2 * drivers/base/memory.c - basic Memory class support
4 * Written by Matt Tolentino <matthew.e.tolentino@intel.com>
5 * Dave Hansen <haveblue@us.ibm.com>
7 * This file provides the necessary infrastructure to represent
8 * a SPARSEMEM-memory-model system's physical memory in /sysfs.
9 * All arch-independent code that assumes MEMORY_HOTPLUG requires
10 * SPARSEMEM should be contained here, or in mm/memory_hotplug.c.
13 #include <linux/sysdev.h>
14 #include <linux/module.h>
15 #include <linux/init.h>
16 #include <linux/topology.h>
17 #include <linux/capability.h>
18 #include <linux/device.h>
19 #include <linux/memory.h>
20 #include <linux/kobject.h>
21 #include <linux/memory_hotplug.h>
23 #include <asm/atomic.h>
24 #include <asm/uaccess.h>
26 #define MEMORY_CLASS_NAME "memory"
28 static struct sysdev_class memory_sysdev_class = {
29 set_kset_name(MEMORY_CLASS_NAME),
32 static const char *memory_uevent_name(struct kset *kset, struct kobject *kobj)
34 return MEMORY_CLASS_NAME;
37 static int memory_uevent(struct kset *kset, struct kobject *obj, struct kobj_uevent_env *env)
44 static struct kset_uevent_ops memory_uevent_ops = {
45 .name = memory_uevent_name,
46 .uevent = memory_uevent,
49 static BLOCKING_NOTIFIER_HEAD(memory_chain);
51 int register_memory_notifier(struct notifier_block *nb)
53 return blocking_notifier_chain_register(&memory_chain, nb);
56 void unregister_memory_notifier(struct notifier_block *nb)
58 blocking_notifier_chain_unregister(&memory_chain, nb);
62 * register_memory - Setup a sysfs device for a memory block
64 int register_memory(struct memory_block *memory, struct mem_section *section,
69 memory->sysdev.cls = &memory_sysdev_class;
70 memory->sysdev.id = __section_nr(section);
72 error = sysdev_register(&memory->sysdev);
75 error = sysfs_create_link(&root->sysdev.kobj,
77 kobject_name(&memory->sysdev.kobj));
83 unregister_memory(struct memory_block *memory, struct mem_section *section,
86 BUG_ON(memory->sysdev.cls != &memory_sysdev_class);
87 BUG_ON(memory->sysdev.id != __section_nr(section));
89 sysdev_unregister(&memory->sysdev);
91 sysfs_remove_link(&root->sysdev.kobj,
92 kobject_name(&memory->sysdev.kobj));
96 * use this as the physical section index that this memsection
100 static ssize_t show_mem_phys_index(struct sys_device *dev, char *buf)
102 struct memory_block *mem =
103 container_of(dev, struct memory_block, sysdev);
104 return sprintf(buf, "%08lx\n", mem->phys_index);
108 * online, offline, going offline, etc.
110 static ssize_t show_mem_state(struct sys_device *dev, char *buf)
112 struct memory_block *mem =
113 container_of(dev, struct memory_block, sysdev);
117 * We can probably put these states in a nice little array
118 * so that they're not open-coded
120 switch (mem->state) {
122 len = sprintf(buf, "online\n");
125 len = sprintf(buf, "offline\n");
127 case MEM_GOING_OFFLINE:
128 len = sprintf(buf, "going-offline\n");
131 len = sprintf(buf, "ERROR-UNKNOWN-%ld\n",
140 static inline int memory_notify(unsigned long val, void *v)
142 return blocking_notifier_call_chain(&memory_chain, val, v);
146 * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is
147 * OK to have direct references to sparsemem variables in here.
150 memory_block_action(struct memory_block *mem, unsigned long action)
153 unsigned long psection;
154 unsigned long start_pfn, start_paddr;
155 struct page *first_page;
157 int old_state = mem->state;
159 psection = mem->phys_index;
160 first_page = pfn_to_page(psection << PFN_SECTION_SHIFT);
163 * The probe routines leave the pages reserved, just
164 * as the bootmem code does. Make sure they're still
167 if (action == MEM_ONLINE) {
168 for (i = 0; i < PAGES_PER_SECTION; i++) {
169 if (PageReserved(first_page+i))
172 printk(KERN_WARNING "section number %ld page number %d "
173 "not reserved, was it already online? \n",
181 start_pfn = page_to_pfn(first_page);
182 ret = online_pages(start_pfn, PAGES_PER_SECTION);
185 mem->state = MEM_GOING_OFFLINE;
186 memory_notify(MEM_GOING_OFFLINE, NULL);
187 start_paddr = page_to_pfn(first_page) << PAGE_SHIFT;
188 ret = remove_memory(start_paddr,
189 PAGES_PER_SECTION << PAGE_SHIFT);
191 mem->state = old_state;
194 memory_notify(MEM_MAPPING_INVALID, NULL);
197 printk(KERN_WARNING "%s(%p, %ld) unknown action: %ld\n",
198 __FUNCTION__, mem, action, action);
203 * For now, only notify on successful memory operations
206 memory_notify(action, NULL);
211 static int memory_block_change_state(struct memory_block *mem,
212 unsigned long to_state, unsigned long from_state_req)
215 down(&mem->state_sem);
217 if (mem->state != from_state_req) {
222 ret = memory_block_action(mem, to_state);
224 mem->state = to_state;
232 store_mem_state(struct sys_device *dev, const char *buf, size_t count)
234 struct memory_block *mem;
235 unsigned int phys_section_nr;
238 mem = container_of(dev, struct memory_block, sysdev);
239 phys_section_nr = mem->phys_index;
241 if (!valid_section_nr(phys_section_nr))
244 if (!strncmp(buf, "online", min((int)count, 6)))
245 ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
246 else if(!strncmp(buf, "offline", min((int)count, 7)))
247 ret = memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE);
255 * phys_device is a bad name for this. What I really want
256 * is a way to differentiate between memory ranges that
257 * are part of physical devices that constitute
258 * a complete removable unit or fru.
259 * i.e. do these ranges belong to the same physical device,
260 * s.t. if I offline all of these sections I can then
261 * remove the physical device?
263 static ssize_t show_phys_device(struct sys_device *dev, char *buf)
265 struct memory_block *mem =
266 container_of(dev, struct memory_block, sysdev);
267 return sprintf(buf, "%d\n", mem->phys_device);
270 static SYSDEV_ATTR(phys_index, 0444, show_mem_phys_index, NULL);
271 static SYSDEV_ATTR(state, 0644, show_mem_state, store_mem_state);
272 static SYSDEV_ATTR(phys_device, 0444, show_phys_device, NULL);
274 #define mem_create_simple_file(mem, attr_name) \
275 sysdev_create_file(&mem->sysdev, &attr_##attr_name)
276 #define mem_remove_simple_file(mem, attr_name) \
277 sysdev_remove_file(&mem->sysdev, &attr_##attr_name)
280 * Block size attribute stuff
283 print_block_size(struct class *class, char *buf)
285 return sprintf(buf, "%lx\n", (unsigned long)PAGES_PER_SECTION * PAGE_SIZE);
288 static CLASS_ATTR(block_size_bytes, 0444, print_block_size, NULL);
290 static int block_size_init(void)
292 return sysfs_create_file(&memory_sysdev_class.kset.kobj,
293 &class_attr_block_size_bytes.attr);
297 * Some architectures will have custom drivers to do this, and
298 * will not need to do it from userspace. The fake hot-add code
299 * as well as ppc64 will do all of their discovery in userspace
300 * and will require this interface.
302 #ifdef CONFIG_ARCH_MEMORY_PROBE
304 memory_probe_store(struct class *class, const char *buf, size_t count)
310 phys_addr = simple_strtoull(buf, NULL, 0);
312 nid = memory_add_physaddr_to_nid(phys_addr);
313 ret = add_memory(nid, phys_addr, PAGES_PER_SECTION << PAGE_SHIFT);
320 static CLASS_ATTR(probe, 0700, NULL, memory_probe_store);
322 static int memory_probe_init(void)
324 return sysfs_create_file(&memory_sysdev_class.kset.kobj,
325 &class_attr_probe.attr);
328 static inline int memory_probe_init(void)
335 * Note that phys_device is optional. It is here to allow for
336 * differentiation between which *physical* devices each
337 * section belongs to...
340 static int add_memory_block(unsigned long node_id, struct mem_section *section,
341 unsigned long state, int phys_device)
343 struct memory_block *mem = kzalloc(sizeof(*mem), GFP_KERNEL);
349 mem->phys_index = __section_nr(section);
351 init_MUTEX(&mem->state_sem);
352 mem->phys_device = phys_device;
354 ret = register_memory(mem, section, NULL);
356 ret = mem_create_simple_file(mem, phys_index);
358 ret = mem_create_simple_file(mem, state);
360 ret = mem_create_simple_file(mem, phys_device);
366 * For now, we have a linear search to go find the appropriate
367 * memory_block corresponding to a particular phys_index. If
368 * this gets to be a real problem, we can always use a radix
369 * tree or something here.
371 * This could be made generic for all sysdev classes.
373 static struct memory_block *find_memory_block(struct mem_section *section)
375 struct kobject *kobj;
376 struct sys_device *sysdev;
377 struct memory_block *mem;
378 char name[sizeof(MEMORY_CLASS_NAME) + 9 + 1];
381 * This only works because we know that section == sysdev->id
382 * slightly redundant with sysdev_register()
384 sprintf(&name[0], "%s%d", MEMORY_CLASS_NAME, __section_nr(section));
386 kobj = kset_find_obj(&memory_sysdev_class.kset, name);
390 sysdev = container_of(kobj, struct sys_device, kobj);
391 mem = container_of(sysdev, struct memory_block, sysdev);
396 int remove_memory_block(unsigned long node_id, struct mem_section *section,
399 struct memory_block *mem;
401 mem = find_memory_block(section);
402 mem_remove_simple_file(mem, phys_index);
403 mem_remove_simple_file(mem, state);
404 mem_remove_simple_file(mem, phys_device);
405 unregister_memory(mem, section, NULL);
411 * need an interface for the VM to add new memory regions,
412 * but without onlining it.
414 int register_new_memory(struct mem_section *section)
416 return add_memory_block(0, section, MEM_OFFLINE, 0);
419 int unregister_memory_section(struct mem_section *section)
421 if (!valid_section(section))
424 return remove_memory_block(0, section, 0);
428 * Initialize the sysfs support for memory devices...
430 int __init memory_dev_init(void)
436 memory_sysdev_class.kset.uevent_ops = &memory_uevent_ops;
437 ret = sysdev_class_register(&memory_sysdev_class);
442 * Create entries for memory sections that were found
443 * during boot and have been initialized
445 for (i = 0; i < NR_MEM_SECTIONS; i++) {
446 if (!valid_section_nr(i))
448 err = add_memory_block(0, __nr_to_section(i), MEM_ONLINE, 0);
453 err = memory_probe_init();
456 err = block_size_init();
461 printk(KERN_ERR "%s() failed: %d\n", __FUNCTION__, ret);