail
555          * to register the node state class attribute files.
556          */
557         return ret;
558 }
559 postcore_initcall(register_node_type);