visorbus: switch from ioremap_cache to memremap
[linux/fpc-iii.git] / drivers / base / node.c
blob31df474d72f4a275ba0c87fb79a26c538fc661ca
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/notifier.h>
11 #include <linux/node.h>
12 #include <linux/hugetlb.h>
13 #include <linux/compaction.h>
14 #include <linux/cpumask.h>
15 #include <linux/topology.h>
16 #include <linux/nodemask.h>
17 #include <linux/cpu.h>
18 #include <linux/device.h>
19 #include <linux/swap.h>
20 #include <linux/slab.h>
22 static struct bus_type node_subsys = {
23 .name = "node",
24 .dev_name = "node",
28 static ssize_t node_read_cpumap(struct device *dev, bool list, char *buf)
30 struct node *node_dev = to_node(dev);
31 const struct cpumask *mask = cpumask_of_node(node_dev->dev.id);
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 return cpumap_print_to_pagebuf(list, buf, mask);
39 static inline ssize_t node_read_cpumask(struct device *dev,
40 struct device_attribute *attr, char *buf)
42 return node_read_cpumap(dev, false, buf);
44 static inline ssize_t node_read_cpulist(struct device *dev,
45 struct device_attribute *attr, char *buf)
47 return node_read_cpumap(dev, true, buf);
50 static DEVICE_ATTR(cpumap, S_IRUGO, node_read_cpumask, NULL);
51 static DEVICE_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL);
53 #define K(x) ((x) << (PAGE_SHIFT - 10))
54 static ssize_t node_read_meminfo(struct device *dev,
55 struct device_attribute *attr, char *buf)
57 int n;
58 int nid = dev->id;
59 struct sysinfo i;
61 si_meminfo_node(&i, nid);
62 n = sprintf(buf,
63 "Node %d MemTotal: %8lu kB\n"
64 "Node %d MemFree: %8lu kB\n"
65 "Node %d MemUsed: %8lu kB\n"
66 "Node %d Active: %8lu kB\n"
67 "Node %d Inactive: %8lu kB\n"
68 "Node %d Active(anon): %8lu kB\n"
69 "Node %d Inactive(anon): %8lu kB\n"
70 "Node %d Active(file): %8lu kB\n"
71 "Node %d Inactive(file): %8lu kB\n"
72 "Node %d Unevictable: %8lu kB\n"
73 "Node %d Mlocked: %8lu kB\n",
74 nid, K(i.totalram),
75 nid, K(i.freeram),
76 nid, K(i.totalram - i.freeram),
77 nid, K(node_page_state(nid, NR_ACTIVE_ANON) +
78 node_page_state(nid, NR_ACTIVE_FILE)),
79 nid, K(node_page_state(nid, NR_INACTIVE_ANON) +
80 node_page_state(nid, NR_INACTIVE_FILE)),
81 nid, K(node_page_state(nid, NR_ACTIVE_ANON)),
82 nid, K(node_page_state(nid, NR_INACTIVE_ANON)),
83 nid, K(node_page_state(nid, NR_ACTIVE_FILE)),
84 nid, K(node_page_state(nid, NR_INACTIVE_FILE)),
85 nid, K(node_page_state(nid, NR_UNEVICTABLE)),
86 nid, K(node_page_state(nid, NR_MLOCK)));
88 #ifdef CONFIG_HIGHMEM
89 n += sprintf(buf + n,
90 "Node %d HighTotal: %8lu kB\n"
91 "Node %d HighFree: %8lu kB\n"
92 "Node %d LowTotal: %8lu kB\n"
93 "Node %d LowFree: %8lu kB\n",
94 nid, K(i.totalhigh),
95 nid, K(i.freehigh),
96 nid, K(i.totalram - i.totalhigh),
97 nid, K(i.freeram - i.freehigh));
98 #endif
99 n += sprintf(buf + n,
100 "Node %d Dirty: %8lu kB\n"
101 "Node %d Writeback: %8lu kB\n"
102 "Node %d FilePages: %8lu kB\n"
103 "Node %d Mapped: %8lu kB\n"
104 "Node %d AnonPages: %8lu kB\n"
105 "Node %d Shmem: %8lu kB\n"
106 "Node %d KernelStack: %8lu kB\n"
107 "Node %d PageTables: %8lu kB\n"
108 "Node %d NFS_Unstable: %8lu kB\n"
109 "Node %d Bounce: %8lu kB\n"
110 "Node %d WritebackTmp: %8lu kB\n"
111 "Node %d Slab: %8lu kB\n"
112 "Node %d SReclaimable: %8lu kB\n"
113 "Node %d SUnreclaim: %8lu kB\n"
114 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
115 "Node %d AnonHugePages: %8lu kB\n"
116 #endif
118 nid, K(node_page_state(nid, NR_FILE_DIRTY)),
119 nid, K(node_page_state(nid, NR_WRITEBACK)),
120 nid, K(node_page_state(nid, NR_FILE_PAGES)),
121 nid, K(node_page_state(nid, NR_FILE_MAPPED)),
122 nid, K(node_page_state(nid, NR_ANON_PAGES)),
123 nid, K(i.sharedram),
124 nid, node_page_state(nid, NR_KERNEL_STACK) *
125 THREAD_SIZE / 1024,
126 nid, K(node_page_state(nid, NR_PAGETABLE)),
127 nid, K(node_page_state(nid, NR_UNSTABLE_NFS)),
128 nid, K(node_page_state(nid, NR_BOUNCE)),
129 nid, K(node_page_state(nid, NR_WRITEBACK_TEMP)),
130 nid, K(node_page_state(nid, NR_SLAB_RECLAIMABLE) +
131 node_page_state(nid, NR_SLAB_UNRECLAIMABLE)),
132 nid, K(node_page_state(nid, NR_SLAB_RECLAIMABLE)),
133 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
134 nid, K(node_page_state(nid, NR_SLAB_UNRECLAIMABLE))
135 , nid,
136 K(node_page_state(nid, NR_ANON_TRANSPARENT_HUGEPAGES) *
137 HPAGE_PMD_NR));
138 #else
139 nid, K(node_page_state(nid, NR_SLAB_UNRECLAIMABLE)));
140 #endif
141 n += hugetlb_report_node_meminfo(nid, buf + n);
142 return n;
145 #undef K
146 static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
148 static ssize_t node_read_numastat(struct device *dev,
149 struct device_attribute *attr, char *buf)
151 return sprintf(buf,
152 "numa_hit %lu\n"
153 "numa_miss %lu\n"
154 "numa_foreign %lu\n"
155 "interleave_hit %lu\n"
156 "local_node %lu\n"
157 "other_node %lu\n",
158 node_page_state(dev->id, NUMA_HIT),
159 node_page_state(dev->id, NUMA_MISS),
160 node_page_state(dev->id, NUMA_FOREIGN),
161 node_page_state(dev->id, NUMA_INTERLEAVE_HIT),
162 node_page_state(dev->id, NUMA_LOCAL),
163 node_page_state(dev->id, NUMA_OTHER));
165 static DEVICE_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
167 static ssize_t node_read_vmstat(struct device *dev,
168 struct device_attribute *attr, char *buf)
170 int nid = dev->id;
171 int i;
172 int n = 0;
174 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
175 n += sprintf(buf+n, "%s %lu\n", vmstat_text[i],
176 node_page_state(nid, i));
178 return n;
180 static DEVICE_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL);
182 static ssize_t node_read_distance(struct device *dev,
183 struct device_attribute *attr, char *buf)
185 int nid = dev->id;
186 int len = 0;
187 int i;
190 * buf is currently PAGE_SIZE in length and each node needs 4 chars
191 * at the most (distance + space or newline).
193 BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
195 for_each_online_node(i)
196 len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i));
198 len += sprintf(buf + len, "\n");
199 return len;
201 static DEVICE_ATTR(distance, S_IRUGO, node_read_distance, NULL);
203 static struct attribute *node_dev_attrs[] = {
204 &dev_attr_cpumap.attr,
205 &dev_attr_cpulist.attr,
206 &dev_attr_meminfo.attr,
207 &dev_attr_numastat.attr,
208 &dev_attr_distance.attr,
209 &dev_attr_vmstat.attr,
210 NULL
212 ATTRIBUTE_GROUPS(node_dev);
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 node->dev.groups = node_dev_groups;
288 error = device_register(&node->dev);
290 if (!error){
291 hugetlb_register_node(node);
293 compaction_register_node(node);
295 return error;
299 * unregister_node - unregister a node device
300 * @node: node going away
302 * Unregisters a node device @node. All the devices on the node must be
303 * unregistered before calling this function.
305 void unregister_node(struct node *node)
307 hugetlb_unregister_node(node); /* no-op, if memoryless node */
309 device_unregister(&node->dev);
312 struct node *node_devices[MAX_NUMNODES];
315 * register cpu under node
317 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
319 int ret;
320 struct device *obj;
322 if (!node_online(nid))
323 return 0;
325 obj = get_cpu_device(cpu);
326 if (!obj)
327 return 0;
329 ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
330 &obj->kobj,
331 kobject_name(&obj->kobj));
332 if (ret)
333 return ret;
335 return sysfs_create_link(&obj->kobj,
336 &node_devices[nid]->dev.kobj,
337 kobject_name(&node_devices[nid]->dev.kobj));
340 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
342 struct device *obj;
344 if (!node_online(nid))
345 return 0;
347 obj = get_cpu_device(cpu);
348 if (!obj)
349 return 0;
351 sysfs_remove_link(&node_devices[nid]->dev.kobj,
352 kobject_name(&obj->kobj));
353 sysfs_remove_link(&obj->kobj,
354 kobject_name(&node_devices[nid]->dev.kobj));
356 return 0;
359 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
360 #define page_initialized(page) (page->lru.next)
362 static int __init_refok get_nid_for_pfn(unsigned long pfn)
364 struct page *page;
366 if (!pfn_valid_within(pfn))
367 return -1;
368 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
369 if (system_state == SYSTEM_BOOTING)
370 return early_pfn_to_nid(pfn);
371 #endif
372 page = pfn_to_page(pfn);
373 if (!page_initialized(page))
374 return -1;
375 return pfn_to_nid(pfn);
378 /* register memory section under specified node if it spans that node */
379 int register_mem_sect_under_node(struct memory_block *mem_blk, int nid)
381 int ret;
382 unsigned long pfn, sect_start_pfn, sect_end_pfn;
384 if (!mem_blk)
385 return -EFAULT;
386 if (!node_online(nid))
387 return 0;
389 sect_start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
390 sect_end_pfn = section_nr_to_pfn(mem_blk->end_section_nr);
391 sect_end_pfn += PAGES_PER_SECTION - 1;
392 for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
393 int page_nid;
395 page_nid = get_nid_for_pfn(pfn);
396 if (page_nid < 0)
397 continue;
398 if (page_nid != nid)
399 continue;
400 ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
401 &mem_blk->dev.kobj,
402 kobject_name(&mem_blk->dev.kobj));
403 if (ret)
404 return ret;
406 return sysfs_create_link_nowarn(&mem_blk->dev.kobj,
407 &node_devices[nid]->dev.kobj,
408 kobject_name(&node_devices[nid]->dev.kobj));
410 /* mem section does not span the specified node */
411 return 0;
414 /* unregister memory section under all nodes that it spans */
415 int unregister_mem_sect_under_nodes(struct memory_block *mem_blk,
416 unsigned long phys_index)
418 NODEMASK_ALLOC(nodemask_t, unlinked_nodes, GFP_KERNEL);
419 unsigned long pfn, sect_start_pfn, sect_end_pfn;
421 if (!mem_blk) {
422 NODEMASK_FREE(unlinked_nodes);
423 return -EFAULT;
425 if (!unlinked_nodes)
426 return -ENOMEM;
427 nodes_clear(*unlinked_nodes);
429 sect_start_pfn = section_nr_to_pfn(phys_index);
430 sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
431 for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
432 int nid;
434 nid = get_nid_for_pfn(pfn);
435 if (nid < 0)
436 continue;
437 if (!node_online(nid))
438 continue;
439 if (node_test_and_set(nid, *unlinked_nodes))
440 continue;
441 sysfs_remove_link(&node_devices[nid]->dev.kobj,
442 kobject_name(&mem_blk->dev.kobj));
443 sysfs_remove_link(&mem_blk->dev.kobj,
444 kobject_name(&node_devices[nid]->dev.kobj));
446 NODEMASK_FREE(unlinked_nodes);
447 return 0;
450 static int link_mem_sections(int nid)
452 unsigned long start_pfn = NODE_DATA(nid)->node_start_pfn;
453 unsigned long end_pfn = start_pfn + NODE_DATA(nid)->node_spanned_pages;
454 unsigned long pfn;
455 struct memory_block *mem_blk = NULL;
456 int err = 0;
458 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
459 unsigned long section_nr = pfn_to_section_nr(pfn);
460 struct mem_section *mem_sect;
461 int ret;
463 if (!present_section_nr(section_nr))
464 continue;
465 mem_sect = __nr_to_section(section_nr);
467 /* same memblock ? */
468 if (mem_blk)
469 if ((section_nr >= mem_blk->start_section_nr) &&
470 (section_nr <= mem_blk->end_section_nr))
471 continue;
473 mem_blk = find_memory_block_hinted(mem_sect, mem_blk);
475 ret = register_mem_sect_under_node(mem_blk, nid);
476 if (!err)
477 err = ret;
479 /* discard ref obtained in find_memory_block() */
482 if (mem_blk)
483 kobject_put(&mem_blk->dev.kobj);
484 return err;
487 #ifdef CONFIG_HUGETLBFS
489 * Handle per node hstate attribute [un]registration on transistions
490 * to/from memoryless state.
492 static void node_hugetlb_work(struct work_struct *work)
494 struct node *node = container_of(work, struct node, node_work);
497 * We only get here when a node transitions to/from memoryless state.
498 * We can detect which transition occurred by examining whether the
499 * node has memory now. hugetlb_register_node() already check this
500 * so we try to register the attributes. If that fails, then the
501 * node has transitioned to memoryless, try to unregister the
502 * attributes.
504 if (!hugetlb_register_node(node))
505 hugetlb_unregister_node(node);
508 static void init_node_hugetlb_work(int nid)
510 INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
513 static int node_memory_callback(struct notifier_block *self,
514 unsigned long action, void *arg)
516 struct memory_notify *mnb = arg;
517 int nid = mnb->status_change_nid;
519 switch (action) {
520 case MEM_ONLINE:
521 case MEM_OFFLINE:
523 * offload per node hstate [un]registration to a work thread
524 * when transitioning to/from memoryless state.
526 if (nid != NUMA_NO_NODE)
527 schedule_work(&node_devices[nid]->node_work);
528 break;
530 case MEM_GOING_ONLINE:
531 case MEM_GOING_OFFLINE:
532 case MEM_CANCEL_ONLINE:
533 case MEM_CANCEL_OFFLINE:
534 default:
535 break;
538 return NOTIFY_OK;
540 #endif /* CONFIG_HUGETLBFS */
541 #else /* !CONFIG_MEMORY_HOTPLUG_SPARSE */
543 static int link_mem_sections(int nid) { return 0; }
544 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
546 #if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
547 !defined(CONFIG_HUGETLBFS)
548 static inline int node_memory_callback(struct notifier_block *self,
549 unsigned long action, void *arg)
551 return NOTIFY_OK;
554 static void init_node_hugetlb_work(int nid) { }
556 #endif
558 int register_one_node(int nid)
560 int error = 0;
561 int cpu;
563 if (node_online(nid)) {
564 int p_node = parent_node(nid);
565 struct node *parent = NULL;
567 if (p_node != nid)
568 parent = node_devices[p_node];
570 node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
571 if (!node_devices[nid])
572 return -ENOMEM;
574 error = register_node(node_devices[nid], nid, parent);
576 /* link cpu under this node */
577 for_each_present_cpu(cpu) {
578 if (cpu_to_node(cpu) == nid)
579 register_cpu_under_node(cpu, nid);
582 /* link memory sections under this node */
583 error = link_mem_sections(nid);
585 /* initialize work queue for memory hot plug */
586 init_node_hugetlb_work(nid);
589 return error;
593 void unregister_one_node(int nid)
595 if (!node_devices[nid])
596 return;
598 unregister_node(node_devices[nid]);
599 node_devices[nid] = NULL;
603 * node states attributes
606 static ssize_t print_nodes_state(enum node_states state, char *buf)
608 int n;
610 n = scnprintf(buf, PAGE_SIZE - 1, "%*pbl",
611 nodemask_pr_args(&node_states[state]));
612 buf[n++] = '\n';
613 buf[n] = '\0';
614 return n;
617 struct node_attr {
618 struct device_attribute attr;
619 enum node_states state;
622 static ssize_t show_node_state(struct device *dev,
623 struct device_attribute *attr, char *buf)
625 struct node_attr *na = container_of(attr, struct node_attr, attr);
626 return print_nodes_state(na->state, buf);
629 #define _NODE_ATTR(name, state) \
630 { __ATTR(name, 0444, show_node_state, NULL), state }
632 static struct node_attr node_state_attr[] = {
633 [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
634 [N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
635 [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
636 #ifdef CONFIG_HIGHMEM
637 [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
638 #endif
639 #ifdef CONFIG_MOVABLE_NODE
640 [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
641 #endif
642 [N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
645 static struct attribute *node_state_attrs[] = {
646 &node_state_attr[N_POSSIBLE].attr.attr,
647 &node_state_attr[N_ONLINE].attr.attr,
648 &node_state_attr[N_NORMAL_MEMORY].attr.attr,
649 #ifdef CONFIG_HIGHMEM
650 &node_state_attr[N_HIGH_MEMORY].attr.attr,
651 #endif
652 #ifdef CONFIG_MOVABLE_NODE
653 &node_state_attr[N_MEMORY].attr.attr,
654 #endif
655 &node_state_attr[N_CPU].attr.attr,
656 NULL
659 static struct attribute_group memory_root_attr_group = {
660 .attrs = node_state_attrs,
663 static const struct attribute_group *cpu_root_attr_groups[] = {
664 &memory_root_attr_group,
665 NULL,
668 #define NODE_CALLBACK_PRI 2 /* lower than SLAB */
669 static int __init register_node_type(void)
671 int ret;
673 BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
674 BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
676 ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
677 if (!ret) {
678 static struct notifier_block node_memory_callback_nb = {
679 .notifier_call = node_memory_callback,
680 .priority = NODE_CALLBACK_PRI,
682 register_hotmemory_notifier(&node_memory_callback_nb);
686 * Note: we're not going to unregister the node class if we fail
687 * to register the node state class attribute files.
689 return ret;
691 postcore_initcall(register_node_type);