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 *kobj, char **envp,
38 int num_envp, char *buffer, int buffer_size)
45 static struct kset_uevent_ops memory_uevent_ops = {
46 .name = memory_uevent_name,
47 .uevent = memory_uevent,
50 static struct notifier_block *memory_chain;
52 int register_memory_notifier(struct notifier_block *nb)
54 return notifier_chain_register(&memory_chain, nb);
57 void unregister_memory_notifier(struct notifier_block *nb)
59 notifier_chain_unregister(&memory_chain, nb);
63 * register_memory - Setup a sysfs device for a memory block
65 int register_memory(struct memory_block *memory, struct mem_section *section,
70 memory->sysdev.cls = &memory_sysdev_class;
71 memory->sysdev.id = __section_nr(section);
73 error = sysdev_register(&memory->sysdev);
76 error = sysfs_create_link(&root->sysdev.kobj,
78 kobject_name(&memory->sysdev.kobj));
84 unregister_memory(struct memory_block *memory, struct mem_section *section,
87 BUG_ON(memory->sysdev.cls != &memory_sysdev_class);
88 BUG_ON(memory->sysdev.id != __section_nr(section));
90 sysdev_unregister(&memory->sysdev);
92 sysfs_remove_link(&root->sysdev.kobj,
93 kobject_name(&memory->sysdev.kobj));
97 * use this as the physical section index that this memsection
101 static ssize_t show_mem_phys_index(struct sys_device *dev, char *buf)
103 struct memory_block *mem =
104 container_of(dev, struct memory_block, sysdev);
105 return sprintf(buf, "%08lx\n", mem->phys_index);
109 * online, offline, going offline, etc.
111 static ssize_t show_mem_state(struct sys_device *dev, char *buf)
113 struct memory_block *mem =
114 container_of(dev, struct memory_block, sysdev);
118 * We can probably put these states in a nice little array
119 * so that they're not open-coded
121 switch (mem->state) {
123 len = sprintf(buf, "online\n");
126 len = sprintf(buf, "offline\n");
128 case MEM_GOING_OFFLINE:
129 len = sprintf(buf, "going-offline\n");
132 len = sprintf(buf, "ERROR-UNKNOWN-%ld\n",
141 static inline int memory_notify(unsigned long val, void *v)
143 return notifier_call_chain(&memory_chain, val, v);
147 * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is
148 * OK to have direct references to sparsemem variables in here.
151 memory_block_action(struct memory_block *mem, unsigned long action)
154 unsigned long psection;
155 unsigned long start_pfn, start_paddr;
156 struct page *first_page;
158 int old_state = mem->state;
160 psection = mem->phys_index;
161 first_page = pfn_to_page(psection << PFN_SECTION_SHIFT);
164 * The probe routines leave the pages reserved, just
165 * as the bootmem code does. Make sure they're still
168 if (action == MEM_ONLINE) {
169 for (i = 0; i < PAGES_PER_SECTION; i++) {
170 if (PageReserved(first_page+i))
173 printk(KERN_WARNING "section number %ld page number %d "
174 "not reserved, was it already online? \n",
182 start_pfn = page_to_pfn(first_page);
183 ret = online_pages(start_pfn, PAGES_PER_SECTION);
186 mem->state = MEM_GOING_OFFLINE;
187 memory_notify(MEM_GOING_OFFLINE, NULL);
188 start_paddr = page_to_pfn(first_page) << PAGE_SHIFT;
189 ret = remove_memory(start_paddr,
190 PAGES_PER_SECTION << PAGE_SHIFT);
192 mem->state = old_state;
195 memory_notify(MEM_MAPPING_INVALID, NULL);
198 printk(KERN_WARNING "%s(%p, %ld) unknown action: %ld\n",
199 __FUNCTION__, mem, action, action);
204 * For now, only notify on successful memory operations
207 memory_notify(action, NULL);
212 static int memory_block_change_state(struct memory_block *mem,
213 unsigned long to_state, unsigned long from_state_req)
216 down(&mem->state_sem);
218 if (mem->state != from_state_req) {
223 ret = memory_block_action(mem, to_state);
225 mem->state = to_state;
233 store_mem_state(struct sys_device *dev, const char *buf, size_t count)
235 struct memory_block *mem;
236 unsigned int phys_section_nr;
239 mem = container_of(dev, struct memory_block, sysdev);
240 phys_section_nr = mem->phys_index;
242 if (!valid_section_nr(phys_section_nr))
245 if (!strncmp(buf, "online", min((int)count, 6)))
246 ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
247 else if(!strncmp(buf, "offline", min((int)count, 7)))
248 ret = memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE);
256 * phys_device is a bad name for this. What I really want
257 * is a way to differentiate between memory ranges that
258 * are part of physical devices that constitute
259 * a complete removable unit or fru.
260 * i.e. do these ranges belong to the same physical device,
261 * s.t. if I offline all of these sections I can then
262 * remove the physical device?
264 static ssize_t show_phys_device(struct sys_device *dev, char *buf)
266 struct memory_block *mem =
267 container_of(dev, struct memory_block, sysdev);
268 return sprintf(buf, "%d\n", mem->phys_device);
271 static SYSDEV_ATTR(phys_index, 0444, show_mem_phys_index, NULL);
272 static SYSDEV_ATTR(state, 0644, show_mem_state, store_mem_state);
273 static SYSDEV_ATTR(phys_device, 0444, show_phys_device, NULL);
275 #define mem_create_simple_file(mem, attr_name) \
276 sysdev_create_file(&mem->sysdev, &attr_##attr_name)
277 #define mem_remove_simple_file(mem, attr_name) \
278 sysdev_remove_file(&mem->sysdev, &attr_##attr_name)
281 * Block size attribute stuff
284 print_block_size(struct class *class, char *buf)
286 return sprintf(buf, "%lx\n", (unsigned long)PAGES_PER_SECTION * PAGE_SIZE);
289 static CLASS_ATTR(block_size_bytes, 0444, print_block_size, NULL);
291 static int block_size_init(void)
293 sysfs_create_file(&memory_sysdev_class.kset.kobj,
294 &class_attr_block_size_bytes.attr);
299 * Some architectures will have custom drivers to do this, and
300 * will not need to do it from userspace. The fake hot-add code
301 * as well as ppc64 will do all of their discovery in userspace
302 * and will require this interface.
304 #ifdef CONFIG_ARCH_MEMORY_PROBE
306 memory_probe_store(struct class *class, const char *buf, size_t count)
311 phys_addr = simple_strtoull(buf, NULL, 0);
313 ret = add_memory(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 sysfs_create_file(&memory_sysdev_class.kset.kobj,
325 &class_attr_probe.attr);
329 #define memory_probe_init(...) do {} while (0)
333 * Note that phys_device is optional. It is here to allow for
334 * differentiation between which *physical* devices each
335 * section belongs to...
338 static int add_memory_block(unsigned long node_id, struct mem_section *section,
339 unsigned long state, int phys_device)
341 struct memory_block *mem = kzalloc(sizeof(*mem), GFP_KERNEL);
347 mem->phys_index = __section_nr(section);
349 init_MUTEX(&mem->state_sem);
350 mem->phys_device = phys_device;
352 ret = register_memory(mem, section, NULL);
354 ret = mem_create_simple_file(mem, phys_index);
356 ret = mem_create_simple_file(mem, state);
358 ret = mem_create_simple_file(mem, phys_device);
364 * For now, we have a linear search to go find the appropriate
365 * memory_block corresponding to a particular phys_index. If
366 * this gets to be a real problem, we can always use a radix
367 * tree or something here.
369 * This could be made generic for all sysdev classes.
371 static struct memory_block *find_memory_block(struct mem_section *section)
373 struct kobject *kobj;
374 struct sys_device *sysdev;
375 struct memory_block *mem;
376 char name[sizeof(MEMORY_CLASS_NAME) + 9 + 1];
379 * This only works because we know that section == sysdev->id
380 * slightly redundant with sysdev_register()
382 sprintf(&name[0], "%s%d", MEMORY_CLASS_NAME, __section_nr(section));
384 kobj = kset_find_obj(&memory_sysdev_class.kset, name);
388 sysdev = container_of(kobj, struct sys_device, kobj);
389 mem = container_of(sysdev, struct memory_block, sysdev);
394 int remove_memory_block(unsigned long node_id, struct mem_section *section,
397 struct memory_block *mem;
399 mem = find_memory_block(section);
400 mem_remove_simple_file(mem, phys_index);
401 mem_remove_simple_file(mem, state);
402 mem_remove_simple_file(mem, phys_device);
403 unregister_memory(mem, section, NULL);
409 * need an interface for the VM to add new memory regions,
410 * but without onlining it.
412 int register_new_memory(struct mem_section *section)
414 return add_memory_block(0, section, MEM_OFFLINE, 0);
417 int unregister_memory_section(struct mem_section *section)
419 if (!valid_section(section))
422 return remove_memory_block(0, section, 0);
426 * Initialize the sysfs support for memory devices...
428 int __init memory_dev_init(void)
433 memory_sysdev_class.kset.uevent_ops = &memory_uevent_ops;
434 ret = sysdev_class_register(&memory_sysdev_class);
437 * Create entries for memory sections that were found
438 * during boot and have been initialized
440 for (i = 0; i < NR_MEM_SECTIONS; i++) {
441 if (!valid_section_nr(i))
443 add_memory_block(0, __nr_to_section(i), MEM_ONLINE, 0);