Commit | Line | Data |
---|---|---|
3947be19 DH |
1 | /* |
2 | * drivers/base/memory.c - basic Memory class support | |
3 | * | |
4 | * Written by Matt Tolentino <matthew.e.tolentino@intel.com> | |
5 | * Dave Hansen <haveblue@us.ibm.com> | |
6 | * | |
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. | |
11 | */ | |
12 | ||
13 | #include <linux/sysdev.h> | |
14 | #include <linux/module.h> | |
15 | #include <linux/init.h> | |
3947be19 | 16 | #include <linux/topology.h> |
c59ede7b | 17 | #include <linux/capability.h> |
3947be19 DH |
18 | #include <linux/device.h> |
19 | #include <linux/memory.h> | |
20 | #include <linux/kobject.h> | |
21 | #include <linux/memory_hotplug.h> | |
22 | #include <linux/mm.h> | |
da19cbcf | 23 | #include <linux/mutex.h> |
3947be19 DH |
24 | #include <asm/atomic.h> |
25 | #include <asm/uaccess.h> | |
26 | ||
27 | #define MEMORY_CLASS_NAME "memory" | |
28 | ||
29 | static struct sysdev_class memory_sysdev_class = { | |
af5ca3f4 | 30 | .name = MEMORY_CLASS_NAME, |
3947be19 | 31 | }; |
3947be19 | 32 | |
312c004d | 33 | static const char *memory_uevent_name(struct kset *kset, struct kobject *kobj) |
3947be19 DH |
34 | { |
35 | return MEMORY_CLASS_NAME; | |
36 | } | |
37 | ||
9ec0fd4e | 38 | static int memory_uevent(struct kset *kset, struct kobject *obj, struct kobj_uevent_env *env) |
3947be19 DH |
39 | { |
40 | int retval = 0; | |
41 | ||
42 | return retval; | |
43 | } | |
44 | ||
312c004d KS |
45 | static struct kset_uevent_ops memory_uevent_ops = { |
46 | .name = memory_uevent_name, | |
47 | .uevent = memory_uevent, | |
3947be19 DH |
48 | }; |
49 | ||
e041c683 | 50 | static BLOCKING_NOTIFIER_HEAD(memory_chain); |
3947be19 | 51 | |
98a38ebd | 52 | int register_memory_notifier(struct notifier_block *nb) |
3947be19 | 53 | { |
e041c683 | 54 | return blocking_notifier_chain_register(&memory_chain, nb); |
3947be19 DH |
55 | } |
56 | ||
98a38ebd | 57 | void unregister_memory_notifier(struct notifier_block *nb) |
3947be19 | 58 | { |
e041c683 | 59 | blocking_notifier_chain_unregister(&memory_chain, nb); |
3947be19 DH |
60 | } |
61 | ||
62 | /* | |
63 | * register_memory - Setup a sysfs device for a memory block | |
64 | */ | |
00a41db5 BP |
65 | static |
66 | int register_memory(struct memory_block *memory, struct mem_section *section) | |
3947be19 DH |
67 | { |
68 | int error; | |
69 | ||
70 | memory->sysdev.cls = &memory_sysdev_class; | |
71 | memory->sysdev.id = __section_nr(section); | |
72 | ||
73 | error = sysdev_register(&memory->sysdev); | |
3947be19 DH |
74 | return error; |
75 | } | |
76 | ||
77 | static void | |
00a41db5 | 78 | unregister_memory(struct memory_block *memory, struct mem_section *section) |
3947be19 DH |
79 | { |
80 | BUG_ON(memory->sysdev.cls != &memory_sysdev_class); | |
81 | BUG_ON(memory->sysdev.id != __section_nr(section)); | |
82 | ||
00a41db5 BP |
83 | /* drop the ref. we got in remove_memory_block() */ |
84 | kobject_put(&memory->sysdev.kobj); | |
3947be19 | 85 | sysdev_unregister(&memory->sysdev); |
3947be19 DH |
86 | } |
87 | ||
88 | /* | |
89 | * use this as the physical section index that this memsection | |
90 | * uses. | |
91 | */ | |
92 | ||
93 | static ssize_t show_mem_phys_index(struct sys_device *dev, char *buf) | |
94 | { | |
95 | struct memory_block *mem = | |
96 | container_of(dev, struct memory_block, sysdev); | |
97 | return sprintf(buf, "%08lx\n", mem->phys_index); | |
98 | } | |
99 | ||
100 | /* | |
101 | * online, offline, going offline, etc. | |
102 | */ | |
103 | static ssize_t show_mem_state(struct sys_device *dev, char *buf) | |
104 | { | |
105 | struct memory_block *mem = | |
106 | container_of(dev, struct memory_block, sysdev); | |
107 | ssize_t len = 0; | |
108 | ||
109 | /* | |
110 | * We can probably put these states in a nice little array | |
111 | * so that they're not open-coded | |
112 | */ | |
113 | switch (mem->state) { | |
114 | case MEM_ONLINE: | |
115 | len = sprintf(buf, "online\n"); | |
116 | break; | |
117 | case MEM_OFFLINE: | |
118 | len = sprintf(buf, "offline\n"); | |
119 | break; | |
120 | case MEM_GOING_OFFLINE: | |
121 | len = sprintf(buf, "going-offline\n"); | |
122 | break; | |
123 | default: | |
124 | len = sprintf(buf, "ERROR-UNKNOWN-%ld\n", | |
125 | mem->state); | |
126 | WARN_ON(1); | |
127 | break; | |
128 | } | |
129 | ||
130 | return len; | |
131 | } | |
132 | ||
7b78d335 | 133 | int memory_notify(unsigned long val, void *v) |
3947be19 | 134 | { |
e041c683 | 135 | return blocking_notifier_call_chain(&memory_chain, val, v); |
3947be19 DH |
136 | } |
137 | ||
138 | /* | |
139 | * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is | |
140 | * OK to have direct references to sparsemem variables in here. | |
141 | */ | |
142 | static int | |
143 | memory_block_action(struct memory_block *mem, unsigned long action) | |
144 | { | |
145 | int i; | |
146 | unsigned long psection; | |
147 | unsigned long start_pfn, start_paddr; | |
148 | struct page *first_page; | |
149 | int ret; | |
150 | int old_state = mem->state; | |
151 | ||
152 | psection = mem->phys_index; | |
153 | first_page = pfn_to_page(psection << PFN_SECTION_SHIFT); | |
154 | ||
155 | /* | |
156 | * The probe routines leave the pages reserved, just | |
157 | * as the bootmem code does. Make sure they're still | |
158 | * that way. | |
159 | */ | |
160 | if (action == MEM_ONLINE) { | |
161 | for (i = 0; i < PAGES_PER_SECTION; i++) { | |
162 | if (PageReserved(first_page+i)) | |
163 | continue; | |
164 | ||
165 | printk(KERN_WARNING "section number %ld page number %d " | |
166 | "not reserved, was it already online? \n", | |
167 | psection, i); | |
168 | return -EBUSY; | |
169 | } | |
170 | } | |
171 | ||
172 | switch (action) { | |
173 | case MEM_ONLINE: | |
174 | start_pfn = page_to_pfn(first_page); | |
175 | ret = online_pages(start_pfn, PAGES_PER_SECTION); | |
176 | break; | |
177 | case MEM_OFFLINE: | |
178 | mem->state = MEM_GOING_OFFLINE; | |
3947be19 DH |
179 | start_paddr = page_to_pfn(first_page) << PAGE_SHIFT; |
180 | ret = remove_memory(start_paddr, | |
181 | PAGES_PER_SECTION << PAGE_SHIFT); | |
182 | if (ret) { | |
183 | mem->state = old_state; | |
184 | break; | |
185 | } | |
3947be19 DH |
186 | break; |
187 | default: | |
188 | printk(KERN_WARNING "%s(%p, %ld) unknown action: %ld\n", | |
189 | __FUNCTION__, mem, action, action); | |
190 | WARN_ON(1); | |
191 | ret = -EINVAL; | |
192 | } | |
3947be19 DH |
193 | |
194 | return ret; | |
195 | } | |
196 | ||
197 | static int memory_block_change_state(struct memory_block *mem, | |
198 | unsigned long to_state, unsigned long from_state_req) | |
199 | { | |
200 | int ret = 0; | |
da19cbcf | 201 | mutex_lock(&mem->state_mutex); |
3947be19 DH |
202 | |
203 | if (mem->state != from_state_req) { | |
204 | ret = -EINVAL; | |
205 | goto out; | |
206 | } | |
207 | ||
208 | ret = memory_block_action(mem, to_state); | |
209 | if (!ret) | |
210 | mem->state = to_state; | |
211 | ||
212 | out: | |
da19cbcf | 213 | mutex_unlock(&mem->state_mutex); |
3947be19 DH |
214 | return ret; |
215 | } | |
216 | ||
217 | static ssize_t | |
218 | store_mem_state(struct sys_device *dev, const char *buf, size_t count) | |
219 | { | |
220 | struct memory_block *mem; | |
221 | unsigned int phys_section_nr; | |
222 | int ret = -EINVAL; | |
223 | ||
224 | mem = container_of(dev, struct memory_block, sysdev); | |
225 | phys_section_nr = mem->phys_index; | |
226 | ||
540557b9 | 227 | if (!present_section_nr(phys_section_nr)) |
3947be19 DH |
228 | goto out; |
229 | ||
230 | if (!strncmp(buf, "online", min((int)count, 6))) | |
231 | ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE); | |
232 | else if(!strncmp(buf, "offline", min((int)count, 7))) | |
233 | ret = memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE); | |
234 | out: | |
235 | if (ret) | |
236 | return ret; | |
237 | return count; | |
238 | } | |
239 | ||
240 | /* | |
241 | * phys_device is a bad name for this. What I really want | |
242 | * is a way to differentiate between memory ranges that | |
243 | * are part of physical devices that constitute | |
244 | * a complete removable unit or fru. | |
245 | * i.e. do these ranges belong to the same physical device, | |
246 | * s.t. if I offline all of these sections I can then | |
247 | * remove the physical device? | |
248 | */ | |
249 | static ssize_t show_phys_device(struct sys_device *dev, char *buf) | |
250 | { | |
251 | struct memory_block *mem = | |
252 | container_of(dev, struct memory_block, sysdev); | |
253 | return sprintf(buf, "%d\n", mem->phys_device); | |
254 | } | |
255 | ||
256 | static SYSDEV_ATTR(phys_index, 0444, show_mem_phys_index, NULL); | |
257 | static SYSDEV_ATTR(state, 0644, show_mem_state, store_mem_state); | |
258 | static SYSDEV_ATTR(phys_device, 0444, show_phys_device, NULL); | |
259 | ||
260 | #define mem_create_simple_file(mem, attr_name) \ | |
261 | sysdev_create_file(&mem->sysdev, &attr_##attr_name) | |
262 | #define mem_remove_simple_file(mem, attr_name) \ | |
263 | sysdev_remove_file(&mem->sysdev, &attr_##attr_name) | |
264 | ||
265 | /* | |
266 | * Block size attribute stuff | |
267 | */ | |
268 | static ssize_t | |
269 | print_block_size(struct class *class, char *buf) | |
270 | { | |
271 | return sprintf(buf, "%lx\n", (unsigned long)PAGES_PER_SECTION * PAGE_SIZE); | |
272 | } | |
273 | ||
274 | static CLASS_ATTR(block_size_bytes, 0444, print_block_size, NULL); | |
275 | ||
276 | static int block_size_init(void) | |
277 | { | |
28ec24e2 AM |
278 | return sysfs_create_file(&memory_sysdev_class.kset.kobj, |
279 | &class_attr_block_size_bytes.attr); | |
3947be19 DH |
280 | } |
281 | ||
282 | /* | |
283 | * Some architectures will have custom drivers to do this, and | |
284 | * will not need to do it from userspace. The fake hot-add code | |
285 | * as well as ppc64 will do all of their discovery in userspace | |
286 | * and will require this interface. | |
287 | */ | |
288 | #ifdef CONFIG_ARCH_MEMORY_PROBE | |
289 | static ssize_t | |
be7ee9b2 | 290 | memory_probe_store(struct class *class, const char *buf, size_t count) |
3947be19 DH |
291 | { |
292 | u64 phys_addr; | |
bc02af93 | 293 | int nid; |
3947be19 DH |
294 | int ret; |
295 | ||
296 | phys_addr = simple_strtoull(buf, NULL, 0); | |
297 | ||
bc02af93 YG |
298 | nid = memory_add_physaddr_to_nid(phys_addr); |
299 | ret = add_memory(nid, phys_addr, PAGES_PER_SECTION << PAGE_SHIFT); | |
3947be19 DH |
300 | |
301 | if (ret) | |
302 | count = ret; | |
303 | ||
304 | return count; | |
305 | } | |
306 | static CLASS_ATTR(probe, 0700, NULL, memory_probe_store); | |
307 | ||
308 | static int memory_probe_init(void) | |
309 | { | |
28ec24e2 AM |
310 | return sysfs_create_file(&memory_sysdev_class.kset.kobj, |
311 | &class_attr_probe.attr); | |
3947be19 DH |
312 | } |
313 | #else | |
28ec24e2 AM |
314 | static inline int memory_probe_init(void) |
315 | { | |
316 | return 0; | |
317 | } | |
3947be19 DH |
318 | #endif |
319 | ||
320 | /* | |
321 | * Note that phys_device is optional. It is here to allow for | |
322 | * differentiation between which *physical* devices each | |
323 | * section belongs to... | |
324 | */ | |
325 | ||
326 | static int add_memory_block(unsigned long node_id, struct mem_section *section, | |
327 | unsigned long state, int phys_device) | |
328 | { | |
0b0acbec | 329 | struct memory_block *mem = kzalloc(sizeof(*mem), GFP_KERNEL); |
3947be19 DH |
330 | int ret = 0; |
331 | ||
332 | if (!mem) | |
333 | return -ENOMEM; | |
334 | ||
3947be19 DH |
335 | mem->phys_index = __section_nr(section); |
336 | mem->state = state; | |
da19cbcf | 337 | mutex_init(&mem->state_mutex); |
3947be19 DH |
338 | mem->phys_device = phys_device; |
339 | ||
00a41db5 | 340 | ret = register_memory(mem, section); |
3947be19 DH |
341 | if (!ret) |
342 | ret = mem_create_simple_file(mem, phys_index); | |
343 | if (!ret) | |
344 | ret = mem_create_simple_file(mem, state); | |
345 | if (!ret) | |
346 | ret = mem_create_simple_file(mem, phys_device); | |
347 | ||
348 | return ret; | |
349 | } | |
350 | ||
351 | /* | |
352 | * For now, we have a linear search to go find the appropriate | |
353 | * memory_block corresponding to a particular phys_index. If | |
354 | * this gets to be a real problem, we can always use a radix | |
355 | * tree or something here. | |
356 | * | |
357 | * This could be made generic for all sysdev classes. | |
358 | */ | |
359 | static struct memory_block *find_memory_block(struct mem_section *section) | |
360 | { | |
361 | struct kobject *kobj; | |
362 | struct sys_device *sysdev; | |
363 | struct memory_block *mem; | |
364 | char name[sizeof(MEMORY_CLASS_NAME) + 9 + 1]; | |
365 | ||
366 | /* | |
367 | * This only works because we know that section == sysdev->id | |
368 | * slightly redundant with sysdev_register() | |
369 | */ | |
370 | sprintf(&name[0], "%s%d", MEMORY_CLASS_NAME, __section_nr(section)); | |
371 | ||
372 | kobj = kset_find_obj(&memory_sysdev_class.kset, name); | |
373 | if (!kobj) | |
374 | return NULL; | |
375 | ||
376 | sysdev = container_of(kobj, struct sys_device, kobj); | |
377 | mem = container_of(sysdev, struct memory_block, sysdev); | |
378 | ||
379 | return mem; | |
380 | } | |
381 | ||
382 | int remove_memory_block(unsigned long node_id, struct mem_section *section, | |
383 | int phys_device) | |
384 | { | |
385 | struct memory_block *mem; | |
386 | ||
387 | mem = find_memory_block(section); | |
388 | mem_remove_simple_file(mem, phys_index); | |
389 | mem_remove_simple_file(mem, state); | |
390 | mem_remove_simple_file(mem, phys_device); | |
00a41db5 | 391 | unregister_memory(mem, section); |
3947be19 DH |
392 | |
393 | return 0; | |
394 | } | |
395 | ||
396 | /* | |
397 | * need an interface for the VM to add new memory regions, | |
398 | * but without onlining it. | |
399 | */ | |
400 | int register_new_memory(struct mem_section *section) | |
401 | { | |
402 | return add_memory_block(0, section, MEM_OFFLINE, 0); | |
403 | } | |
404 | ||
405 | int unregister_memory_section(struct mem_section *section) | |
406 | { | |
540557b9 | 407 | if (!present_section(section)) |
3947be19 DH |
408 | return -EINVAL; |
409 | ||
410 | return remove_memory_block(0, section, 0); | |
411 | } | |
412 | ||
413 | /* | |
414 | * Initialize the sysfs support for memory devices... | |
415 | */ | |
416 | int __init memory_dev_init(void) | |
417 | { | |
418 | unsigned int i; | |
419 | int ret; | |
28ec24e2 | 420 | int err; |
3947be19 | 421 | |
312c004d | 422 | memory_sysdev_class.kset.uevent_ops = &memory_uevent_ops; |
3947be19 | 423 | ret = sysdev_class_register(&memory_sysdev_class); |
28ec24e2 AM |
424 | if (ret) |
425 | goto out; | |
3947be19 DH |
426 | |
427 | /* | |
428 | * Create entries for memory sections that were found | |
429 | * during boot and have been initialized | |
430 | */ | |
431 | for (i = 0; i < NR_MEM_SECTIONS; i++) { | |
540557b9 | 432 | if (!present_section_nr(i)) |
3947be19 | 433 | continue; |
28ec24e2 AM |
434 | err = add_memory_block(0, __nr_to_section(i), MEM_ONLINE, 0); |
435 | if (!ret) | |
436 | ret = err; | |
3947be19 DH |
437 | } |
438 | ||
28ec24e2 AM |
439 | err = memory_probe_init(); |
440 | if (!ret) | |
441 | ret = err; | |
442 | err = block_size_init(); | |
443 | if (!ret) | |
444 | ret = err; | |
445 | out: | |
446 | if (ret) | |
447 | printk(KERN_ERR "%s() failed: %d\n", __FUNCTION__, ret); | |
3947be19 DH |
448 | return ret; |
449 | } |