memory hotplug: Update phys_index to [start|end]_section_nr
[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/compaction.h>
13 #include <linux/cpumask.h>
14 #include <linux/topology.h>
15 #include <linux/nodemask.h>
16 #include <linux/cpu.h>
17 #include <linux/device.h>
18 #include <linux/swap.h>
19 #include <linux/slab.h>
20
21 static struct sysdev_class_attribute *node_state_attrs[];
22
23 static struct sysdev_class node_class = {
24         .name = "node",
25         .attrs = node_state_attrs,
26 };
27
28
29 static ssize_t node_read_cpumap(struct sys_device *dev, int type, char *buf)
30 {
31         struct node *node_dev = to_node(dev);
32         const struct cpumask *mask = cpumask_of_node(node_dev->sysdev.id);
33         int len;
34
35         /* 2008/04/07: buf currently PAGE_SIZE, need 9 chars per 32 bits. */
36         BUILD_BUG_ON((NR_CPUS/32 * 9) > (PAGE_SIZE-1));
37
38         len = type?
39                 cpulist_scnprintf(buf, PAGE_SIZE-2, mask) :
40                 cpumask_scnprintf(buf, PAGE_SIZE-2, mask);
41         buf[len++] = '\n';
42         buf[len] = '\0';
43         return len;
44 }
45
46 static inline ssize_t node_read_cpumask(struct sys_device *dev,
47                                 struct sysdev_attribute *attr, char *buf)
48 {
49         return node_read_cpumap(dev, 0, buf);
50 }
51 static inline ssize_t node_read_cpulist(struct sys_device *dev,
52                                 struct sysdev_attribute *attr, char *buf)
53 {
54         return node_read_cpumap(dev, 1, buf);
55 }
56
57 static SYSDEV_ATTR(cpumap,  S_IRUGO, node_read_cpumask, NULL);
58 static SYSDEV_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL);
59
60 #define K(x) ((x) << (PAGE_SHIFT - 10))
61 static ssize_t node_read_meminfo(struct sys_device * dev,
62                         struct sysdev_attribute *attr, char * buf)
63 {
64         int n;
65         int nid = dev->id;
66         struct sysinfo i;
67
68         si_meminfo_node(&i, nid);
69         n = sprintf(buf,
70                        "Node %d MemTotal:       %8lu kB\n"
71                        "Node %d MemFree:        %8lu kB\n"
72                        "Node %d MemUsed:        %8lu kB\n"
73                        "Node %d Active:         %8lu kB\n"
74                        "Node %d Inactive:       %8lu kB\n"
75                        "Node %d Active(anon):   %8lu kB\n"
76                        "Node %d Inactive(anon): %8lu kB\n"
77                        "Node %d Active(file):   %8lu kB\n"
78                        "Node %d Inactive(file): %8lu kB\n"
79                        "Node %d Unevictable:    %8lu kB\n"
80                        "Node %d Mlocked:        %8lu kB\n",
81                        nid, K(i.totalram),
82                        nid, K(i.freeram),
83                        nid, K(i.totalram - i.freeram),
84                        nid, K(node_page_state(nid, NR_ACTIVE_ANON) +
85                                 node_page_state(nid, NR_ACTIVE_FILE)),
86                        nid, K(node_page_state(nid, NR_INACTIVE_ANON) +
87                                 node_page_state(nid, NR_INACTIVE_FILE)),
88                        nid, K(node_page_state(nid, NR_ACTIVE_ANON)),
89                        nid, K(node_page_state(nid, NR_INACTIVE_ANON)),
90                        nid, K(node_page_state(nid, NR_ACTIVE_FILE)),
91                        nid, K(node_page_state(nid, NR_INACTIVE_FILE)),
92                        nid, K(node_page_state(nid, NR_UNEVICTABLE)),
93                        nid, K(node_page_state(nid, NR_MLOCK)));
94
95 #ifdef CONFIG_HIGHMEM
96         n += sprintf(buf + n,
97                        "Node %d HighTotal:      %8lu kB\n"
98                        "Node %d HighFree:       %8lu kB\n"
99                        "Node %d LowTotal:       %8lu kB\n"
100                        "Node %d LowFree:        %8lu kB\n",
101                        nid, K(i.totalhigh),
102                        nid, K(i.freehigh),
103                        nid, K(i.totalram - i.totalhigh),
104                        nid, K(i.freeram - i.freehigh));
105 #endif
106         n += sprintf(buf + n,
107                        "Node %d Dirty:          %8lu kB\n"
108                        "Node %d Writeback:      %8lu kB\n"
109                        "Node %d FilePages:      %8lu kB\n"
110                        "Node %d Mapped:         %8lu kB\n"
111                        "Node %d AnonPages:      %8lu kB\n"
112                        "Node %d Shmem:          %8lu kB\n"
113                        "Node %d KernelStack:    %8lu kB\n"
114                        "Node %d PageTables:     %8lu kB\n"
115                        "Node %d NFS_Unstable:   %8lu kB\n"
116                        "Node %d Bounce:         %8lu kB\n"
117                        "Node %d WritebackTmp:   %8lu kB\n"
118                        "Node %d Slab:           %8lu kB\n"
119                        "Node %d SReclaimable:   %8lu kB\n"
120                        "Node %d SUnreclaim:     %8lu kB\n"
121 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
122                        "Node %d AnonHugePages:  %8lu kB\n"
123 #endif
124                         ,
125                        nid, K(node_page_state(nid, NR_FILE_DIRTY)),
126                        nid, K(node_page_state(nid, NR_WRITEBACK)),
127                        nid, K(node_page_state(nid, NR_FILE_PAGES)),
128                        nid, K(node_page_state(nid, NR_FILE_MAPPED)),
129                        nid, K(node_page_state(nid, NR_ANON_PAGES)
130 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
131                         + node_page_state(nid, NR_ANON_TRANSPARENT_HUGEPAGES) *
132                         HPAGE_PMD_NR
133 #endif
134                        ),
135                        nid, K(node_page_state(nid, NR_SHMEM)),
136                        nid, node_page_state(nid, NR_KERNEL_STACK) *
137                                 THREAD_SIZE / 1024,
138                        nid, K(node_page_state(nid, NR_PAGETABLE)),
139                        nid, K(node_page_state(nid, NR_UNSTABLE_NFS)),
140                        nid, K(node_page_state(nid, NR_BOUNCE)),
141                        nid, K(node_page_state(nid, NR_WRITEBACK_TEMP)),
142                        nid, K(node_page_state(nid, NR_SLAB_RECLAIMABLE) +
143                                 node_page_state(nid, NR_SLAB_UNRECLAIMABLE)),
144                        nid, K(node_page_state(nid, NR_SLAB_RECLAIMABLE)),
145                        nid, K(node_page_state(nid, NR_SLAB_UNRECLAIMABLE))
146 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
147                         , nid,
148                         K(node_page_state(nid, NR_ANON_TRANSPARENT_HUGEPAGES) *
149                         HPAGE_PMD_NR)
150 #endif
151                        );
152         n += hugetlb_report_node_meminfo(nid, buf + n);
153         return n;
154 }
155
156 #undef K
157 static SYSDEV_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
158
159 static ssize_t node_read_numastat(struct sys_device * dev,
160                                 struct sysdev_attribute *attr, char * buf)
161 {
162         return sprintf(buf,
163                        "numa_hit %lu\n"
164                        "numa_miss %lu\n"
165                        "numa_foreign %lu\n"
166                        "interleave_hit %lu\n"
167                        "local_node %lu\n"
168                        "other_node %lu\n",
169                        node_page_state(dev->id, NUMA_HIT),
170                        node_page_state(dev->id, NUMA_MISS),
171                        node_page_state(dev->id, NUMA_FOREIGN),
172                        node_page_state(dev->id, NUMA_INTERLEAVE_HIT),
173                        node_page_state(dev->id, NUMA_LOCAL),
174                        node_page_state(dev->id, NUMA_OTHER));
175 }
176 static SYSDEV_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
177
178 static ssize_t node_read_vmstat(struct sys_device *dev,
179                                 struct sysdev_attribute *attr, char *buf)
180 {
181         int nid = dev->id;
182         return sprintf(buf,
183                 "nr_written %lu\n"
184                 "nr_dirtied %lu\n",
185                 node_page_state(nid, NR_WRITTEN),
186                 node_page_state(nid, NR_DIRTIED));
187 }
188 static SYSDEV_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL);
189
190 static ssize_t node_read_distance(struct sys_device * dev,
191                         struct sysdev_attribute *attr, char * buf)
192 {
193         int nid = dev->id;
194         int len = 0;
195         int i;
196
197         /*
198          * buf is currently PAGE_SIZE in length and each node needs 4 chars
199          * at the most (distance + space or newline).
200          */
201         BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
202
203         for_each_online_node(i)
204                 len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i));
205
206         len += sprintf(buf + len, "\n");
207         return len;
208 }
209 static SYSDEV_ATTR(distance, S_IRUGO, node_read_distance, NULL);
210
211 #ifdef CONFIG_HUGETLBFS
212 /*
213  * hugetlbfs per node attributes registration interface:
214  * When/if hugetlb[fs] subsystem initializes [sometime after this module],
215  * it will register its per node attributes for all online nodes with
216  * memory.  It will also call register_hugetlbfs_with_node(), below, to
217  * register its attribute registration functions with this node driver.
218  * Once these hooks have been initialized, the node driver will call into
219  * the hugetlb module to [un]register attributes for hot-plugged nodes.
220  */
221 static node_registration_func_t __hugetlb_register_node;
222 static node_registration_func_t __hugetlb_unregister_node;
223
224 static inline bool hugetlb_register_node(struct node *node)
225 {
226         if (__hugetlb_register_node &&
227                         node_state(node->sysdev.id, N_HIGH_MEMORY)) {
228                 __hugetlb_register_node(node);
229                 return true;
230         }
231         return false;
232 }
233
234 static inline void hugetlb_unregister_node(struct node *node)
235 {
236         if (__hugetlb_unregister_node)
237                 __hugetlb_unregister_node(node);
238 }
239
240 void register_hugetlbfs_with_node(node_registration_func_t doregister,
241                                   node_registration_func_t unregister)
242 {
243         __hugetlb_register_node   = doregister;
244         __hugetlb_unregister_node = unregister;
245 }
246 #else
247 static inline void hugetlb_register_node(struct node *node) {}
248
249 static inline void hugetlb_unregister_node(struct node *node) {}
250 #endif
251
252
253 /*
254  * register_node - Setup a sysfs device for a node.
255  * @num - Node number to use when creating the device.
256  *
257  * Initialize and register the node device.
258  */
259 int register_node(struct node *node, int num, struct node *parent)
260 {
261         int error;
262
263         node->sysdev.id = num;
264         node->sysdev.cls = &node_class;
265         error = sysdev_register(&node->sysdev);
266
267         if (!error){
268                 sysdev_create_file(&node->sysdev, &attr_cpumap);
269                 sysdev_create_file(&node->sysdev, &attr_cpulist);
270                 sysdev_create_file(&node->sysdev, &attr_meminfo);
271                 sysdev_create_file(&node->sysdev, &attr_numastat);
272                 sysdev_create_file(&node->sysdev, &attr_distance);
273                 sysdev_create_file(&node->sysdev, &attr_vmstat);
274
275                 scan_unevictable_register_node(node);
276
277                 hugetlb_register_node(node);
278
279                 compaction_register_node(node);
280         }
281         return error;
282 }
283
284 /**
285  * unregister_node - unregister a node device
286  * @node: node going away
287  *
288  * Unregisters a node device @node.  All the devices on the node must be
289  * unregistered before calling this function.
290  */
291 void unregister_node(struct node *node)
292 {
293         sysdev_remove_file(&node->sysdev, &attr_cpumap);
294         sysdev_remove_file(&node->sysdev, &attr_cpulist);
295         sysdev_remove_file(&node->sysdev, &attr_meminfo);
296         sysdev_remove_file(&node->sysdev, &attr_numastat);
297         sysdev_remove_file(&node->sysdev, &attr_distance);
298         sysdev_remove_file(&node->sysdev, &attr_vmstat);
299
300         scan_unevictable_unregister_node(node);
301         hugetlb_unregister_node(node);          /* no-op, if memoryless node */
302
303         sysdev_unregister(&node->sysdev);
304 }
305
306 struct node node_devices[MAX_NUMNODES];
307
308 /*
309  * register cpu under node
310  */
311 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
312 {
313         int ret;
314         struct sys_device *obj;
315
316         if (!node_online(nid))
317                 return 0;
318
319         obj = get_cpu_sysdev(cpu);
320         if (!obj)
321                 return 0;
322
323         ret = sysfs_create_link(&node_devices[nid].sysdev.kobj,
324                                 &obj->kobj,
325                                 kobject_name(&obj->kobj));
326         if (ret)
327                 return ret;
328
329         return sysfs_create_link(&obj->kobj,
330                                  &node_devices[nid].sysdev.kobj,
331                                  kobject_name(&node_devices[nid].sysdev.kobj));
332 }
333
334 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
335 {
336         struct sys_device *obj;
337
338         if (!node_online(nid))
339                 return 0;
340
341         obj = get_cpu_sysdev(cpu);
342         if (!obj)
343                 return 0;
344
345         sysfs_remove_link(&node_devices[nid].sysdev.kobj,
346                           kobject_name(&obj->kobj));
347         sysfs_remove_link(&obj->kobj,
348                           kobject_name(&node_devices[nid].sysdev.kobj));
349
350         return 0;
351 }
352
353 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
354 #define page_initialized(page)  (page->lru.next)
355
356 static int get_nid_for_pfn(unsigned long pfn)
357 {
358         struct page *page;
359
360         if (!pfn_valid_within(pfn))
361                 return -1;
362         page = pfn_to_page(pfn);
363         if (!page_initialized(page))
364                 return -1;
365         return pfn_to_nid(pfn);
366 }
367
368 /* register memory section under specified node if it spans that node */
369 int register_mem_sect_under_node(struct memory_block *mem_blk, int nid)
370 {
371         int ret;
372         unsigned long pfn, sect_start_pfn, sect_end_pfn;
373
374         if (!mem_blk)
375                 return -EFAULT;
376         if (!node_online(nid))
377                 return 0;
378
379         sect_start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
380         sect_end_pfn = section_nr_to_pfn(mem_blk->end_section_nr);
381         sect_end_pfn += PAGES_PER_SECTION - 1;
382         for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
383                 int page_nid;
384
385                 page_nid = get_nid_for_pfn(pfn);
386                 if (page_nid < 0)
387                         continue;
388                 if (page_nid != nid)
389                         continue;
390                 ret = sysfs_create_link_nowarn(&node_devices[nid].sysdev.kobj,
391                                         &mem_blk->sysdev.kobj,
392                                         kobject_name(&mem_blk->sysdev.kobj));
393                 if (ret)
394                         return ret;
395
396                 return sysfs_create_link_nowarn(&mem_blk->sysdev.kobj,
397                                 &node_devices[nid].sysdev.kobj,
398                                 kobject_name(&node_devices[nid].sysdev.kobj));
399         }
400         /* mem section does not span the specified node */
401         return 0;
402 }
403
404 /* unregister memory section under all nodes that it spans */
405 int unregister_mem_sect_under_nodes(struct memory_block *mem_blk,
406                                     unsigned long phys_index)
407 {
408         NODEMASK_ALLOC(nodemask_t, unlinked_nodes, GFP_KERNEL);
409         unsigned long pfn, sect_start_pfn, sect_end_pfn;
410
411         if (!mem_blk) {
412                 NODEMASK_FREE(unlinked_nodes);
413                 return -EFAULT;
414         }
415         if (!unlinked_nodes)
416                 return -ENOMEM;
417         nodes_clear(*unlinked_nodes);
418
419         sect_start_pfn = section_nr_to_pfn(phys_index);
420         sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
421         for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
422                 int nid;
423
424                 nid = get_nid_for_pfn(pfn);
425                 if (nid < 0)
426                         continue;
427                 if (!node_online(nid))
428                         continue;
429                 if (node_test_and_set(nid, *unlinked_nodes))
430                         continue;
431                 sysfs_remove_link(&node_devices[nid].sysdev.kobj,
432                          kobject_name(&mem_blk->sysdev.kobj));
433                 sysfs_remove_link(&mem_blk->sysdev.kobj,
434                          kobject_name(&node_devices[nid].sysdev.kobj));
435         }
436         NODEMASK_FREE(unlinked_nodes);
437         return 0;
438 }
439
440 static int link_mem_sections(int nid)
441 {
442         unsigned long start_pfn = NODE_DATA(nid)->node_start_pfn;
443         unsigned long end_pfn = start_pfn + NODE_DATA(nid)->node_spanned_pages;
444         unsigned long pfn;
445         struct memory_block *mem_blk = NULL;
446         int err = 0;
447
448         for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
449                 unsigned long section_nr = pfn_to_section_nr(pfn);
450                 struct mem_section *mem_sect;
451                 int ret;
452
453                 if (!present_section_nr(section_nr))
454                         continue;
455                 mem_sect = __nr_to_section(section_nr);
456                 mem_blk = find_memory_block_hinted(mem_sect, mem_blk);
457                 ret = register_mem_sect_under_node(mem_blk, nid);
458                 if (!err)
459                         err = ret;
460
461                 /* discard ref obtained in find_memory_block() */
462         }
463
464         if (mem_blk)
465                 kobject_put(&mem_blk->sysdev.kobj);
466         return err;
467 }
468
469 #ifdef CONFIG_HUGETLBFS
470 /*
471  * Handle per node hstate attribute [un]registration on transistions
472  * to/from memoryless state.
473  */
474 static void node_hugetlb_work(struct work_struct *work)
475 {
476         struct node *node = container_of(work, struct node, node_work);
477
478         /*
479          * We only get here when a node transitions to/from memoryless state.
480          * We can detect which transition occurred by examining whether the
481          * node has memory now.  hugetlb_register_node() already check this
482          * so we try to register the attributes.  If that fails, then the
483          * node has transitioned to memoryless, try to unregister the
484          * attributes.
485          */
486         if (!hugetlb_register_node(node))
487                 hugetlb_unregister_node(node);
488 }
489
490 static void init_node_hugetlb_work(int nid)
491 {
492         INIT_WORK(&node_devices[nid].node_work, node_hugetlb_work);
493 }
494
495 static int node_memory_callback(struct notifier_block *self,
496                                 unsigned long action, void *arg)
497 {
498         struct memory_notify *mnb = arg;
499         int nid = mnb->status_change_nid;
500
501         switch (action) {
502         case MEM_ONLINE:
503         case MEM_OFFLINE:
504                 /*
505                  * offload per node hstate [un]registration to a work thread
506                  * when transitioning to/from memoryless state.
507                  */
508                 if (nid != NUMA_NO_NODE)
509                         schedule_work(&node_devices[nid].node_work);
510                 break;
511
512         case MEM_GOING_ONLINE:
513         case MEM_GOING_OFFLINE:
514         case MEM_CANCEL_ONLINE:
515         case MEM_CANCEL_OFFLINE:
516         default:
517                 break;
518         }
519
520         return NOTIFY_OK;
521 }
522 #endif  /* CONFIG_HUGETLBFS */
523 #else   /* !CONFIG_MEMORY_HOTPLUG_SPARSE */
524
525 static int link_mem_sections(int nid) { return 0; }
526 #endif  /* CONFIG_MEMORY_HOTPLUG_SPARSE */
527
528 #if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
529     !defined(CONFIG_HUGETLBFS)
530 static inline int node_memory_callback(struct notifier_block *self,
531                                 unsigned long action, void *arg)
532 {
533         return NOTIFY_OK;
534 }
535
536 static void init_node_hugetlb_work(int nid) { }
537
538 #endif
539
540 int register_one_node(int nid)
541 {
542         int error = 0;
543         int cpu;
544
545         if (node_online(nid)) {
546                 int p_node = parent_node(nid);
547                 struct node *parent = NULL;
548
549                 if (p_node != nid)
550                         parent = &node_devices[p_node];
551
552                 error = register_node(&node_devices[nid], nid, parent);
553
554                 /* link cpu under this node */
555                 for_each_present_cpu(cpu) {
556                         if (cpu_to_node(cpu) == nid)
557                                 register_cpu_under_node(cpu, nid);
558                 }
559
560                 /* link memory sections under this node */
561                 error = link_mem_sections(nid);
562
563                 /* initialize work queue for memory hot plug */
564                 init_node_hugetlb_work(nid);
565         }
566
567         return error;
568
569 }
570
571 void unregister_one_node(int nid)
572 {
573         unregister_node(&node_devices[nid]);
574 }
575
576 /*
577  * node states attributes
578  */
579
580 static ssize_t print_nodes_state(enum node_states state, char *buf)
581 {
582         int n;
583
584         n = nodelist_scnprintf(buf, PAGE_SIZE, node_states[state]);
585         if (n > 0 && PAGE_SIZE > n + 1) {
586                 *(buf + n++) = '\n';
587                 *(buf + n++) = '\0';
588         }
589         return n;
590 }
591
592 struct node_attr {
593         struct sysdev_class_attribute attr;
594         enum node_states state;
595 };
596
597 static ssize_t show_node_state(struct sysdev_class *class,
598                                struct sysdev_class_attribute *attr, char *buf)
599 {
600         struct node_attr *na = container_of(attr, struct node_attr, attr);
601         return print_nodes_state(na->state, buf);
602 }
603
604 #define _NODE_ATTR(name, state) \
605         { _SYSDEV_CLASS_ATTR(name, 0444, show_node_state, NULL), state }
606
607 static struct node_attr node_state_attr[] = {
608         _NODE_ATTR(possible, N_POSSIBLE),
609         _NODE_ATTR(online, N_ONLINE),
610         _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
611         _NODE_ATTR(has_cpu, N_CPU),
612 #ifdef CONFIG_HIGHMEM
613         _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
614 #endif
615 };
616
617 static struct sysdev_class_attribute *node_state_attrs[] = {
618         &node_state_attr[0].attr,
619         &node_state_attr[1].attr,
620         &node_state_attr[2].attr,
621         &node_state_attr[3].attr,
622 #ifdef CONFIG_HIGHMEM
623         &node_state_attr[4].attr,
624 #endif
625         NULL
626 };
627
628 #define NODE_CALLBACK_PRI       2       /* lower than SLAB */
629 static int __init register_node_type(void)
630 {
631         int ret;
632
633         BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
634         BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
635
636         ret = sysdev_class_register(&node_class);
637         if (!ret) {
638                 hotplug_memory_notifier(node_memory_callback,
639                                         NODE_CALLBACK_PRI);
640         }
641
642         /*
643          * Note:  we're not going to unregister the node class if we fail
644          * to register the node state class attribute files.
645          */
646         return ret;
647 }
648 postcore_initcall(register_node_type);