]>
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> |
9f1b16a5 | 24 | #include <linux/stat.h> |
5a0e3ad6 | 25 | #include <linux/slab.h> |
9f1b16a5 | 26 | |
3947be19 DH |
27 | #include <asm/atomic.h> |
28 | #include <asm/uaccess.h> | |
29 | ||
2938ffbd NF |
30 | static DEFINE_MUTEX(mem_sysfs_mutex); |
31 | ||
3947be19 | 32 | #define MEMORY_CLASS_NAME "memory" |
0c2c99b1 NF |
33 | #define MIN_MEMORY_BLOCK_SIZE (1 << SECTION_SIZE_BITS) |
34 | ||
35 | static int sections_per_block; | |
36 | ||
37 | static inline int base_memory_block_id(int section_nr) | |
38 | { | |
39 | return section_nr / sections_per_block; | |
40 | } | |
3947be19 DH |
41 | |
42 | static struct sysdev_class memory_sysdev_class = { | |
af5ca3f4 | 43 | .name = MEMORY_CLASS_NAME, |
3947be19 | 44 | }; |
3947be19 | 45 | |
312c004d | 46 | static const char *memory_uevent_name(struct kset *kset, struct kobject *kobj) |
3947be19 DH |
47 | { |
48 | return MEMORY_CLASS_NAME; | |
49 | } | |
50 | ||
9ec0fd4e | 51 | static int memory_uevent(struct kset *kset, struct kobject *obj, struct kobj_uevent_env *env) |
3947be19 DH |
52 | { |
53 | int retval = 0; | |
54 | ||
55 | return retval; | |
56 | } | |
57 | ||
9cd43611 | 58 | static const struct kset_uevent_ops memory_uevent_ops = { |
312c004d KS |
59 | .name = memory_uevent_name, |
60 | .uevent = memory_uevent, | |
3947be19 DH |
61 | }; |
62 | ||
e041c683 | 63 | static BLOCKING_NOTIFIER_HEAD(memory_chain); |
3947be19 | 64 | |
98a38ebd | 65 | int register_memory_notifier(struct notifier_block *nb) |
3947be19 | 66 | { |
e041c683 | 67 | return blocking_notifier_chain_register(&memory_chain, nb); |
3947be19 | 68 | } |
3c82c30c | 69 | EXPORT_SYMBOL(register_memory_notifier); |
3947be19 | 70 | |
98a38ebd | 71 | void unregister_memory_notifier(struct notifier_block *nb) |
3947be19 | 72 | { |
e041c683 | 73 | blocking_notifier_chain_unregister(&memory_chain, nb); |
3947be19 | 74 | } |
3c82c30c | 75 | EXPORT_SYMBOL(unregister_memory_notifier); |
3947be19 | 76 | |
925cc71e RJ |
77 | static ATOMIC_NOTIFIER_HEAD(memory_isolate_chain); |
78 | ||
79 | int register_memory_isolate_notifier(struct notifier_block *nb) | |
80 | { | |
81 | return atomic_notifier_chain_register(&memory_isolate_chain, nb); | |
82 | } | |
83 | EXPORT_SYMBOL(register_memory_isolate_notifier); | |
84 | ||
85 | void unregister_memory_isolate_notifier(struct notifier_block *nb) | |
86 | { | |
87 | atomic_notifier_chain_unregister(&memory_isolate_chain, nb); | |
88 | } | |
89 | EXPORT_SYMBOL(unregister_memory_isolate_notifier); | |
90 | ||
3947be19 DH |
91 | /* |
92 | * register_memory - Setup a sysfs device for a memory block | |
93 | */ | |
00a41db5 | 94 | static |
0c2c99b1 | 95 | int register_memory(struct memory_block *memory) |
3947be19 DH |
96 | { |
97 | int error; | |
98 | ||
99 | memory->sysdev.cls = &memory_sysdev_class; | |
0c2c99b1 | 100 | memory->sysdev.id = memory->phys_index / sections_per_block; |
3947be19 DH |
101 | |
102 | error = sysdev_register(&memory->sysdev); | |
3947be19 DH |
103 | return error; |
104 | } | |
105 | ||
106 | static void | |
0c2c99b1 | 107 | unregister_memory(struct memory_block *memory) |
3947be19 DH |
108 | { |
109 | BUG_ON(memory->sysdev.cls != &memory_sysdev_class); | |
3947be19 | 110 | |
00a41db5 BP |
111 | /* drop the ref. we got in remove_memory_block() */ |
112 | kobject_put(&memory->sysdev.kobj); | |
3947be19 | 113 | sysdev_unregister(&memory->sysdev); |
3947be19 DH |
114 | } |
115 | ||
0c2c99b1 NF |
116 | unsigned long __weak memory_block_size_bytes(void) |
117 | { | |
118 | return MIN_MEMORY_BLOCK_SIZE; | |
119 | } | |
120 | ||
121 | static unsigned long get_memory_block_size(void) | |
122 | { | |
123 | unsigned long block_sz; | |
124 | ||
125 | block_sz = memory_block_size_bytes(); | |
126 | ||
127 | /* Validate blk_sz is a power of 2 and not less than section size */ | |
128 | if ((block_sz & (block_sz - 1)) || (block_sz < MIN_MEMORY_BLOCK_SIZE)) { | |
129 | WARN_ON(1); | |
130 | block_sz = MIN_MEMORY_BLOCK_SIZE; | |
131 | } | |
132 | ||
133 | return block_sz; | |
134 | } | |
135 | ||
3947be19 DH |
136 | /* |
137 | * use this as the physical section index that this memsection | |
138 | * uses. | |
139 | */ | |
140 | ||
4a0b2b4d AK |
141 | static ssize_t show_mem_phys_index(struct sys_device *dev, |
142 | struct sysdev_attribute *attr, char *buf) | |
3947be19 DH |
143 | { |
144 | struct memory_block *mem = | |
145 | container_of(dev, struct memory_block, sysdev); | |
0c2c99b1 | 146 | return sprintf(buf, "%08lx\n", mem->phys_index / sections_per_block); |
3947be19 DH |
147 | } |
148 | ||
5c755e9f BP |
149 | /* |
150 | * Show whether the section of memory is likely to be hot-removable | |
151 | */ | |
1f07be1c SR |
152 | static ssize_t show_mem_removable(struct sys_device *dev, |
153 | struct sysdev_attribute *attr, char *buf) | |
5c755e9f | 154 | { |
0c2c99b1 NF |
155 | unsigned long i, pfn; |
156 | int ret = 1; | |
5c755e9f BP |
157 | struct memory_block *mem = |
158 | container_of(dev, struct memory_block, sysdev); | |
159 | ||
0c2c99b1 NF |
160 | for (i = 0; i < sections_per_block; i++) { |
161 | pfn = section_nr_to_pfn(mem->phys_index + i); | |
162 | ret &= is_mem_section_removable(pfn, PAGES_PER_SECTION); | |
163 | } | |
164 | ||
5c755e9f BP |
165 | return sprintf(buf, "%d\n", ret); |
166 | } | |
167 | ||
3947be19 DH |
168 | /* |
169 | * online, offline, going offline, etc. | |
170 | */ | |
4a0b2b4d AK |
171 | static ssize_t show_mem_state(struct sys_device *dev, |
172 | struct sysdev_attribute *attr, char *buf) | |
3947be19 DH |
173 | { |
174 | struct memory_block *mem = | |
175 | container_of(dev, struct memory_block, sysdev); | |
176 | ssize_t len = 0; | |
177 | ||
178 | /* | |
179 | * We can probably put these states in a nice little array | |
180 | * so that they're not open-coded | |
181 | */ | |
182 | switch (mem->state) { | |
183 | case MEM_ONLINE: | |
184 | len = sprintf(buf, "online\n"); | |
185 | break; | |
186 | case MEM_OFFLINE: | |
187 | len = sprintf(buf, "offline\n"); | |
188 | break; | |
189 | case MEM_GOING_OFFLINE: | |
190 | len = sprintf(buf, "going-offline\n"); | |
191 | break; | |
192 | default: | |
193 | len = sprintf(buf, "ERROR-UNKNOWN-%ld\n", | |
194 | mem->state); | |
195 | WARN_ON(1); | |
196 | break; | |
197 | } | |
198 | ||
199 | return len; | |
200 | } | |
201 | ||
7b78d335 | 202 | int memory_notify(unsigned long val, void *v) |
3947be19 | 203 | { |
e041c683 | 204 | return blocking_notifier_call_chain(&memory_chain, val, v); |
3947be19 DH |
205 | } |
206 | ||
925cc71e RJ |
207 | int memory_isolate_notify(unsigned long val, void *v) |
208 | { | |
209 | return atomic_notifier_call_chain(&memory_isolate_chain, val, v); | |
210 | } | |
211 | ||
3947be19 DH |
212 | /* |
213 | * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is | |
214 | * OK to have direct references to sparsemem variables in here. | |
215 | */ | |
216 | static int | |
0c2c99b1 | 217 | memory_section_action(unsigned long phys_index, unsigned long action) |
3947be19 DH |
218 | { |
219 | int i; | |
3947be19 DH |
220 | unsigned long start_pfn, start_paddr; |
221 | struct page *first_page; | |
222 | int ret; | |
3947be19 | 223 | |
0c2c99b1 | 224 | first_page = pfn_to_page(phys_index << PFN_SECTION_SHIFT); |
3947be19 DH |
225 | |
226 | /* | |
227 | * The probe routines leave the pages reserved, just | |
228 | * as the bootmem code does. Make sure they're still | |
229 | * that way. | |
230 | */ | |
231 | if (action == MEM_ONLINE) { | |
232 | for (i = 0; i < PAGES_PER_SECTION; i++) { | |
233 | if (PageReserved(first_page+i)) | |
234 | continue; | |
235 | ||
236 | printk(KERN_WARNING "section number %ld page number %d " | |
0c2c99b1 NF |
237 | "not reserved, was it already online?\n", |
238 | phys_index, i); | |
3947be19 DH |
239 | return -EBUSY; |
240 | } | |
241 | } | |
242 | ||
243 | switch (action) { | |
244 | case MEM_ONLINE: | |
245 | start_pfn = page_to_pfn(first_page); | |
246 | ret = online_pages(start_pfn, PAGES_PER_SECTION); | |
247 | break; | |
248 | case MEM_OFFLINE: | |
3947be19 DH |
249 | start_paddr = page_to_pfn(first_page) << PAGE_SHIFT; |
250 | ret = remove_memory(start_paddr, | |
251 | PAGES_PER_SECTION << PAGE_SHIFT); | |
3947be19 DH |
252 | break; |
253 | default: | |
0c2c99b1 NF |
254 | WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: " |
255 | "%ld\n", __func__, phys_index, action, action); | |
3947be19 DH |
256 | ret = -EINVAL; |
257 | } | |
3947be19 DH |
258 | |
259 | return ret; | |
260 | } | |
261 | ||
262 | static int memory_block_change_state(struct memory_block *mem, | |
263 | unsigned long to_state, unsigned long from_state_req) | |
264 | { | |
0c2c99b1 NF |
265 | int i, ret = 0; |
266 | ||
da19cbcf | 267 | mutex_lock(&mem->state_mutex); |
3947be19 DH |
268 | |
269 | if (mem->state != from_state_req) { | |
270 | ret = -EINVAL; | |
271 | goto out; | |
272 | } | |
273 | ||
0c2c99b1 NF |
274 | if (to_state == MEM_OFFLINE) |
275 | mem->state = MEM_GOING_OFFLINE; | |
276 | ||
277 | for (i = 0; i < sections_per_block; i++) { | |
278 | ret = memory_section_action(mem->phys_index + i, to_state); | |
279 | if (ret) | |
280 | break; | |
281 | } | |
282 | ||
283 | if (ret) { | |
284 | for (i = 0; i < sections_per_block; i++) | |
285 | memory_section_action(mem->phys_index + i, | |
286 | from_state_req); | |
287 | ||
288 | mem->state = from_state_req; | |
289 | } else | |
3947be19 DH |
290 | mem->state = to_state; |
291 | ||
292 | out: | |
da19cbcf | 293 | mutex_unlock(&mem->state_mutex); |
3947be19 DH |
294 | return ret; |
295 | } | |
296 | ||
297 | static ssize_t | |
4a0b2b4d AK |
298 | store_mem_state(struct sys_device *dev, |
299 | struct sysdev_attribute *attr, const char *buf, size_t count) | |
3947be19 DH |
300 | { |
301 | struct memory_block *mem; | |
3947be19 DH |
302 | int ret = -EINVAL; |
303 | ||
304 | mem = container_of(dev, struct memory_block, sysdev); | |
3947be19 DH |
305 | |
306 | if (!strncmp(buf, "online", min((int)count, 6))) | |
307 | ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE); | |
308 | else if(!strncmp(buf, "offline", min((int)count, 7))) | |
309 | ret = memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE); | |
0c2c99b1 | 310 | |
3947be19 DH |
311 | if (ret) |
312 | return ret; | |
313 | return count; | |
314 | } | |
315 | ||
316 | /* | |
317 | * phys_device is a bad name for this. What I really want | |
318 | * is a way to differentiate between memory ranges that | |
319 | * are part of physical devices that constitute | |
320 | * a complete removable unit or fru. | |
321 | * i.e. do these ranges belong to the same physical device, | |
322 | * s.t. if I offline all of these sections I can then | |
323 | * remove the physical device? | |
324 | */ | |
4a0b2b4d AK |
325 | static ssize_t show_phys_device(struct sys_device *dev, |
326 | struct sysdev_attribute *attr, char *buf) | |
3947be19 DH |
327 | { |
328 | struct memory_block *mem = | |
329 | container_of(dev, struct memory_block, sysdev); | |
330 | return sprintf(buf, "%d\n", mem->phys_device); | |
331 | } | |
332 | ||
333 | static SYSDEV_ATTR(phys_index, 0444, show_mem_phys_index, NULL); | |
334 | static SYSDEV_ATTR(state, 0644, show_mem_state, store_mem_state); | |
335 | static SYSDEV_ATTR(phys_device, 0444, show_phys_device, NULL); | |
5c755e9f | 336 | static SYSDEV_ATTR(removable, 0444, show_mem_removable, NULL); |
3947be19 DH |
337 | |
338 | #define mem_create_simple_file(mem, attr_name) \ | |
339 | sysdev_create_file(&mem->sysdev, &attr_##attr_name) | |
340 | #define mem_remove_simple_file(mem, attr_name) \ | |
341 | sysdev_remove_file(&mem->sysdev, &attr_##attr_name) | |
342 | ||
343 | /* | |
344 | * Block size attribute stuff | |
345 | */ | |
346 | static ssize_t | |
8564a6c1 AK |
347 | print_block_size(struct sysdev_class *class, struct sysdev_class_attribute *attr, |
348 | char *buf) | |
3947be19 | 349 | { |
0c2c99b1 | 350 | return sprintf(buf, "%lx\n", get_memory_block_size()); |
3947be19 DH |
351 | } |
352 | ||
8564a6c1 | 353 | static SYSDEV_CLASS_ATTR(block_size_bytes, 0444, print_block_size, NULL); |
3947be19 DH |
354 | |
355 | static int block_size_init(void) | |
356 | { | |
28ec24e2 | 357 | return sysfs_create_file(&memory_sysdev_class.kset.kobj, |
8564a6c1 | 358 | &attr_block_size_bytes.attr); |
3947be19 DH |
359 | } |
360 | ||
361 | /* | |
362 | * Some architectures will have custom drivers to do this, and | |
363 | * will not need to do it from userspace. The fake hot-add code | |
364 | * as well as ppc64 will do all of their discovery in userspace | |
365 | * and will require this interface. | |
366 | */ | |
367 | #ifdef CONFIG_ARCH_MEMORY_PROBE | |
368 | static ssize_t | |
28812fe1 AK |
369 | memory_probe_store(struct class *class, struct class_attribute *attr, |
370 | const char *buf, size_t count) | |
3947be19 DH |
371 | { |
372 | u64 phys_addr; | |
bc02af93 | 373 | int nid; |
3947be19 DH |
374 | int ret; |
375 | ||
376 | phys_addr = simple_strtoull(buf, NULL, 0); | |
377 | ||
bc02af93 YG |
378 | nid = memory_add_physaddr_to_nid(phys_addr); |
379 | ret = add_memory(nid, phys_addr, PAGES_PER_SECTION << PAGE_SHIFT); | |
3947be19 DH |
380 | |
381 | if (ret) | |
382 | count = ret; | |
383 | ||
384 | return count; | |
385 | } | |
bd796671 | 386 | static CLASS_ATTR(probe, S_IWUSR, NULL, memory_probe_store); |
3947be19 DH |
387 | |
388 | static int memory_probe_init(void) | |
389 | { | |
28ec24e2 | 390 | return sysfs_create_file(&memory_sysdev_class.kset.kobj, |
bd796671 | 391 | &class_attr_probe.attr); |
3947be19 DH |
392 | } |
393 | #else | |
28ec24e2 AM |
394 | static inline int memory_probe_init(void) |
395 | { | |
396 | return 0; | |
397 | } | |
3947be19 DH |
398 | #endif |
399 | ||
facb6011 AK |
400 | #ifdef CONFIG_MEMORY_FAILURE |
401 | /* | |
402 | * Support for offlining pages of memory | |
403 | */ | |
404 | ||
405 | /* Soft offline a page */ | |
406 | static ssize_t | |
28812fe1 AK |
407 | store_soft_offline_page(struct class *class, |
408 | struct class_attribute *attr, | |
409 | const char *buf, size_t count) | |
facb6011 AK |
410 | { |
411 | int ret; | |
412 | u64 pfn; | |
413 | if (!capable(CAP_SYS_ADMIN)) | |
414 | return -EPERM; | |
415 | if (strict_strtoull(buf, 0, &pfn) < 0) | |
416 | return -EINVAL; | |
417 | pfn >>= PAGE_SHIFT; | |
418 | if (!pfn_valid(pfn)) | |
419 | return -ENXIO; | |
420 | ret = soft_offline_page(pfn_to_page(pfn), 0); | |
421 | return ret == 0 ? count : ret; | |
422 | } | |
423 | ||
424 | /* Forcibly offline a page, including killing processes. */ | |
425 | static ssize_t | |
28812fe1 AK |
426 | store_hard_offline_page(struct class *class, |
427 | struct class_attribute *attr, | |
428 | const char *buf, size_t count) | |
facb6011 AK |
429 | { |
430 | int ret; | |
431 | u64 pfn; | |
432 | if (!capable(CAP_SYS_ADMIN)) | |
433 | return -EPERM; | |
434 | if (strict_strtoull(buf, 0, &pfn) < 0) | |
435 | return -EINVAL; | |
436 | pfn >>= PAGE_SHIFT; | |
437 | ret = __memory_failure(pfn, 0, 0); | |
438 | return ret ? ret : count; | |
439 | } | |
440 | ||
bd796671 GKH |
441 | static CLASS_ATTR(soft_offline_page, 0644, NULL, store_soft_offline_page); |
442 | static CLASS_ATTR(hard_offline_page, 0644, NULL, store_hard_offline_page); | |
facb6011 AK |
443 | |
444 | static __init int memory_fail_init(void) | |
445 | { | |
446 | int err; | |
447 | ||
448 | err = sysfs_create_file(&memory_sysdev_class.kset.kobj, | |
bd796671 | 449 | &class_attr_soft_offline_page.attr); |
facb6011 AK |
450 | if (!err) |
451 | err = sysfs_create_file(&memory_sysdev_class.kset.kobj, | |
bd796671 | 452 | &class_attr_hard_offline_page.attr); |
facb6011 AK |
453 | return err; |
454 | } | |
455 | #else | |
456 | static inline int memory_fail_init(void) | |
457 | { | |
458 | return 0; | |
459 | } | |
460 | #endif | |
461 | ||
3947be19 DH |
462 | /* |
463 | * Note that phys_device is optional. It is here to allow for | |
464 | * differentiation between which *physical* devices each | |
465 | * section belongs to... | |
466 | */ | |
bc32df00 HC |
467 | int __weak arch_get_memory_phys_device(unsigned long start_pfn) |
468 | { | |
469 | return 0; | |
470 | } | |
3947be19 | 471 | |
98383031 RH |
472 | struct memory_block *find_memory_block_hinted(struct mem_section *section, |
473 | struct memory_block *hint) | |
3947be19 DH |
474 | { |
475 | struct kobject *kobj; | |
476 | struct sys_device *sysdev; | |
477 | struct memory_block *mem; | |
478 | char name[sizeof(MEMORY_CLASS_NAME) + 9 + 1]; | |
0c2c99b1 | 479 | int block_id = base_memory_block_id(__section_nr(section)); |
3947be19 | 480 | |
98383031 RH |
481 | kobj = hint ? &hint->sysdev.kobj : NULL; |
482 | ||
3947be19 DH |
483 | /* |
484 | * This only works because we know that section == sysdev->id | |
485 | * slightly redundant with sysdev_register() | |
486 | */ | |
0c2c99b1 | 487 | sprintf(&name[0], "%s%d", MEMORY_CLASS_NAME, block_id); |
3947be19 | 488 | |
98383031 | 489 | kobj = kset_find_obj_hinted(&memory_sysdev_class.kset, name, kobj); |
3947be19 DH |
490 | if (!kobj) |
491 | return NULL; | |
492 | ||
493 | sysdev = container_of(kobj, struct sys_device, kobj); | |
494 | mem = container_of(sysdev, struct memory_block, sysdev); | |
495 | ||
496 | return mem; | |
497 | } | |
498 | ||
98383031 RH |
499 | /* |
500 | * For now, we have a linear search to go find the appropriate | |
501 | * memory_block corresponding to a particular phys_index. If | |
502 | * this gets to be a real problem, we can always use a radix | |
503 | * tree or something here. | |
504 | * | |
505 | * This could be made generic for all sysdev classes. | |
506 | */ | |
507 | struct memory_block *find_memory_block(struct mem_section *section) | |
508 | { | |
509 | return find_memory_block_hinted(section, NULL); | |
510 | } | |
511 | ||
0c2c99b1 NF |
512 | static int init_memory_block(struct memory_block **memory, |
513 | struct mem_section *section, unsigned long state) | |
e4619c85 | 514 | { |
0c2c99b1 | 515 | struct memory_block *mem; |
e4619c85 | 516 | unsigned long start_pfn; |
0c2c99b1 | 517 | int scn_nr; |
e4619c85 NF |
518 | int ret = 0; |
519 | ||
0c2c99b1 | 520 | mem = kzalloc(sizeof(*mem), GFP_KERNEL); |
e4619c85 NF |
521 | if (!mem) |
522 | return -ENOMEM; | |
523 | ||
0c2c99b1 NF |
524 | scn_nr = __section_nr(section); |
525 | mem->phys_index = base_memory_block_id(scn_nr) * sections_per_block; | |
e4619c85 | 526 | mem->state = state; |
07681215 | 527 | mem->section_count++; |
e4619c85 NF |
528 | mutex_init(&mem->state_mutex); |
529 | start_pfn = section_nr_to_pfn(mem->phys_index); | |
530 | mem->phys_device = arch_get_memory_phys_device(start_pfn); | |
531 | ||
0c2c99b1 | 532 | ret = register_memory(mem); |
e4619c85 NF |
533 | if (!ret) |
534 | ret = mem_create_simple_file(mem, phys_index); | |
535 | if (!ret) | |
536 | ret = mem_create_simple_file(mem, state); | |
537 | if (!ret) | |
538 | ret = mem_create_simple_file(mem, phys_device); | |
539 | if (!ret) | |
540 | ret = mem_create_simple_file(mem, removable); | |
0c2c99b1 NF |
541 | |
542 | *memory = mem; | |
543 | return ret; | |
544 | } | |
545 | ||
546 | static int add_memory_section(int nid, struct mem_section *section, | |
547 | unsigned long state, enum mem_add_context context) | |
548 | { | |
549 | struct memory_block *mem; | |
550 | int ret = 0; | |
551 | ||
552 | mutex_lock(&mem_sysfs_mutex); | |
553 | ||
554 | mem = find_memory_block(section); | |
555 | if (mem) { | |
556 | mem->section_count++; | |
557 | kobject_put(&mem->sysdev.kobj); | |
558 | } else | |
559 | ret = init_memory_block(&mem, section, state); | |
560 | ||
e4619c85 | 561 | if (!ret) { |
0c2c99b1 NF |
562 | if (context == HOTPLUG && |
563 | mem->section_count == sections_per_block) | |
e4619c85 NF |
564 | ret = register_mem_sect_under_node(mem, nid); |
565 | } | |
566 | ||
2938ffbd | 567 | mutex_unlock(&mem_sysfs_mutex); |
e4619c85 NF |
568 | return ret; |
569 | } | |
570 | ||
3947be19 DH |
571 | int remove_memory_block(unsigned long node_id, struct mem_section *section, |
572 | int phys_device) | |
573 | { | |
574 | struct memory_block *mem; | |
575 | ||
2938ffbd | 576 | mutex_lock(&mem_sysfs_mutex); |
3947be19 | 577 | mem = find_memory_block(section); |
07681215 NF |
578 | |
579 | mem->section_count--; | |
580 | if (mem->section_count == 0) { | |
581 | unregister_mem_sect_under_nodes(mem); | |
582 | mem_remove_simple_file(mem, phys_index); | |
583 | mem_remove_simple_file(mem, state); | |
584 | mem_remove_simple_file(mem, phys_device); | |
585 | mem_remove_simple_file(mem, removable); | |
0c2c99b1 NF |
586 | unregister_memory(mem); |
587 | kfree(mem); | |
588 | } else | |
589 | kobject_put(&mem->sysdev.kobj); | |
3947be19 | 590 | |
2938ffbd | 591 | mutex_unlock(&mem_sysfs_mutex); |
3947be19 DH |
592 | return 0; |
593 | } | |
594 | ||
595 | /* | |
596 | * need an interface for the VM to add new memory regions, | |
597 | * but without onlining it. | |
598 | */ | |
c04fc586 | 599 | int register_new_memory(int nid, struct mem_section *section) |
3947be19 | 600 | { |
0c2c99b1 | 601 | return add_memory_section(nid, section, MEM_OFFLINE, HOTPLUG); |
3947be19 DH |
602 | } |
603 | ||
604 | int unregister_memory_section(struct mem_section *section) | |
605 | { | |
540557b9 | 606 | if (!present_section(section)) |
3947be19 DH |
607 | return -EINVAL; |
608 | ||
609 | return remove_memory_block(0, section, 0); | |
610 | } | |
611 | ||
612 | /* | |
613 | * Initialize the sysfs support for memory devices... | |
614 | */ | |
615 | int __init memory_dev_init(void) | |
616 | { | |
617 | unsigned int i; | |
618 | int ret; | |
28ec24e2 | 619 | int err; |
0c2c99b1 | 620 | unsigned long block_sz; |
3947be19 | 621 | |
312c004d | 622 | memory_sysdev_class.kset.uevent_ops = &memory_uevent_ops; |
3947be19 | 623 | ret = sysdev_class_register(&memory_sysdev_class); |
28ec24e2 AM |
624 | if (ret) |
625 | goto out; | |
3947be19 | 626 | |
0c2c99b1 NF |
627 | block_sz = get_memory_block_size(); |
628 | sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE; | |
629 | ||
3947be19 DH |
630 | /* |
631 | * Create entries for memory sections that were found | |
632 | * during boot and have been initialized | |
633 | */ | |
634 | for (i = 0; i < NR_MEM_SECTIONS; i++) { | |
540557b9 | 635 | if (!present_section_nr(i)) |
3947be19 | 636 | continue; |
0c2c99b1 NF |
637 | err = add_memory_section(0, __nr_to_section(i), MEM_ONLINE, |
638 | BOOT); | |
28ec24e2 AM |
639 | if (!ret) |
640 | ret = err; | |
3947be19 DH |
641 | } |
642 | ||
28ec24e2 | 643 | err = memory_probe_init(); |
facb6011 AK |
644 | if (!ret) |
645 | ret = err; | |
646 | err = memory_fail_init(); | |
28ec24e2 AM |
647 | if (!ret) |
648 | ret = err; | |
649 | err = block_size_init(); | |
650 | if (!ret) | |
651 | ret = err; | |
652 | out: | |
653 | if (ret) | |
2b3a302a | 654 | printk(KERN_ERR "%s() failed: %d\n", __func__, ret); |
3947be19 DH |
655 | return ret; |
656 | } |