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 <linux/mutex.h>
24 #include <asm/atomic.h>
25 #include <asm/uaccess.h>
27 #define MEMORY_CLASS_NAME "memory"
29 static struct sysdev_class memory_sysdev_class = {
30 .name = MEMORY_CLASS_NAME,
33 static const char *memory_uevent_name(struct kset *kset, struct kobject *kobj)
35 return MEMORY_CLASS_NAME;
38 static int memory_uevent(struct kset *kset, struct kobject *obj, struct kobj_uevent_env *env)
45 static struct kset_uevent_ops memory_uevent_ops = {
46 .name = memory_uevent_name,
47 .uevent = memory_uevent,
50 static BLOCKING_NOTIFIER_HEAD(memory_chain);
52 int register_memory_notifier(struct notifier_block *nb)
54 return blocking_notifier_chain_register(&memory_chain, nb);
56 EXPORT_SYMBOL(register_memory_notifier);
58 void unregister_memory_notifier(struct notifier_block *nb)
60 blocking_notifier_chain_unregister(&memory_chain, nb);
62 EXPORT_SYMBOL(unregister_memory_notifier);
65 * register_memory - Setup a sysfs device for a memory block
68 int register_memory(struct memory_block *memory, struct mem_section *section)
72 memory->sysdev.cls = &memory_sysdev_class;
73 memory->sysdev.id = __section_nr(section);
75 error = sysdev_register(&memory->sysdev);
80 unregister_memory(struct memory_block *memory, struct mem_section *section)
82 BUG_ON(memory->sysdev.cls != &memory_sysdev_class);
83 BUG_ON(memory->sysdev.id != __section_nr(section));
85 /* drop the ref. we got in remove_memory_block() */
86 kobject_put(&memory->sysdev.kobj);
87 sysdev_unregister(&memory->sysdev);
91 * use this as the physical section index that this memsection
95 static ssize_t show_mem_phys_index(struct sys_device *dev,
96 struct sysdev_attribute *attr, char *buf)
98 struct memory_block *mem =
99 container_of(dev, struct memory_block, sysdev);
100 return sprintf(buf, "%08lx\n", mem->phys_index);
104 * Show whether the section of memory is likely to be hot-removable
106 static ssize_t show_mem_removable(struct sys_device *dev, char *buf)
108 unsigned long start_pfn;
110 struct memory_block *mem =
111 container_of(dev, struct memory_block, sysdev);
113 start_pfn = section_nr_to_pfn(mem->phys_index);
114 ret = is_mem_section_removable(start_pfn, PAGES_PER_SECTION);
115 return sprintf(buf, "%d\n", ret);
119 * online, offline, going offline, etc.
121 static ssize_t show_mem_state(struct sys_device *dev,
122 struct sysdev_attribute *attr, char *buf)
124 struct memory_block *mem =
125 container_of(dev, struct memory_block, sysdev);
129 * We can probably put these states in a nice little array
130 * so that they're not open-coded
132 switch (mem->state) {
134 len = sprintf(buf, "online\n");
137 len = sprintf(buf, "offline\n");
139 case MEM_GOING_OFFLINE:
140 len = sprintf(buf, "going-offline\n");
143 len = sprintf(buf, "ERROR-UNKNOWN-%ld\n",
152 int memory_notify(unsigned long val, void *v)
154 return blocking_notifier_call_chain(&memory_chain, val, v);
158 * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is
159 * OK to have direct references to sparsemem variables in here.
162 memory_block_action(struct memory_block *mem, unsigned long action)
165 unsigned long psection;
166 unsigned long start_pfn, start_paddr;
167 struct page *first_page;
169 int old_state = mem->state;
171 psection = mem->phys_index;
172 first_page = pfn_to_page(psection << PFN_SECTION_SHIFT);
175 * The probe routines leave the pages reserved, just
176 * as the bootmem code does. Make sure they're still
179 if (action == MEM_ONLINE) {
180 for (i = 0; i < PAGES_PER_SECTION; i++) {
181 if (PageReserved(first_page+i))
184 printk(KERN_WARNING "section number %ld page number %d "
185 "not reserved, was it already online? \n",
193 start_pfn = page_to_pfn(first_page);
194 ret = online_pages(start_pfn, PAGES_PER_SECTION);
197 mem->state = MEM_GOING_OFFLINE;
198 start_paddr = page_to_pfn(first_page) << PAGE_SHIFT;
199 ret = remove_memory(start_paddr,
200 PAGES_PER_SECTION << PAGE_SHIFT);
202 mem->state = old_state;
207 WARN(1, KERN_WARNING "%s(%p, %ld) unknown action: %ld\n",
208 __func__, mem, action, action);
215 static int memory_block_change_state(struct memory_block *mem,
216 unsigned long to_state, unsigned long from_state_req)
219 mutex_lock(&mem->state_mutex);
221 if (mem->state != from_state_req) {
226 ret = memory_block_action(mem, to_state);
228 mem->state = to_state;
231 mutex_unlock(&mem->state_mutex);
236 store_mem_state(struct sys_device *dev,
237 struct sysdev_attribute *attr, const char *buf, size_t count)
239 struct memory_block *mem;
240 unsigned int phys_section_nr;
243 mem = container_of(dev, struct memory_block, sysdev);
244 phys_section_nr = mem->phys_index;
246 if (!present_section_nr(phys_section_nr))
249 if (!strncmp(buf, "online", min((int)count, 6)))
250 ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
251 else if(!strncmp(buf, "offline", min((int)count, 7)))
252 ret = memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE);
260 * phys_device is a bad name for this. What I really want
261 * is a way to differentiate between memory ranges that
262 * are part of physical devices that constitute
263 * a complete removable unit or fru.
264 * i.e. do these ranges belong to the same physical device,
265 * s.t. if I offline all of these sections I can then
266 * remove the physical device?
268 static ssize_t show_phys_device(struct sys_device *dev,
269 struct sysdev_attribute *attr, char *buf)
271 struct memory_block *mem =
272 container_of(dev, struct memory_block, sysdev);
273 return sprintf(buf, "%d\n", mem->phys_device);
276 static SYSDEV_ATTR(phys_index, 0444, show_mem_phys_index, NULL);
277 static SYSDEV_ATTR(state, 0644, show_mem_state, store_mem_state);
278 static SYSDEV_ATTR(phys_device, 0444, show_phys_device, NULL);
279 static SYSDEV_ATTR(removable, 0444, show_mem_removable, NULL);
281 #define mem_create_simple_file(mem, attr_name) \
282 sysdev_create_file(&mem->sysdev, &attr_##attr_name)
283 #define mem_remove_simple_file(mem, attr_name) \
284 sysdev_remove_file(&mem->sysdev, &attr_##attr_name)
287 * Block size attribute stuff
290 print_block_size(struct class *class, char *buf)
292 return sprintf(buf, "%lx\n", (unsigned long)PAGES_PER_SECTION * PAGE_SIZE);
295 static CLASS_ATTR(block_size_bytes, 0444, print_block_size, NULL);
297 static int block_size_init(void)
299 return sysfs_create_file(&memory_sysdev_class.kset.kobj,
300 &class_attr_block_size_bytes.attr);
304 * Some architectures will have custom drivers to do this, and
305 * will not need to do it from userspace. The fake hot-add code
306 * as well as ppc64 will do all of their discovery in userspace
307 * and will require this interface.
309 #ifdef CONFIG_ARCH_MEMORY_PROBE
311 memory_probe_store(struct class *class, const char *buf, size_t count)
317 phys_addr = simple_strtoull(buf, NULL, 0);
319 nid = memory_add_physaddr_to_nid(phys_addr);
320 ret = add_memory(nid, phys_addr, PAGES_PER_SECTION << PAGE_SHIFT);
327 static CLASS_ATTR(probe, 0700, NULL, memory_probe_store);
329 static int memory_probe_init(void)
331 return sysfs_create_file(&memory_sysdev_class.kset.kobj,
332 &class_attr_probe.attr);
335 static inline int memory_probe_init(void)
342 * Note that phys_device is optional. It is here to allow for
343 * differentiation between which *physical* devices each
344 * section belongs to...
347 static int add_memory_block(unsigned long node_id, struct mem_section *section,
348 unsigned long state, int phys_device)
350 struct memory_block *mem = kzalloc(sizeof(*mem), GFP_KERNEL);
356 mem->phys_index = __section_nr(section);
358 mutex_init(&mem->state_mutex);
359 mem->phys_device = phys_device;
361 ret = register_memory(mem, section);
363 ret = mem_create_simple_file(mem, phys_index);
365 ret = mem_create_simple_file(mem, state);
367 ret = mem_create_simple_file(mem, phys_device);
369 ret = mem_create_simple_file(mem, removable);
375 * For now, we have a linear search to go find the appropriate
376 * memory_block corresponding to a particular phys_index. If
377 * this gets to be a real problem, we can always use a radix
378 * tree or something here.
380 * This could be made generic for all sysdev classes.
382 static struct memory_block *find_memory_block(struct mem_section *section)
384 struct kobject *kobj;
385 struct sys_device *sysdev;
386 struct memory_block *mem;
387 char name[sizeof(MEMORY_CLASS_NAME) + 9 + 1];
390 * This only works because we know that section == sysdev->id
391 * slightly redundant with sysdev_register()
393 sprintf(&name[0], "%s%d", MEMORY_CLASS_NAME, __section_nr(section));
395 kobj = kset_find_obj(&memory_sysdev_class.kset, name);
399 sysdev = container_of(kobj, struct sys_device, kobj);
400 mem = container_of(sysdev, struct memory_block, sysdev);
405 int remove_memory_block(unsigned long node_id, struct mem_section *section,
408 struct memory_block *mem;
410 mem = find_memory_block(section);
411 mem_remove_simple_file(mem, phys_index);
412 mem_remove_simple_file(mem, state);
413 mem_remove_simple_file(mem, phys_device);
414 mem_remove_simple_file(mem, removable);
415 unregister_memory(mem, section);
421 * need an interface for the VM to add new memory regions,
422 * but without onlining it.
424 int register_new_memory(struct mem_section *section)
426 return add_memory_block(0, section, MEM_OFFLINE, 0);
429 int unregister_memory_section(struct mem_section *section)
431 if (!present_section(section))
434 return remove_memory_block(0, section, 0);
438 * Initialize the sysfs support for memory devices...
440 int __init memory_dev_init(void)
446 memory_sysdev_class.kset.uevent_ops = &memory_uevent_ops;
447 ret = sysdev_class_register(&memory_sysdev_class);
452 * Create entries for memory sections that were found
453 * during boot and have been initialized
455 for (i = 0; i < NR_MEM_SECTIONS; i++) {
456 if (!present_section_nr(i))
458 err = add_memory_block(0, __nr_to_section(i), MEM_ONLINE, 0);
463 err = memory_probe_init();
466 err = block_size_init();
471 printk(KERN_ERR "%s() failed: %d\n", __func__, ret);