mmc: mxs-mmc: move to use generic DMA helper
[linux-2.6.git] / drivers / base / node.c
blobfac124a7e1c5b0203b4953e6fd30ce9fbfe5c290
1 /*
2 * Basic Node interface support
3 */
5 #include <linux/module.h>
6 #include <linux/init.h>
7 #include <linux/mm.h>
8 #include <linux/memory.h>
9 #include <linux/vmstat.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>
21 static struct bus_type node_subsys = {
22 .name = "node",
23 .dev_name = "node",
27 static ssize_t node_read_cpumap(struct device *dev, int type, char *buf)
29 struct node *node_dev = to_node(dev);
30 const struct cpumask *mask = cpumask_of_node(node_dev->dev.id);
31 int len;
33 /* 2008/04/07: buf currently PAGE_SIZE, need 9 chars per 32 bits. */
34 BUILD_BUG_ON((NR_CPUS/32 * 9) > (PAGE_SIZE-1));
36 len = type?
37 cpulist_scnprintf(buf, PAGE_SIZE-2, mask) :
38 cpumask_scnprintf(buf, PAGE_SIZE-2, mask);
39 buf[len++] = '\n';
40 buf[len] = '\0';
41 return len;
44 static inline ssize_t node_read_cpumask(struct device *dev,
45 struct device_attribute *attr, char *buf)
47 return node_read_cpumap(dev, 0, buf);
49 static inline ssize_t node_read_cpulist(struct device *dev,
50 struct device_attribute *attr, char *buf)
52 return node_read_cpumap(dev, 1, buf);
55 static DEVICE_ATTR(cpumap, S_IRUGO, node_read_cpumask, NULL);
56 static DEVICE_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL);
58 #define K(x) ((x) << (PAGE_SHIFT - 10))
59 static ssize_t node_read_meminfo(struct device *dev,
60 struct device_attribute *attr, char *buf)
62 int n;
63 int nid = dev->id;
64 struct sysinfo i;
66 si_meminfo_node(&i, nid);
67 n = sprintf(buf,
68 "Node %d MemTotal: %8lu kB\n"
69 "Node %d MemFree: %8lu kB\n"
70 "Node %d MemUsed: %8lu kB\n"
71 "Node %d Active: %8lu kB\n"
72 "Node %d Inactive: %8lu kB\n"
73 "Node %d Active(anon): %8lu kB\n"
74 "Node %d Inactive(anon): %8lu kB\n"
75 "Node %d Active(file): %8lu kB\n"
76 "Node %d Inactive(file): %8lu kB\n"
77 "Node %d Unevictable: %8lu kB\n"
78 "Node %d Mlocked: %8lu kB\n",
79 nid, K(i.totalram),
80 nid, K(i.freeram),
81 nid, K(i.totalram - i.freeram),
82 nid, K(node_page_state(nid, NR_ACTIVE_ANON) +
83 node_page_state(nid, NR_ACTIVE_FILE)),
84 nid, K(node_page_state(nid, NR_INACTIVE_ANON) +
85 node_page_state(nid, NR_INACTIVE_FILE)),
86 nid, K(node_page_state(nid, NR_ACTIVE_ANON)),
87 nid, K(node_page_state(nid, NR_INACTIVE_ANON)),
88 nid, K(node_page_state(nid, NR_ACTIVE_FILE)),
89 nid, K(node_page_state(nid, NR_INACTIVE_FILE)),
90 nid, K(node_page_state(nid, NR_UNEVICTABLE)),
91 nid, K(node_page_state(nid, NR_MLOCK)));
93 #ifdef CONFIG_HIGHMEM
94 n += sprintf(buf + n,
95 "Node %d HighTotal: %8lu kB\n"
96 "Node %d HighFree: %8lu kB\n"
97 "Node %d LowTotal: %8lu kB\n"
98 "Node %d LowFree: %8lu kB\n",
99 nid, K(i.totalhigh),
100 nid, K(i.freehigh),
101 nid, K(i.totalram - i.totalhigh),
102 nid, K(i.freeram - i.freehigh));
103 #endif
104 n += sprintf(buf + n,
105 "Node %d Dirty: %8lu kB\n"
106 "Node %d Writeback: %8lu kB\n"
107 "Node %d FilePages: %8lu kB\n"
108 "Node %d Mapped: %8lu kB\n"
109 "Node %d AnonPages: %8lu kB\n"
110 "Node %d Shmem: %8lu kB\n"
111 "Node %d KernelStack: %8lu kB\n"
112 "Node %d PageTables: %8lu kB\n"
113 "Node %d NFS_Unstable: %8lu kB\n"
114 "Node %d Bounce: %8lu kB\n"
115 "Node %d WritebackTmp: %8lu kB\n"
116 "Node %d Slab: %8lu kB\n"
117 "Node %d SReclaimable: %8lu kB\n"
118 "Node %d SUnreclaim: %8lu kB\n"
119 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
120 "Node %d AnonHugePages: %8lu kB\n"
121 #endif
123 nid, K(node_page_state(nid, NR_FILE_DIRTY)),
124 nid, K(node_page_state(nid, NR_WRITEBACK)),
125 nid, K(node_page_state(nid, NR_FILE_PAGES)),
126 nid, K(node_page_state(nid, NR_FILE_MAPPED)),
127 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
128 nid, K(node_page_state(nid, NR_ANON_PAGES)
129 + node_page_state(nid, NR_ANON_TRANSPARENT_HUGEPAGES) *
130 HPAGE_PMD_NR),
131 #else
132 nid, K(node_page_state(nid, NR_ANON_PAGES)),
133 #endif
134 nid, K(node_page_state(nid, NR_SHMEM)),
135 nid, node_page_state(nid, NR_KERNEL_STACK) *
136 THREAD_SIZE / 1024,
137 nid, K(node_page_state(nid, NR_PAGETABLE)),
138 nid, K(node_page_state(nid, NR_UNSTABLE_NFS)),
139 nid, K(node_page_state(nid, NR_BOUNCE)),
140 nid, K(node_page_state(nid, NR_WRITEBACK_TEMP)),
141 nid, K(node_page_state(nid, NR_SLAB_RECLAIMABLE) +
142 node_page_state(nid, NR_SLAB_UNRECLAIMABLE)),
143 nid, K(node_page_state(nid, NR_SLAB_RECLAIMABLE)),
144 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
145 nid, K(node_page_state(nid, NR_SLAB_UNRECLAIMABLE))
146 , nid,
147 K(node_page_state(nid, NR_ANON_TRANSPARENT_HUGEPAGES) *
148 HPAGE_PMD_NR));
149 #else
150 nid, K(node_page_state(nid, NR_SLAB_UNRECLAIMABLE)));
151 #endif
152 n += hugetlb_report_node_meminfo(nid, buf + n);
153 return n;
156 #undef K
157 static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
159 static ssize_t node_read_numastat(struct device *dev,
160 struct device_attribute *attr, char *buf)
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));
176 static DEVICE_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
178 static ssize_t node_read_vmstat(struct device *dev,
179 struct device_attribute *attr, char *buf)
181 int nid = dev->id;
182 int i;
183 int n = 0;
185 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
186 n += sprintf(buf+n, "%s %lu\n", vmstat_text[i],
187 node_page_state(nid, i));
189 return n;
191 static DEVICE_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL);
193 static ssize_t node_read_distance(struct device *dev,
194 struct device_attribute *attr, char * buf)
196 int nid = dev->id;
197 int len = 0;
198 int i;
201 * buf is currently PAGE_SIZE in length and each node needs 4 chars
202 * at the most (distance + space or newline).
204 BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
206 for_each_online_node(i)
207 len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i));
209 len += sprintf(buf + len, "\n");
210 return len;
212 static DEVICE_ATTR(distance, S_IRUGO, node_read_distance, NULL);
214 #ifdef CONFIG_HUGETLBFS
216 * hugetlbfs per node attributes registration interface:
217 * When/if hugetlb[fs] subsystem initializes [sometime after this module],
218 * it will register its per node attributes for all online nodes with
219 * memory. It will also call register_hugetlbfs_with_node(), below, to
220 * register its attribute registration functions with this node driver.
221 * Once these hooks have been initialized, the node driver will call into
222 * the hugetlb module to [un]register attributes for hot-plugged nodes.
224 static node_registration_func_t __hugetlb_register_node;
225 static node_registration_func_t __hugetlb_unregister_node;
227 static inline bool hugetlb_register_node(struct node *node)
229 if (__hugetlb_register_node &&
230 node_state(node->dev.id, N_MEMORY)) {
231 __hugetlb_register_node(node);
232 return true;
234 return false;
237 static inline void hugetlb_unregister_node(struct node *node)
239 if (__hugetlb_unregister_node)
240 __hugetlb_unregister_node(node);
243 void register_hugetlbfs_with_node(node_registration_func_t doregister,
244 node_registration_func_t unregister)
246 __hugetlb_register_node = doregister;
247 __hugetlb_unregister_node = unregister;
249 #else
250 static inline void hugetlb_register_node(struct node *node) {}
252 static inline void hugetlb_unregister_node(struct node *node) {}
253 #endif
255 static void node_device_release(struct device *dev)
257 struct node *node = to_node(dev);
259 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
261 * We schedule the work only when a memory section is
262 * onlined/offlined on this node. When we come here,
263 * all the memory on this node has been offlined,
264 * so we won't enqueue new work to this work.
266 * The work is using node->node_work, so we should
267 * flush work before freeing the memory.
269 flush_work(&node->node_work);
270 #endif
271 kfree(node);
275 * register_node - Setup a sysfs device for a node.
276 * @num - Node number to use when creating the device.
278 * Initialize and register the node device.
280 static int register_node(struct node *node, int num, struct node *parent)
282 int error;
284 node->dev.id = num;
285 node->dev.bus = &node_subsys;
286 node->dev.release = node_device_release;
287 error = device_register(&node->dev);
289 if (!error){
290 device_create_file(&node->dev, &dev_attr_cpumap);
291 device_create_file(&node->dev, &dev_attr_cpulist);
292 device_create_file(&node->dev, &dev_attr_meminfo);
293 device_create_file(&node->dev, &dev_attr_numastat);
294 device_create_file(&node->dev, &dev_attr_distance);
295 device_create_file(&node->dev, &dev_attr_vmstat);
297 scan_unevictable_register_node(node);
299 hugetlb_register_node(node);
301 compaction_register_node(node);
303 return error;
307 * unregister_node - unregister a node device
308 * @node: node going away
310 * Unregisters a node device @node. All the devices on the node must be
311 * unregistered before calling this function.
313 void unregister_node(struct node *node)
315 device_remove_file(&node->dev, &dev_attr_cpumap);
316 device_remove_file(&node->dev, &dev_attr_cpulist);
317 device_remove_file(&node->dev, &dev_attr_meminfo);
318 device_remove_file(&node->dev, &dev_attr_numastat);
319 device_remove_file(&node->dev, &dev_attr_distance);
320 device_remove_file(&node->dev, &dev_attr_vmstat);
322 scan_unevictable_unregister_node(node);
323 hugetlb_unregister_node(node); /* no-op, if memoryless node */
325 device_unregister(&node->dev);
328 struct node *node_devices[MAX_NUMNODES];
331 * register cpu under node
333 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
335 int ret;
336 struct device *obj;
338 if (!node_online(nid))
339 return 0;
341 obj = get_cpu_device(cpu);
342 if (!obj)
343 return 0;
345 ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
346 &obj->kobj,
347 kobject_name(&obj->kobj));
348 if (ret)
349 return ret;
351 return sysfs_create_link(&obj->kobj,
352 &node_devices[nid]->dev.kobj,
353 kobject_name(&node_devices[nid]->dev.kobj));
356 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
358 struct device *obj;
360 if (!node_online(nid))
361 return 0;
363 obj = get_cpu_device(cpu);
364 if (!obj)
365 return 0;
367 sysfs_remove_link(&node_devices[nid]->dev.kobj,
368 kobject_name(&obj->kobj));
369 sysfs_remove_link(&obj->kobj,
370 kobject_name(&node_devices[nid]->dev.kobj));
372 return 0;
375 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
376 #define page_initialized(page) (page->lru.next)
378 static int get_nid_for_pfn(unsigned long pfn)
380 struct page *page;
382 if (!pfn_valid_within(pfn))
383 return -1;
384 page = pfn_to_page(pfn);
385 if (!page_initialized(page))
386 return -1;
387 return pfn_to_nid(pfn);
390 /* register memory section under specified node if it spans that node */
391 int register_mem_sect_under_node(struct memory_block *mem_blk, int nid)
393 int ret;
394 unsigned long pfn, sect_start_pfn, sect_end_pfn;
396 if (!mem_blk)
397 return -EFAULT;
398 if (!node_online(nid))
399 return 0;
401 sect_start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
402 sect_end_pfn = section_nr_to_pfn(mem_blk->end_section_nr);
403 sect_end_pfn += PAGES_PER_SECTION - 1;
404 for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
405 int page_nid;
407 page_nid = get_nid_for_pfn(pfn);
408 if (page_nid < 0)
409 continue;
410 if (page_nid != nid)
411 continue;
412 ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
413 &mem_blk->dev.kobj,
414 kobject_name(&mem_blk->dev.kobj));
415 if (ret)
416 return ret;
418 return sysfs_create_link_nowarn(&mem_blk->dev.kobj,
419 &node_devices[nid]->dev.kobj,
420 kobject_name(&node_devices[nid]->dev.kobj));
422 /* mem section does not span the specified node */
423 return 0;
426 /* unregister memory section under all nodes that it spans */
427 int unregister_mem_sect_under_nodes(struct memory_block *mem_blk,
428 unsigned long phys_index)
430 NODEMASK_ALLOC(nodemask_t, unlinked_nodes, GFP_KERNEL);
431 unsigned long pfn, sect_start_pfn, sect_end_pfn;
433 if (!mem_blk) {
434 NODEMASK_FREE(unlinked_nodes);
435 return -EFAULT;
437 if (!unlinked_nodes)
438 return -ENOMEM;
439 nodes_clear(*unlinked_nodes);
441 sect_start_pfn = section_nr_to_pfn(phys_index);
442 sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
443 for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
444 int nid;
446 nid = get_nid_for_pfn(pfn);
447 if (nid < 0)
448 continue;
449 if (!node_online(nid))
450 continue;
451 if (node_test_and_set(nid, *unlinked_nodes))
452 continue;
453 sysfs_remove_link(&node_devices[nid]->dev.kobj,
454 kobject_name(&mem_blk->dev.kobj));
455 sysfs_remove_link(&mem_blk->dev.kobj,
456 kobject_name(&node_devices[nid]->dev.kobj));
458 NODEMASK_FREE(unlinked_nodes);
459 return 0;
462 static int link_mem_sections(int nid)
464 unsigned long start_pfn = NODE_DATA(nid)->node_start_pfn;
465 unsigned long end_pfn = start_pfn + NODE_DATA(nid)->node_spanned_pages;
466 unsigned long pfn;
467 struct memory_block *mem_blk = NULL;
468 int err = 0;
470 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
471 unsigned long section_nr = pfn_to_section_nr(pfn);
472 struct mem_section *mem_sect;
473 int ret;
475 if (!present_section_nr(section_nr))
476 continue;
477 mem_sect = __nr_to_section(section_nr);
479 /* same memblock ? */
480 if (mem_blk)
481 if ((section_nr >= mem_blk->start_section_nr) &&
482 (section_nr <= mem_blk->end_section_nr))
483 continue;
485 mem_blk = find_memory_block_hinted(mem_sect, mem_blk);
487 ret = register_mem_sect_under_node(mem_blk, nid);
488 if (!err)
489 err = ret;
491 /* discard ref obtained in find_memory_block() */
494 if (mem_blk)
495 kobject_put(&mem_blk->dev.kobj);
496 return err;
499 #ifdef CONFIG_HUGETLBFS
501 * Handle per node hstate attribute [un]registration on transistions
502 * to/from memoryless state.
504 static void node_hugetlb_work(struct work_struct *work)
506 struct node *node = container_of(work, struct node, node_work);
509 * We only get here when a node transitions to/from memoryless state.
510 * We can detect which transition occurred by examining whether the
511 * node has memory now. hugetlb_register_node() already check this
512 * so we try to register the attributes. If that fails, then the
513 * node has transitioned to memoryless, try to unregister the
514 * attributes.
516 if (!hugetlb_register_node(node))
517 hugetlb_unregister_node(node);
520 static void init_node_hugetlb_work(int nid)
522 INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
525 static int node_memory_callback(struct notifier_block *self,
526 unsigned long action, void *arg)
528 struct memory_notify *mnb = arg;
529 int nid = mnb->status_change_nid;
531 switch (action) {
532 case MEM_ONLINE:
533 case MEM_OFFLINE:
535 * offload per node hstate [un]registration to a work thread
536 * when transitioning to/from memoryless state.
538 if (nid != NUMA_NO_NODE)
539 schedule_work(&node_devices[nid]->node_work);
540 break;
542 case MEM_GOING_ONLINE:
543 case MEM_GOING_OFFLINE:
544 case MEM_CANCEL_ONLINE:
545 case MEM_CANCEL_OFFLINE:
546 default:
547 break;
550 return NOTIFY_OK;
552 #endif /* CONFIG_HUGETLBFS */
553 #else /* !CONFIG_MEMORY_HOTPLUG_SPARSE */
555 static int link_mem_sections(int nid) { return 0; }
556 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
558 #if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
559 !defined(CONFIG_HUGETLBFS)
560 static inline int node_memory_callback(struct notifier_block *self,
561 unsigned long action, void *arg)
563 return NOTIFY_OK;
566 static void init_node_hugetlb_work(int nid) { }
568 #endif
570 int register_one_node(int nid)
572 int error = 0;
573 int cpu;
575 if (node_online(nid)) {
576 int p_node = parent_node(nid);
577 struct node *parent = NULL;
579 if (p_node != nid)
580 parent = node_devices[p_node];
582 node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
583 if (!node_devices[nid])
584 return -ENOMEM;
586 error = register_node(node_devices[nid], nid, parent);
588 /* link cpu under this node */
589 for_each_present_cpu(cpu) {
590 if (cpu_to_node(cpu) == nid)
591 register_cpu_under_node(cpu, nid);
594 /* link memory sections under this node */
595 error = link_mem_sections(nid);
597 /* initialize work queue for memory hot plug */
598 init_node_hugetlb_work(nid);
601 return error;
605 void unregister_one_node(int nid)
607 unregister_node(node_devices[nid]);
608 node_devices[nid] = NULL;
612 * node states attributes
615 static ssize_t print_nodes_state(enum node_states state, char *buf)
617 int n;
619 n = nodelist_scnprintf(buf, PAGE_SIZE-2, node_states[state]);
620 buf[n++] = '\n';
621 buf[n] = '\0';
622 return n;
625 struct node_attr {
626 struct device_attribute attr;
627 enum node_states state;
630 static ssize_t show_node_state(struct device *dev,
631 struct device_attribute *attr, char *buf)
633 struct node_attr *na = container_of(attr, struct node_attr, attr);
634 return print_nodes_state(na->state, buf);
637 #define _NODE_ATTR(name, state) \
638 { __ATTR(name, 0444, show_node_state, NULL), state }
640 static struct node_attr node_state_attr[] = {
641 [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
642 [N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
643 [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
644 #ifdef CONFIG_HIGHMEM
645 [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
646 #endif
647 #ifdef CONFIG_MOVABLE_NODE
648 [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
649 #endif
650 [N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
653 static struct attribute *node_state_attrs[] = {
654 &node_state_attr[N_POSSIBLE].attr.attr,
655 &node_state_attr[N_ONLINE].attr.attr,
656 &node_state_attr[N_NORMAL_MEMORY].attr.attr,
657 #ifdef CONFIG_HIGHMEM
658 &node_state_attr[N_HIGH_MEMORY].attr.attr,
659 #endif
660 #ifdef CONFIG_MOVABLE_NODE
661 &node_state_attr[N_MEMORY].attr.attr,
662 #endif
663 &node_state_attr[N_CPU].attr.attr,
664 NULL
667 static struct attribute_group memory_root_attr_group = {
668 .attrs = node_state_attrs,
671 static const struct attribute_group *cpu_root_attr_groups[] = {
672 &memory_root_attr_group,
673 NULL,
676 #define NODE_CALLBACK_PRI 2 /* lower than SLAB */
677 static int __init register_node_type(void)
679 int ret;
681 BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
682 BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
684 ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
685 if (!ret) {
686 hotplug_memory_notifier(node_memory_callback,
687 NODE_CALLBACK_PRI);
691 * Note: we're not going to unregister the node class if we fail
692 * to register the node state class attribute files.
694 return ret;
696 postcore_initcall(register_node_type);