2 * Basic Node interface support
5 #include <linux/module.h>
6 #include <linux/init.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
= {
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
)
61 si_meminfo_node(&i
, nid
);
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",
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
)));
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",
96 nid
, K(i
.totalram
- i
.totalhigh
),
97 nid
, K(i
.freeram
- i
.freehigh
));
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 "Node %d ShmemHugePages: %8lu kB\n"
117 "Node %d ShmemPmdMapped: %8lu kB\n"
120 nid
, K(node_page_state(nid
, NR_FILE_DIRTY
)),
121 nid
, K(node_page_state(nid
, NR_WRITEBACK
)),
122 nid
, K(node_page_state(nid
, NR_FILE_PAGES
)),
123 nid
, K(node_page_state(nid
, NR_FILE_MAPPED
)),
124 nid
, K(node_page_state(nid
, NR_ANON_PAGES
)),
126 nid
, node_page_state(nid
, NR_KERNEL_STACK
) *
128 nid
, K(node_page_state(nid
, NR_PAGETABLE
)),
129 nid
, K(node_page_state(nid
, NR_UNSTABLE_NFS
)),
130 nid
, K(node_page_state(nid
, NR_BOUNCE
)),
131 nid
, K(node_page_state(nid
, NR_WRITEBACK_TEMP
)),
132 nid
, K(node_page_state(nid
, NR_SLAB_RECLAIMABLE
) +
133 node_page_state(nid
, NR_SLAB_UNRECLAIMABLE
)),
134 nid
, K(node_page_state(nid
, NR_SLAB_RECLAIMABLE
)),
135 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
136 nid
, K(node_page_state(nid
, NR_SLAB_UNRECLAIMABLE
)),
137 nid
, K(node_page_state(nid
, NR_ANON_THPS
) *
139 nid
, K(node_page_state(nid
, NR_SHMEM_THPS
) *
141 nid
, K(node_page_state(nid
, NR_SHMEM_PMDMAPPED
) *
144 nid
, K(node_page_state(nid
, NR_SLAB_UNRECLAIMABLE
)));
146 n
+= hugetlb_report_node_meminfo(nid
, buf
+ n
);
151 static DEVICE_ATTR(meminfo
, S_IRUGO
, node_read_meminfo
, NULL
);
153 static ssize_t
node_read_numastat(struct device
*dev
,
154 struct device_attribute
*attr
, char *buf
)
160 "interleave_hit %lu\n"
163 node_page_state(dev
->id
, NUMA_HIT
),
164 node_page_state(dev
->id
, NUMA_MISS
),
165 node_page_state(dev
->id
, NUMA_FOREIGN
),
166 node_page_state(dev
->id
, NUMA_INTERLEAVE_HIT
),
167 node_page_state(dev
->id
, NUMA_LOCAL
),
168 node_page_state(dev
->id
, NUMA_OTHER
));
170 static DEVICE_ATTR(numastat
, S_IRUGO
, node_read_numastat
, NULL
);
172 static ssize_t
node_read_vmstat(struct device
*dev
,
173 struct device_attribute
*attr
, char *buf
)
179 for (i
= 0; i
< NR_VM_ZONE_STAT_ITEMS
; i
++)
180 n
+= sprintf(buf
+n
, "%s %lu\n", vmstat_text
[i
],
181 node_page_state(nid
, i
));
185 static DEVICE_ATTR(vmstat
, S_IRUGO
, node_read_vmstat
, NULL
);
187 static ssize_t
node_read_distance(struct device
*dev
,
188 struct device_attribute
*attr
, char *buf
)
195 * buf is currently PAGE_SIZE in length and each node needs 4 chars
196 * at the most (distance + space or newline).
198 BUILD_BUG_ON(MAX_NUMNODES
* 4 > PAGE_SIZE
);
200 for_each_online_node(i
)
201 len
+= sprintf(buf
+ len
, "%s%d", i
? " " : "", node_distance(nid
, i
));
203 len
+= sprintf(buf
+ len
, "\n");
206 static DEVICE_ATTR(distance
, S_IRUGO
, node_read_distance
, NULL
);
208 static struct attribute
*node_dev_attrs
[] = {
209 &dev_attr_cpumap
.attr
,
210 &dev_attr_cpulist
.attr
,
211 &dev_attr_meminfo
.attr
,
212 &dev_attr_numastat
.attr
,
213 &dev_attr_distance
.attr
,
214 &dev_attr_vmstat
.attr
,
217 ATTRIBUTE_GROUPS(node_dev
);
219 #ifdef CONFIG_HUGETLBFS
221 * hugetlbfs per node attributes registration interface:
222 * When/if hugetlb[fs] subsystem initializes [sometime after this module],
223 * it will register its per node attributes for all online nodes with
224 * memory. It will also call register_hugetlbfs_with_node(), below, to
225 * register its attribute registration functions with this node driver.
226 * Once these hooks have been initialized, the node driver will call into
227 * the hugetlb module to [un]register attributes for hot-plugged nodes.
229 static node_registration_func_t __hugetlb_register_node
;
230 static node_registration_func_t __hugetlb_unregister_node
;
232 static inline bool hugetlb_register_node(struct node
*node
)
234 if (__hugetlb_register_node
&&
235 node_state(node
->dev
.id
, N_MEMORY
)) {
236 __hugetlb_register_node(node
);
242 static inline void hugetlb_unregister_node(struct node
*node
)
244 if (__hugetlb_unregister_node
)
245 __hugetlb_unregister_node(node
);
248 void register_hugetlbfs_with_node(node_registration_func_t doregister
,
249 node_registration_func_t unregister
)
251 __hugetlb_register_node
= doregister
;
252 __hugetlb_unregister_node
= unregister
;
255 static inline void hugetlb_register_node(struct node
*node
) {}
257 static inline void hugetlb_unregister_node(struct node
*node
) {}
260 static void node_device_release(struct device
*dev
)
262 struct node
*node
= to_node(dev
);
264 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
266 * We schedule the work only when a memory section is
267 * onlined/offlined on this node. When we come here,
268 * all the memory on this node has been offlined,
269 * so we won't enqueue new work to this work.
271 * The work is using node->node_work, so we should
272 * flush work before freeing the memory.
274 flush_work(&node
->node_work
);
280 * register_node - Setup a sysfs device for a node.
281 * @num - Node number to use when creating the device.
283 * Initialize and register the node device.
285 static int register_node(struct node
*node
, int num
, struct node
*parent
)
290 node
->dev
.bus
= &node_subsys
;
291 node
->dev
.release
= node_device_release
;
292 node
->dev
.groups
= node_dev_groups
;
293 error
= device_register(&node
->dev
);
296 hugetlb_register_node(node
);
298 compaction_register_node(node
);
304 * unregister_node - unregister a node device
305 * @node: node going away
307 * Unregisters a node device @node. All the devices on the node must be
308 * unregistered before calling this function.
310 void unregister_node(struct node
*node
)
312 hugetlb_unregister_node(node
); /* no-op, if memoryless node */
314 device_unregister(&node
->dev
);
317 struct node
*node_devices
[MAX_NUMNODES
];
320 * register cpu under node
322 int register_cpu_under_node(unsigned int cpu
, unsigned int nid
)
327 if (!node_online(nid
))
330 obj
= get_cpu_device(cpu
);
334 ret
= sysfs_create_link(&node_devices
[nid
]->dev
.kobj
,
336 kobject_name(&obj
->kobj
));
340 return sysfs_create_link(&obj
->kobj
,
341 &node_devices
[nid
]->dev
.kobj
,
342 kobject_name(&node_devices
[nid
]->dev
.kobj
));
345 int unregister_cpu_under_node(unsigned int cpu
, unsigned int nid
)
349 if (!node_online(nid
))
352 obj
= get_cpu_device(cpu
);
356 sysfs_remove_link(&node_devices
[nid
]->dev
.kobj
,
357 kobject_name(&obj
->kobj
));
358 sysfs_remove_link(&obj
->kobj
,
359 kobject_name(&node_devices
[nid
]->dev
.kobj
));
364 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
365 #define page_initialized(page) (page->lru.next)
367 static int __init_refok
get_nid_for_pfn(unsigned long pfn
)
371 if (!pfn_valid_within(pfn
))
373 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
374 if (system_state
== SYSTEM_BOOTING
)
375 return early_pfn_to_nid(pfn
);
377 page
= pfn_to_page(pfn
);
378 if (!page_initialized(page
))
380 return pfn_to_nid(pfn
);
383 /* register memory section under specified node if it spans that node */
384 int register_mem_sect_under_node(struct memory_block
*mem_blk
, int nid
)
387 unsigned long pfn
, sect_start_pfn
, sect_end_pfn
;
391 if (!node_online(nid
))
394 sect_start_pfn
= section_nr_to_pfn(mem_blk
->start_section_nr
);
395 sect_end_pfn
= section_nr_to_pfn(mem_blk
->end_section_nr
);
396 sect_end_pfn
+= PAGES_PER_SECTION
- 1;
397 for (pfn
= sect_start_pfn
; pfn
<= sect_end_pfn
; pfn
++) {
401 * memory block could have several absent sections from start.
402 * skip pfn range from absent section
404 if (!pfn_present(pfn
)) {
405 pfn
= round_down(pfn
+ PAGES_PER_SECTION
,
406 PAGES_PER_SECTION
) - 1;
410 page_nid
= get_nid_for_pfn(pfn
);
415 ret
= sysfs_create_link_nowarn(&node_devices
[nid
]->dev
.kobj
,
417 kobject_name(&mem_blk
->dev
.kobj
));
421 return sysfs_create_link_nowarn(&mem_blk
->dev
.kobj
,
422 &node_devices
[nid
]->dev
.kobj
,
423 kobject_name(&node_devices
[nid
]->dev
.kobj
));
425 /* mem section does not span the specified node */
429 /* unregister memory section under all nodes that it spans */
430 int unregister_mem_sect_under_nodes(struct memory_block
*mem_blk
,
431 unsigned long phys_index
)
433 NODEMASK_ALLOC(nodemask_t
, unlinked_nodes
, GFP_KERNEL
);
434 unsigned long pfn
, sect_start_pfn
, sect_end_pfn
;
437 NODEMASK_FREE(unlinked_nodes
);
442 nodes_clear(*unlinked_nodes
);
444 sect_start_pfn
= section_nr_to_pfn(phys_index
);
445 sect_end_pfn
= sect_start_pfn
+ PAGES_PER_SECTION
- 1;
446 for (pfn
= sect_start_pfn
; pfn
<= sect_end_pfn
; pfn
++) {
449 nid
= get_nid_for_pfn(pfn
);
452 if (!node_online(nid
))
454 if (node_test_and_set(nid
, *unlinked_nodes
))
456 sysfs_remove_link(&node_devices
[nid
]->dev
.kobj
,
457 kobject_name(&mem_blk
->dev
.kobj
));
458 sysfs_remove_link(&mem_blk
->dev
.kobj
,
459 kobject_name(&node_devices
[nid
]->dev
.kobj
));
461 NODEMASK_FREE(unlinked_nodes
);
465 static int link_mem_sections(int nid
)
467 unsigned long start_pfn
= NODE_DATA(nid
)->node_start_pfn
;
468 unsigned long end_pfn
= start_pfn
+ NODE_DATA(nid
)->node_spanned_pages
;
470 struct memory_block
*mem_blk
= NULL
;
473 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
474 unsigned long section_nr
= pfn_to_section_nr(pfn
);
475 struct mem_section
*mem_sect
;
478 if (!present_section_nr(section_nr
))
480 mem_sect
= __nr_to_section(section_nr
);
482 /* same memblock ? */
484 if ((section_nr
>= mem_blk
->start_section_nr
) &&
485 (section_nr
<= mem_blk
->end_section_nr
))
488 mem_blk
= find_memory_block_hinted(mem_sect
, mem_blk
);
490 ret
= register_mem_sect_under_node(mem_blk
, nid
);
494 /* discard ref obtained in find_memory_block() */
498 kobject_put(&mem_blk
->dev
.kobj
);
502 #ifdef CONFIG_HUGETLBFS
504 * Handle per node hstate attribute [un]registration on transistions
505 * to/from memoryless state.
507 static void node_hugetlb_work(struct work_struct
*work
)
509 struct node
*node
= container_of(work
, struct node
, node_work
);
512 * We only get here when a node transitions to/from memoryless state.
513 * We can detect which transition occurred by examining whether the
514 * node has memory now. hugetlb_register_node() already check this
515 * so we try to register the attributes. If that fails, then the
516 * node has transitioned to memoryless, try to unregister the
519 if (!hugetlb_register_node(node
))
520 hugetlb_unregister_node(node
);
523 static void init_node_hugetlb_work(int nid
)
525 INIT_WORK(&node_devices
[nid
]->node_work
, node_hugetlb_work
);
528 static int node_memory_callback(struct notifier_block
*self
,
529 unsigned long action
, void *arg
)
531 struct memory_notify
*mnb
= arg
;
532 int nid
= mnb
->status_change_nid
;
538 * offload per node hstate [un]registration to a work thread
539 * when transitioning to/from memoryless state.
541 if (nid
!= NUMA_NO_NODE
)
542 schedule_work(&node_devices
[nid
]->node_work
);
545 case MEM_GOING_ONLINE
:
546 case MEM_GOING_OFFLINE
:
547 case MEM_CANCEL_ONLINE
:
548 case MEM_CANCEL_OFFLINE
:
555 #endif /* CONFIG_HUGETLBFS */
556 #else /* !CONFIG_MEMORY_HOTPLUG_SPARSE */
558 static int link_mem_sections(int nid
) { return 0; }
559 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
561 #if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
562 !defined(CONFIG_HUGETLBFS)
563 static inline int node_memory_callback(struct notifier_block
*self
,
564 unsigned long action
, void *arg
)
569 static void init_node_hugetlb_work(int nid
) { }
573 int register_one_node(int nid
)
578 if (node_online(nid
)) {
579 int p_node
= parent_node(nid
);
580 struct node
*parent
= NULL
;
583 parent
= node_devices
[p_node
];
585 node_devices
[nid
] = kzalloc(sizeof(struct node
), GFP_KERNEL
);
586 if (!node_devices
[nid
])
589 error
= register_node(node_devices
[nid
], nid
, parent
);
591 /* link cpu under this node */
592 for_each_present_cpu(cpu
) {
593 if (cpu_to_node(cpu
) == nid
)
594 register_cpu_under_node(cpu
, nid
);
597 /* link memory sections under this node */
598 error
= link_mem_sections(nid
);
600 /* initialize work queue for memory hot plug */
601 init_node_hugetlb_work(nid
);
608 void unregister_one_node(int nid
)
610 if (!node_devices
[nid
])
613 unregister_node(node_devices
[nid
]);
614 node_devices
[nid
] = NULL
;
618 * node states attributes
621 static ssize_t
print_nodes_state(enum node_states state
, char *buf
)
625 n
= scnprintf(buf
, PAGE_SIZE
- 1, "%*pbl",
626 nodemask_pr_args(&node_states
[state
]));
633 struct device_attribute attr
;
634 enum node_states state
;
637 static ssize_t
show_node_state(struct device
*dev
,
638 struct device_attribute
*attr
, char *buf
)
640 struct node_attr
*na
= container_of(attr
, struct node_attr
, attr
);
641 return print_nodes_state(na
->state
, buf
);
644 #define _NODE_ATTR(name, state) \
645 { __ATTR(name, 0444, show_node_state, NULL), state }
647 static struct node_attr node_state_attr
[] = {
648 [N_POSSIBLE
] = _NODE_ATTR(possible
, N_POSSIBLE
),
649 [N_ONLINE
] = _NODE_ATTR(online
, N_ONLINE
),
650 [N_NORMAL_MEMORY
] = _NODE_ATTR(has_normal_memory
, N_NORMAL_MEMORY
),
651 #ifdef CONFIG_HIGHMEM
652 [N_HIGH_MEMORY
] = _NODE_ATTR(has_high_memory
, N_HIGH_MEMORY
),
654 #ifdef CONFIG_MOVABLE_NODE
655 [N_MEMORY
] = _NODE_ATTR(has_memory
, N_MEMORY
),
657 [N_CPU
] = _NODE_ATTR(has_cpu
, N_CPU
),
660 static struct attribute
*node_state_attrs
[] = {
661 &node_state_attr
[N_POSSIBLE
].attr
.attr
,
662 &node_state_attr
[N_ONLINE
].attr
.attr
,
663 &node_state_attr
[N_NORMAL_MEMORY
].attr
.attr
,
664 #ifdef CONFIG_HIGHMEM
665 &node_state_attr
[N_HIGH_MEMORY
].attr
.attr
,
667 #ifdef CONFIG_MOVABLE_NODE
668 &node_state_attr
[N_MEMORY
].attr
.attr
,
670 &node_state_attr
[N_CPU
].attr
.attr
,
674 static struct attribute_group memory_root_attr_group
= {
675 .attrs
= node_state_attrs
,
678 static const struct attribute_group
*cpu_root_attr_groups
[] = {
679 &memory_root_attr_group
,
683 #define NODE_CALLBACK_PRI 2 /* lower than SLAB */
684 static int __init
register_node_type(void)
688 BUILD_BUG_ON(ARRAY_SIZE(node_state_attr
) != NR_NODE_STATES
);
689 BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs
)-1 != NR_NODE_STATES
);
691 ret
= subsys_system_register(&node_subsys
, cpu_root_attr_groups
);
693 static struct notifier_block node_memory_callback_nb
= {
694 .notifier_call
= node_memory_callback
,
695 .priority
= NODE_CALLBACK_PRI
,
697 register_hotmemory_notifier(&node_memory_callback_nb
);
701 * Note: we're not going to unregister the node class if we fail
702 * to register the node state class attribute files.
706 postcore_initcall(register_node_type
);