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 | 55 | } |
3c82c30c | 56 | EXPORT_SYMBOL(register_memory_notifier); |
3947be19 | 57 | |
98a38ebd | 58 | void unregister_memory_notifier(struct notifier_block *nb) |
3947be19 | 59 | { |
e041c683 | 60 | blocking_notifier_chain_unregister(&memory_chain, nb); |
3947be19 | 61 | } |
3c82c30c | 62 | EXPORT_SYMBOL(unregister_memory_notifier); |
3947be19 DH |
63 | |
64 | /* | |
65 | * register_memory - Setup a sysfs device for a memory block | |
66 | */ | |
00a41db5 BP |
67 | static |
68 | int register_memory(struct memory_block *memory, struct mem_section *section) | |
3947be19 DH |
69 | { |
70 | int error; | |
71 | ||
72 | memory->sysdev.cls = &memory_sysdev_class; | |
73 | memory->sysdev.id = __section_nr(section); | |
74 | ||
75 | error = sysdev_register(&memory->sysdev); | |
3947be19 DH |
76 | return error; |
77 | } | |
78 | ||
79 | static void | |
00a41db5 | 80 | unregister_memory(struct memory_block *memory, struct mem_section *section) |
3947be19 DH |
81 | { |
82 | BUG_ON(memory->sysdev.cls != &memory_sysdev_class); | |
83 | BUG_ON(memory->sysdev.id != __section_nr(section)); | |
84 | ||
00a41db5 BP |
85 | /* drop the ref. we got in remove_memory_block() */ |
86 | kobject_put(&memory->sysdev.kobj); | |
3947be19 | 87 | sysdev_unregister(&memory->sysdev); |
3947be19 DH |
88 | } |
89 | ||
90 | /* | |
91 | * use this as the physical section index that this memsection | |
92 | * uses. | |
93 | */ | |
94 | ||
4a0b2b4d AK |
95 | static ssize_t show_mem_phys_index(struct sys_device *dev, |
96 | struct sysdev_attribute *attr, char *buf) | |
3947be19 DH |
97 | { |
98 | struct memory_block *mem = | |
99 | container_of(dev, struct memory_block, sysdev); | |
100 | return sprintf(buf, "%08lx\n", mem->phys_index); | |
101 | } | |
102 | ||
5c755e9f BP |
103 | /* |
104 | * Show whether the section of memory is likely to be hot-removable | |
105 | */ | |
1f07be1c SR |
106 | static ssize_t show_mem_removable(struct sys_device *dev, |
107 | struct sysdev_attribute *attr, char *buf) | |
5c755e9f BP |
108 | { |
109 | unsigned long start_pfn; | |
110 | int ret; | |
111 | struct memory_block *mem = | |
112 | container_of(dev, struct memory_block, sysdev); | |
113 | ||
114 | start_pfn = section_nr_to_pfn(mem->phys_index); | |
115 | ret = is_mem_section_removable(start_pfn, PAGES_PER_SECTION); | |
116 | return sprintf(buf, "%d\n", ret); | |
117 | } | |
118 | ||
3947be19 DH |
119 | /* |
120 | * online, offline, going offline, etc. | |
121 | */ | |
4a0b2b4d AK |
122 | static ssize_t show_mem_state(struct sys_device *dev, |
123 | struct sysdev_attribute *attr, char *buf) | |
3947be19 DH |
124 | { |
125 | struct memory_block *mem = | |
126 | container_of(dev, struct memory_block, sysdev); | |
127 | ssize_t len = 0; | |
128 | ||
129 | /* | |
130 | * We can probably put these states in a nice little array | |
131 | * so that they're not open-coded | |
132 | */ | |
133 | switch (mem->state) { | |
134 | case MEM_ONLINE: | |
135 | len = sprintf(buf, "online\n"); | |
136 | break; | |
137 | case MEM_OFFLINE: | |
138 | len = sprintf(buf, "offline\n"); | |
139 | break; | |
140 | case MEM_GOING_OFFLINE: | |
141 | len = sprintf(buf, "going-offline\n"); | |
142 | break; | |
143 | default: | |
144 | len = sprintf(buf, "ERROR-UNKNOWN-%ld\n", | |
145 | mem->state); | |
146 | WARN_ON(1); | |
147 | break; | |
148 | } | |
149 | ||
150 | return len; | |
151 | } | |
152 | ||
7b78d335 | 153 | int memory_notify(unsigned long val, void *v) |
3947be19 | 154 | { |
e041c683 | 155 | return blocking_notifier_call_chain(&memory_chain, val, v); |
3947be19 DH |
156 | } |
157 | ||
158 | /* | |
159 | * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is | |
160 | * OK to have direct references to sparsemem variables in here. | |
161 | */ | |
162 | static int | |
163 | memory_block_action(struct memory_block *mem, unsigned long action) | |
164 | { | |
165 | int i; | |
166 | unsigned long psection; | |
167 | unsigned long start_pfn, start_paddr; | |
168 | struct page *first_page; | |
169 | int ret; | |
170 | int old_state = mem->state; | |
171 | ||
172 | psection = mem->phys_index; | |
173 | first_page = pfn_to_page(psection << PFN_SECTION_SHIFT); | |
174 | ||
175 | /* | |
176 | * The probe routines leave the pages reserved, just | |
177 | * as the bootmem code does. Make sure they're still | |
178 | * that way. | |
179 | */ | |
180 | if (action == MEM_ONLINE) { | |
181 | for (i = 0; i < PAGES_PER_SECTION; i++) { | |
182 | if (PageReserved(first_page+i)) | |
183 | continue; | |
184 | ||
185 | printk(KERN_WARNING "section number %ld page number %d " | |
186 | "not reserved, was it already online? \n", | |
187 | psection, i); | |
188 | return -EBUSY; | |
189 | } | |
190 | } | |
191 | ||
192 | switch (action) { | |
193 | case MEM_ONLINE: | |
194 | start_pfn = page_to_pfn(first_page); | |
195 | ret = online_pages(start_pfn, PAGES_PER_SECTION); | |
196 | break; | |
197 | case MEM_OFFLINE: | |
198 | mem->state = MEM_GOING_OFFLINE; | |
3947be19 DH |
199 | start_paddr = page_to_pfn(first_page) << PAGE_SHIFT; |
200 | ret = remove_memory(start_paddr, | |
201 | PAGES_PER_SECTION << PAGE_SHIFT); | |
202 | if (ret) { | |
203 | mem->state = old_state; | |
204 | break; | |
205 | } | |
3947be19 DH |
206 | break; |
207 | default: | |
f810a5cf | 208 | WARN(1, KERN_WARNING "%s(%p, %ld) unknown action: %ld\n", |
2b3a302a | 209 | __func__, mem, action, action); |
3947be19 DH |
210 | ret = -EINVAL; |
211 | } | |
3947be19 DH |
212 | |
213 | return ret; | |
214 | } | |
215 | ||
216 | static int memory_block_change_state(struct memory_block *mem, | |
217 | unsigned long to_state, unsigned long from_state_req) | |
218 | { | |
219 | int ret = 0; | |
da19cbcf | 220 | mutex_lock(&mem->state_mutex); |
3947be19 DH |
221 | |
222 | if (mem->state != from_state_req) { | |
223 | ret = -EINVAL; | |
224 | goto out; | |
225 | } | |
226 | ||
227 | ret = memory_block_action(mem, to_state); | |
228 | if (!ret) | |
229 | mem->state = to_state; | |
230 | ||
231 | out: | |
da19cbcf | 232 | mutex_unlock(&mem->state_mutex); |
3947be19 DH |
233 | return ret; |
234 | } | |
235 | ||
236 | static ssize_t | |
4a0b2b4d AK |
237 | store_mem_state(struct sys_device *dev, |
238 | struct sysdev_attribute *attr, const char *buf, size_t count) | |
3947be19 DH |
239 | { |
240 | struct memory_block *mem; | |
241 | unsigned int phys_section_nr; | |
242 | int ret = -EINVAL; | |
243 | ||
244 | mem = container_of(dev, struct memory_block, sysdev); | |
245 | phys_section_nr = mem->phys_index; | |
246 | ||
540557b9 | 247 | if (!present_section_nr(phys_section_nr)) |
3947be19 DH |
248 | goto out; |
249 | ||
250 | if (!strncmp(buf, "online", min((int)count, 6))) | |
251 | ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE); | |
252 | else if(!strncmp(buf, "offline", min((int)count, 7))) | |
253 | ret = memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE); | |
254 | out: | |
255 | if (ret) | |
256 | return ret; | |
257 | return count; | |
258 | } | |
259 | ||
260 | /* | |
261 | * phys_device is a bad name for this. What I really want | |
262 | * is a way to differentiate between memory ranges that | |
263 | * are part of physical devices that constitute | |
264 | * a complete removable unit or fru. | |
265 | * i.e. do these ranges belong to the same physical device, | |
266 | * s.t. if I offline all of these sections I can then | |
267 | * remove the physical device? | |
268 | */ | |
4a0b2b4d AK |
269 | static ssize_t show_phys_device(struct sys_device *dev, |
270 | struct sysdev_attribute *attr, char *buf) | |
3947be19 DH |
271 | { |
272 | struct memory_block *mem = | |
273 | container_of(dev, struct memory_block, sysdev); | |
274 | return sprintf(buf, "%d\n", mem->phys_device); | |
275 | } | |
276 | ||
277 | static SYSDEV_ATTR(phys_index, 0444, show_mem_phys_index, NULL); | |
278 | static SYSDEV_ATTR(state, 0644, show_mem_state, store_mem_state); | |
279 | static SYSDEV_ATTR(phys_device, 0444, show_phys_device, NULL); | |
5c755e9f | 280 | static SYSDEV_ATTR(removable, 0444, show_mem_removable, NULL); |
3947be19 DH |
281 | |
282 | #define mem_create_simple_file(mem, attr_name) \ | |
283 | sysdev_create_file(&mem->sysdev, &attr_##attr_name) | |
284 | #define mem_remove_simple_file(mem, attr_name) \ | |
285 | sysdev_remove_file(&mem->sysdev, &attr_##attr_name) | |
286 | ||
287 | /* | |
288 | * Block size attribute stuff | |
289 | */ | |
290 | static ssize_t | |
291 | print_block_size(struct class *class, char *buf) | |
292 | { | |
293 | return sprintf(buf, "%lx\n", (unsigned long)PAGES_PER_SECTION * PAGE_SIZE); | |
294 | } | |
295 | ||
296 | static CLASS_ATTR(block_size_bytes, 0444, print_block_size, NULL); | |
297 | ||
298 | static int block_size_init(void) | |
299 | { | |
28ec24e2 AM |
300 | return sysfs_create_file(&memory_sysdev_class.kset.kobj, |
301 | &class_attr_block_size_bytes.attr); | |
3947be19 DH |
302 | } |
303 | ||
304 | /* | |
305 | * Some architectures will have custom drivers to do this, and | |
306 | * will not need to do it from userspace. The fake hot-add code | |
307 | * as well as ppc64 will do all of their discovery in userspace | |
308 | * and will require this interface. | |
309 | */ | |
310 | #ifdef CONFIG_ARCH_MEMORY_PROBE | |
311 | static ssize_t | |
be7ee9b2 | 312 | memory_probe_store(struct class *class, const char *buf, size_t count) |
3947be19 DH |
313 | { |
314 | u64 phys_addr; | |
bc02af93 | 315 | int nid; |
3947be19 DH |
316 | int ret; |
317 | ||
318 | phys_addr = simple_strtoull(buf, NULL, 0); | |
319 | ||
bc02af93 YG |
320 | nid = memory_add_physaddr_to_nid(phys_addr); |
321 | ret = add_memory(nid, phys_addr, PAGES_PER_SECTION << PAGE_SHIFT); | |
3947be19 DH |
322 | |
323 | if (ret) | |
324 | count = ret; | |
325 | ||
326 | return count; | |
327 | } | |
328 | static CLASS_ATTR(probe, 0700, NULL, memory_probe_store); | |
329 | ||
330 | static int memory_probe_init(void) | |
331 | { | |
28ec24e2 AM |
332 | return sysfs_create_file(&memory_sysdev_class.kset.kobj, |
333 | &class_attr_probe.attr); | |
3947be19 DH |
334 | } |
335 | #else | |
28ec24e2 AM |
336 | static inline int memory_probe_init(void) |
337 | { | |
338 | return 0; | |
339 | } | |
3947be19 DH |
340 | #endif |
341 | ||
342 | /* | |
343 | * Note that phys_device is optional. It is here to allow for | |
344 | * differentiation between which *physical* devices each | |
345 | * section belongs to... | |
346 | */ | |
347 | ||
348 | static int add_memory_block(unsigned long node_id, struct mem_section *section, | |
349 | unsigned long state, int phys_device) | |
350 | { | |
0b0acbec | 351 | struct memory_block *mem = kzalloc(sizeof(*mem), GFP_KERNEL); |
3947be19 DH |
352 | int ret = 0; |
353 | ||
354 | if (!mem) | |
355 | return -ENOMEM; | |
356 | ||
3947be19 DH |
357 | mem->phys_index = __section_nr(section); |
358 | mem->state = state; | |
da19cbcf | 359 | mutex_init(&mem->state_mutex); |
3947be19 DH |
360 | mem->phys_device = phys_device; |
361 | ||
00a41db5 | 362 | ret = register_memory(mem, section); |
3947be19 DH |
363 | if (!ret) |
364 | ret = mem_create_simple_file(mem, phys_index); | |
365 | if (!ret) | |
366 | ret = mem_create_simple_file(mem, state); | |
367 | if (!ret) | |
368 | ret = mem_create_simple_file(mem, phys_device); | |
5c755e9f BP |
369 | if (!ret) |
370 | ret = mem_create_simple_file(mem, removable); | |
3947be19 DH |
371 | |
372 | return ret; | |
373 | } | |
374 | ||
375 | /* | |
376 | * For now, we have a linear search to go find the appropriate | |
377 | * memory_block corresponding to a particular phys_index. If | |
378 | * this gets to be a real problem, we can always use a radix | |
379 | * tree or something here. | |
380 | * | |
381 | * This could be made generic for all sysdev classes. | |
382 | */ | |
383 | static struct memory_block *find_memory_block(struct mem_section *section) | |
384 | { | |
385 | struct kobject *kobj; | |
386 | struct sys_device *sysdev; | |
387 | struct memory_block *mem; | |
388 | char name[sizeof(MEMORY_CLASS_NAME) + 9 + 1]; | |
389 | ||
390 | /* | |
391 | * This only works because we know that section == sysdev->id | |
392 | * slightly redundant with sysdev_register() | |
393 | */ | |
394 | sprintf(&name[0], "%s%d", MEMORY_CLASS_NAME, __section_nr(section)); | |
395 | ||
396 | kobj = kset_find_obj(&memory_sysdev_class.kset, name); | |
397 | if (!kobj) | |
398 | return NULL; | |
399 | ||
400 | sysdev = container_of(kobj, struct sys_device, kobj); | |
401 | mem = container_of(sysdev, struct memory_block, sysdev); | |
402 | ||
403 | return mem; | |
404 | } | |
405 | ||
406 | int remove_memory_block(unsigned long node_id, struct mem_section *section, | |
407 | int phys_device) | |
408 | { | |
409 | struct memory_block *mem; | |
410 | ||
411 | mem = find_memory_block(section); | |
412 | mem_remove_simple_file(mem, phys_index); | |
413 | mem_remove_simple_file(mem, state); | |
414 | mem_remove_simple_file(mem, phys_device); | |
5c755e9f | 415 | mem_remove_simple_file(mem, removable); |
00a41db5 | 416 | unregister_memory(mem, section); |
3947be19 DH |
417 | |
418 | return 0; | |
419 | } | |
420 | ||
421 | /* | |
422 | * need an interface for the VM to add new memory regions, | |
423 | * but without onlining it. | |
424 | */ | |
425 | int register_new_memory(struct mem_section *section) | |
426 | { | |
427 | return add_memory_block(0, section, MEM_OFFLINE, 0); | |
428 | } | |
429 | ||
430 | int unregister_memory_section(struct mem_section *section) | |
431 | { | |
540557b9 | 432 | if (!present_section(section)) |
3947be19 DH |
433 | return -EINVAL; |
434 | ||
435 | return remove_memory_block(0, section, 0); | |
436 | } | |
437 | ||
438 | /* | |
439 | * Initialize the sysfs support for memory devices... | |
440 | */ | |
441 | int __init memory_dev_init(void) | |
442 | { | |
443 | unsigned int i; | |
444 | int ret; | |
28ec24e2 | 445 | int err; |
3947be19 | 446 | |
312c004d | 447 | memory_sysdev_class.kset.uevent_ops = &memory_uevent_ops; |
3947be19 | 448 | ret = sysdev_class_register(&memory_sysdev_class); |
28ec24e2 AM |
449 | if (ret) |
450 | goto out; | |
3947be19 DH |
451 | |
452 | /* | |
453 | * Create entries for memory sections that were found | |
454 | * during boot and have been initialized | |
455 | */ | |
456 | for (i = 0; i < NR_MEM_SECTIONS; i++) { | |
540557b9 | 457 | if (!present_section_nr(i)) |
3947be19 | 458 | continue; |
28ec24e2 AM |
459 | err = add_memory_block(0, __nr_to_section(i), MEM_ONLINE, 0); |
460 | if (!ret) | |
461 | ret = err; | |
3947be19 DH |
462 | } |
463 | ||
28ec24e2 AM |
464 | err = memory_probe_init(); |
465 | if (!ret) | |
466 | ret = err; | |
467 | err = block_size_init(); | |
468 | if (!ret) | |
469 | ret = err; | |
470 | out: | |
471 | if (ret) | |
2b3a302a | 472 | printk(KERN_ERR "%s() failed: %d\n", __func__, ret); |
3947be19 DH |
473 | return ret; |
474 | } |