Commit | Line | Data |
---|---|---|
3947be19 | 1 | /* |
10fbcf4c | 2 | * Memory subsystem support |
3947be19 DH |
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 | ||
3947be19 DH |
13 | #include <linux/module.h> |
14 | #include <linux/init.h> | |
3947be19 | 15 | #include <linux/topology.h> |
c59ede7b | 16 | #include <linux/capability.h> |
3947be19 DH |
17 | #include <linux/device.h> |
18 | #include <linux/memory.h> | |
19 | #include <linux/kobject.h> | |
20 | #include <linux/memory_hotplug.h> | |
21 | #include <linux/mm.h> | |
da19cbcf | 22 | #include <linux/mutex.h> |
9f1b16a5 | 23 | #include <linux/stat.h> |
5a0e3ad6 | 24 | #include <linux/slab.h> |
9f1b16a5 | 25 | |
60063497 | 26 | #include <linux/atomic.h> |
3947be19 DH |
27 | #include <asm/uaccess.h> |
28 | ||
2938ffbd NF |
29 | static DEFINE_MUTEX(mem_sysfs_mutex); |
30 | ||
3947be19 | 31 | #define MEMORY_CLASS_NAME "memory" |
0c2c99b1 NF |
32 | |
33 | static int sections_per_block; | |
34 | ||
35 | static inline int base_memory_block_id(int section_nr) | |
36 | { | |
37 | return section_nr / sections_per_block; | |
38 | } | |
3947be19 | 39 | |
10fbcf4c | 40 | static struct bus_type memory_subsys = { |
af5ca3f4 | 41 | .name = MEMORY_CLASS_NAME, |
10fbcf4c | 42 | .dev_name = MEMORY_CLASS_NAME, |
3947be19 DH |
43 | }; |
44 | ||
e041c683 | 45 | static BLOCKING_NOTIFIER_HEAD(memory_chain); |
3947be19 | 46 | |
98a38ebd | 47 | int register_memory_notifier(struct notifier_block *nb) |
3947be19 | 48 | { |
e041c683 | 49 | return blocking_notifier_chain_register(&memory_chain, nb); |
3947be19 | 50 | } |
3c82c30c | 51 | EXPORT_SYMBOL(register_memory_notifier); |
3947be19 | 52 | |
98a38ebd | 53 | void unregister_memory_notifier(struct notifier_block *nb) |
3947be19 | 54 | { |
e041c683 | 55 | blocking_notifier_chain_unregister(&memory_chain, nb); |
3947be19 | 56 | } |
3c82c30c | 57 | EXPORT_SYMBOL(unregister_memory_notifier); |
3947be19 | 58 | |
925cc71e RJ |
59 | static ATOMIC_NOTIFIER_HEAD(memory_isolate_chain); |
60 | ||
61 | int register_memory_isolate_notifier(struct notifier_block *nb) | |
62 | { | |
63 | return atomic_notifier_chain_register(&memory_isolate_chain, nb); | |
64 | } | |
65 | EXPORT_SYMBOL(register_memory_isolate_notifier); | |
66 | ||
67 | void unregister_memory_isolate_notifier(struct notifier_block *nb) | |
68 | { | |
69 | atomic_notifier_chain_unregister(&memory_isolate_chain, nb); | |
70 | } | |
71 | EXPORT_SYMBOL(unregister_memory_isolate_notifier); | |
72 | ||
fa7194eb YI |
73 | static void memory_block_release(struct device *dev) |
74 | { | |
75 | struct memory_block *mem = container_of(dev, struct memory_block, dev); | |
76 | ||
77 | kfree(mem); | |
78 | } | |
79 | ||
3947be19 DH |
80 | /* |
81 | * register_memory - Setup a sysfs device for a memory block | |
82 | */ | |
00a41db5 | 83 | static |
0c2c99b1 | 84 | int register_memory(struct memory_block *memory) |
3947be19 DH |
85 | { |
86 | int error; | |
87 | ||
10fbcf4c KS |
88 | memory->dev.bus = &memory_subsys; |
89 | memory->dev.id = memory->start_section_nr / sections_per_block; | |
fa7194eb | 90 | memory->dev.release = memory_block_release; |
3947be19 | 91 | |
10fbcf4c | 92 | error = device_register(&memory->dev); |
3947be19 DH |
93 | return error; |
94 | } | |
95 | ||
0c2c99b1 NF |
96 | unsigned long __weak memory_block_size_bytes(void) |
97 | { | |
98 | return MIN_MEMORY_BLOCK_SIZE; | |
99 | } | |
100 | ||
101 | static unsigned long get_memory_block_size(void) | |
102 | { | |
103 | unsigned long block_sz; | |
104 | ||
105 | block_sz = memory_block_size_bytes(); | |
106 | ||
107 | /* Validate blk_sz is a power of 2 and not less than section size */ | |
108 | if ((block_sz & (block_sz - 1)) || (block_sz < MIN_MEMORY_BLOCK_SIZE)) { | |
109 | WARN_ON(1); | |
110 | block_sz = MIN_MEMORY_BLOCK_SIZE; | |
111 | } | |
112 | ||
113 | return block_sz; | |
114 | } | |
115 | ||
3947be19 DH |
116 | /* |
117 | * use this as the physical section index that this memsection | |
118 | * uses. | |
119 | */ | |
120 | ||
10fbcf4c KS |
121 | static ssize_t show_mem_start_phys_index(struct device *dev, |
122 | struct device_attribute *attr, char *buf) | |
3947be19 DH |
123 | { |
124 | struct memory_block *mem = | |
10fbcf4c | 125 | container_of(dev, struct memory_block, dev); |
d3360164 NF |
126 | unsigned long phys_index; |
127 | ||
128 | phys_index = mem->start_section_nr / sections_per_block; | |
129 | return sprintf(buf, "%08lx\n", phys_index); | |
130 | } | |
131 | ||
10fbcf4c KS |
132 | static ssize_t show_mem_end_phys_index(struct device *dev, |
133 | struct device_attribute *attr, char *buf) | |
d3360164 NF |
134 | { |
135 | struct memory_block *mem = | |
10fbcf4c | 136 | container_of(dev, struct memory_block, dev); |
d3360164 NF |
137 | unsigned long phys_index; |
138 | ||
139 | phys_index = mem->end_section_nr / sections_per_block; | |
140 | return sprintf(buf, "%08lx\n", phys_index); | |
3947be19 DH |
141 | } |
142 | ||
5c755e9f BP |
143 | /* |
144 | * Show whether the section of memory is likely to be hot-removable | |
145 | */ | |
10fbcf4c KS |
146 | static ssize_t show_mem_removable(struct device *dev, |
147 | struct device_attribute *attr, char *buf) | |
5c755e9f | 148 | { |
0c2c99b1 NF |
149 | unsigned long i, pfn; |
150 | int ret = 1; | |
5c755e9f | 151 | struct memory_block *mem = |
10fbcf4c | 152 | container_of(dev, struct memory_block, dev); |
5c755e9f | 153 | |
0c2c99b1 | 154 | for (i = 0; i < sections_per_block; i++) { |
d3360164 | 155 | pfn = section_nr_to_pfn(mem->start_section_nr + i); |
0c2c99b1 NF |
156 | ret &= is_mem_section_removable(pfn, PAGES_PER_SECTION); |
157 | } | |
158 | ||
5c755e9f BP |
159 | return sprintf(buf, "%d\n", ret); |
160 | } | |
161 | ||
3947be19 DH |
162 | /* |
163 | * online, offline, going offline, etc. | |
164 | */ | |
10fbcf4c KS |
165 | static ssize_t show_mem_state(struct device *dev, |
166 | struct device_attribute *attr, char *buf) | |
3947be19 DH |
167 | { |
168 | struct memory_block *mem = | |
10fbcf4c | 169 | container_of(dev, struct memory_block, dev); |
3947be19 DH |
170 | ssize_t len = 0; |
171 | ||
172 | /* | |
173 | * We can probably put these states in a nice little array | |
174 | * so that they're not open-coded | |
175 | */ | |
176 | switch (mem->state) { | |
177 | case MEM_ONLINE: | |
178 | len = sprintf(buf, "online\n"); | |
179 | break; | |
180 | case MEM_OFFLINE: | |
181 | len = sprintf(buf, "offline\n"); | |
182 | break; | |
183 | case MEM_GOING_OFFLINE: | |
184 | len = sprintf(buf, "going-offline\n"); | |
185 | break; | |
186 | default: | |
187 | len = sprintf(buf, "ERROR-UNKNOWN-%ld\n", | |
188 | mem->state); | |
189 | WARN_ON(1); | |
190 | break; | |
191 | } | |
192 | ||
193 | return len; | |
194 | } | |
195 | ||
7b78d335 | 196 | int memory_notify(unsigned long val, void *v) |
3947be19 | 197 | { |
e041c683 | 198 | return blocking_notifier_call_chain(&memory_chain, val, v); |
3947be19 DH |
199 | } |
200 | ||
925cc71e RJ |
201 | int memory_isolate_notify(unsigned long val, void *v) |
202 | { | |
203 | return atomic_notifier_call_chain(&memory_isolate_chain, val, v); | |
204 | } | |
205 | ||
2bbcb878 MG |
206 | /* |
207 | * The probe routines leave the pages reserved, just as the bootmem code does. | |
208 | * Make sure they're still that way. | |
209 | */ | |
210 | static bool pages_correctly_reserved(unsigned long start_pfn, | |
211 | unsigned long nr_pages) | |
212 | { | |
213 | int i, j; | |
214 | struct page *page; | |
215 | unsigned long pfn = start_pfn; | |
216 | ||
217 | /* | |
218 | * memmap between sections is not contiguous except with | |
219 | * SPARSEMEM_VMEMMAP. We lookup the page once per section | |
220 | * and assume memmap is contiguous within each section | |
221 | */ | |
222 | for (i = 0; i < sections_per_block; i++, pfn += PAGES_PER_SECTION) { | |
223 | if (WARN_ON_ONCE(!pfn_valid(pfn))) | |
224 | return false; | |
225 | page = pfn_to_page(pfn); | |
226 | ||
227 | for (j = 0; j < PAGES_PER_SECTION; j++) { | |
228 | if (PageReserved(page + j)) | |
229 | continue; | |
230 | ||
231 | printk(KERN_WARNING "section number %ld page number %d " | |
232 | "not reserved, was it already online?\n", | |
233 | pfn_to_section_nr(pfn), j); | |
234 | ||
235 | return false; | |
236 | } | |
237 | } | |
238 | ||
239 | return true; | |
240 | } | |
241 | ||
3947be19 DH |
242 | /* |
243 | * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is | |
244 | * OK to have direct references to sparsemem variables in here. | |
245 | */ | |
246 | static int | |
511c2aba | 247 | memory_block_action(unsigned long phys_index, unsigned long action, int online_type) |
3947be19 | 248 | { |
a16cee10 | 249 | unsigned long start_pfn; |
5409d2cd | 250 | unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block; |
de0ed36a | 251 | struct page *first_page; |
3947be19 | 252 | int ret; |
3947be19 | 253 | |
de0ed36a | 254 | first_page = pfn_to_page(phys_index << PFN_SECTION_SHIFT); |
a16cee10 | 255 | start_pfn = page_to_pfn(first_page); |
de0ed36a | 256 | |
3947be19 DH |
257 | switch (action) { |
258 | case MEM_ONLINE: | |
2bbcb878 MG |
259 | if (!pages_correctly_reserved(start_pfn, nr_pages)) |
260 | return -EBUSY; | |
261 | ||
511c2aba | 262 | ret = online_pages(start_pfn, nr_pages, online_type); |
3947be19 DH |
263 | break; |
264 | case MEM_OFFLINE: | |
a16cee10 | 265 | ret = offline_pages(start_pfn, nr_pages); |
3947be19 DH |
266 | break; |
267 | default: | |
0c2c99b1 NF |
268 | WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: " |
269 | "%ld\n", __func__, phys_index, action, action); | |
3947be19 DH |
270 | ret = -EINVAL; |
271 | } | |
3947be19 DH |
272 | |
273 | return ret; | |
274 | } | |
275 | ||
e90bdb7f | 276 | static int __memory_block_change_state(struct memory_block *mem, |
511c2aba LJ |
277 | unsigned long to_state, unsigned long from_state_req, |
278 | int online_type) | |
3947be19 | 279 | { |
de0ed36a | 280 | int ret = 0; |
0c2c99b1 | 281 | |
3947be19 DH |
282 | if (mem->state != from_state_req) { |
283 | ret = -EINVAL; | |
284 | goto out; | |
285 | } | |
286 | ||
0c2c99b1 NF |
287 | if (to_state == MEM_OFFLINE) |
288 | mem->state = MEM_GOING_OFFLINE; | |
289 | ||
511c2aba | 290 | ret = memory_block_action(mem->start_section_nr, to_state, online_type); |
0c2c99b1 | 291 | |
f5138e42 | 292 | if (ret) { |
0c2c99b1 | 293 | mem->state = from_state_req; |
f5138e42 MH |
294 | goto out; |
295 | } | |
3947be19 | 296 | |
f5138e42 MH |
297 | mem->state = to_state; |
298 | switch (mem->state) { | |
299 | case MEM_OFFLINE: | |
300 | kobject_uevent(&mem->dev.kobj, KOBJ_OFFLINE); | |
301 | break; | |
302 | case MEM_ONLINE: | |
303 | kobject_uevent(&mem->dev.kobj, KOBJ_ONLINE); | |
304 | break; | |
305 | default: | |
306 | break; | |
307 | } | |
3947be19 | 308 | out: |
3947be19 DH |
309 | return ret; |
310 | } | |
311 | ||
e90bdb7f | 312 | static int memory_block_change_state(struct memory_block *mem, |
511c2aba LJ |
313 | unsigned long to_state, unsigned long from_state_req, |
314 | int online_type) | |
e90bdb7f WC |
315 | { |
316 | int ret; | |
317 | ||
318 | mutex_lock(&mem->state_mutex); | |
511c2aba LJ |
319 | ret = __memory_block_change_state(mem, to_state, from_state_req, |
320 | online_type); | |
e90bdb7f WC |
321 | mutex_unlock(&mem->state_mutex); |
322 | ||
323 | return ret; | |
324 | } | |
3947be19 | 325 | static ssize_t |
10fbcf4c KS |
326 | store_mem_state(struct device *dev, |
327 | struct device_attribute *attr, const char *buf, size_t count) | |
3947be19 DH |
328 | { |
329 | struct memory_block *mem; | |
3947be19 DH |
330 | int ret = -EINVAL; |
331 | ||
10fbcf4c | 332 | mem = container_of(dev, struct memory_block, dev); |
3947be19 | 333 | |
511c2aba LJ |
334 | if (!strncmp(buf, "online_kernel", min_t(int, count, 13))) |
335 | ret = memory_block_change_state(mem, MEM_ONLINE, | |
336 | MEM_OFFLINE, ONLINE_KERNEL); | |
337 | else if (!strncmp(buf, "online_movable", min_t(int, count, 14))) | |
338 | ret = memory_block_change_state(mem, MEM_ONLINE, | |
339 | MEM_OFFLINE, ONLINE_MOVABLE); | |
340 | else if (!strncmp(buf, "online", min_t(int, count, 6))) | |
341 | ret = memory_block_change_state(mem, MEM_ONLINE, | |
342 | MEM_OFFLINE, ONLINE_KEEP); | |
343 | else if(!strncmp(buf, "offline", min_t(int, count, 7))) | |
344 | ret = memory_block_change_state(mem, MEM_OFFLINE, | |
345 | MEM_ONLINE, -1); | |
0c2c99b1 | 346 | |
3947be19 DH |
347 | if (ret) |
348 | return ret; | |
349 | return count; | |
350 | } | |
351 | ||
352 | /* | |
353 | * phys_device is a bad name for this. What I really want | |
354 | * is a way to differentiate between memory ranges that | |
355 | * are part of physical devices that constitute | |
356 | * a complete removable unit or fru. | |
357 | * i.e. do these ranges belong to the same physical device, | |
358 | * s.t. if I offline all of these sections I can then | |
359 | * remove the physical device? | |
360 | */ | |
10fbcf4c KS |
361 | static ssize_t show_phys_device(struct device *dev, |
362 | struct device_attribute *attr, char *buf) | |
3947be19 DH |
363 | { |
364 | struct memory_block *mem = | |
10fbcf4c | 365 | container_of(dev, struct memory_block, dev); |
3947be19 DH |
366 | return sprintf(buf, "%d\n", mem->phys_device); |
367 | } | |
368 | ||
10fbcf4c KS |
369 | static DEVICE_ATTR(phys_index, 0444, show_mem_start_phys_index, NULL); |
370 | static DEVICE_ATTR(end_phys_index, 0444, show_mem_end_phys_index, NULL); | |
371 | static DEVICE_ATTR(state, 0644, show_mem_state, store_mem_state); | |
372 | static DEVICE_ATTR(phys_device, 0444, show_phys_device, NULL); | |
373 | static DEVICE_ATTR(removable, 0444, show_mem_removable, NULL); | |
3947be19 DH |
374 | |
375 | #define mem_create_simple_file(mem, attr_name) \ | |
10fbcf4c | 376 | device_create_file(&mem->dev, &dev_attr_##attr_name) |
3947be19 | 377 | #define mem_remove_simple_file(mem, attr_name) \ |
10fbcf4c | 378 | device_remove_file(&mem->dev, &dev_attr_##attr_name) |
3947be19 DH |
379 | |
380 | /* | |
381 | * Block size attribute stuff | |
382 | */ | |
383 | static ssize_t | |
10fbcf4c | 384 | print_block_size(struct device *dev, struct device_attribute *attr, |
8564a6c1 | 385 | char *buf) |
3947be19 | 386 | { |
0c2c99b1 | 387 | return sprintf(buf, "%lx\n", get_memory_block_size()); |
3947be19 DH |
388 | } |
389 | ||
10fbcf4c | 390 | static DEVICE_ATTR(block_size_bytes, 0444, print_block_size, NULL); |
3947be19 DH |
391 | |
392 | static int block_size_init(void) | |
393 | { | |
10fbcf4c KS |
394 | return device_create_file(memory_subsys.dev_root, |
395 | &dev_attr_block_size_bytes); | |
3947be19 DH |
396 | } |
397 | ||
398 | /* | |
399 | * Some architectures will have custom drivers to do this, and | |
400 | * will not need to do it from userspace. The fake hot-add code | |
401 | * as well as ppc64 will do all of their discovery in userspace | |
402 | * and will require this interface. | |
403 | */ | |
404 | #ifdef CONFIG_ARCH_MEMORY_PROBE | |
405 | static ssize_t | |
10fbcf4c | 406 | memory_probe_store(struct device *dev, struct device_attribute *attr, |
28812fe1 | 407 | const char *buf, size_t count) |
3947be19 DH |
408 | { |
409 | u64 phys_addr; | |
bc02af93 | 410 | int nid; |
6add7cd6 | 411 | int i, ret; |
61b94fea | 412 | unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block; |
3947be19 DH |
413 | |
414 | phys_addr = simple_strtoull(buf, NULL, 0); | |
415 | ||
61b94fea AB |
416 | if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1)) |
417 | return -EINVAL; | |
418 | ||
6add7cd6 NF |
419 | for (i = 0; i < sections_per_block; i++) { |
420 | nid = memory_add_physaddr_to_nid(phys_addr); | |
421 | ret = add_memory(nid, phys_addr, | |
422 | PAGES_PER_SECTION << PAGE_SHIFT); | |
423 | if (ret) | |
9f0af69b | 424 | goto out; |
6add7cd6 NF |
425 | |
426 | phys_addr += MIN_MEMORY_BLOCK_SIZE; | |
427 | } | |
3947be19 | 428 | |
9f0af69b NK |
429 | ret = count; |
430 | out: | |
431 | return ret; | |
3947be19 | 432 | } |
10fbcf4c | 433 | static DEVICE_ATTR(probe, S_IWUSR, NULL, memory_probe_store); |
3947be19 DH |
434 | |
435 | static int memory_probe_init(void) | |
436 | { | |
10fbcf4c | 437 | return device_create_file(memory_subsys.dev_root, &dev_attr_probe); |
3947be19 DH |
438 | } |
439 | #else | |
28ec24e2 AM |
440 | static inline int memory_probe_init(void) |
441 | { | |
442 | return 0; | |
443 | } | |
3947be19 DH |
444 | #endif |
445 | ||
facb6011 AK |
446 | #ifdef CONFIG_MEMORY_FAILURE |
447 | /* | |
448 | * Support for offlining pages of memory | |
449 | */ | |
450 | ||
451 | /* Soft offline a page */ | |
452 | static ssize_t | |
10fbcf4c KS |
453 | store_soft_offline_page(struct device *dev, |
454 | struct device_attribute *attr, | |
28812fe1 | 455 | const char *buf, size_t count) |
facb6011 AK |
456 | { |
457 | int ret; | |
458 | u64 pfn; | |
459 | if (!capable(CAP_SYS_ADMIN)) | |
460 | return -EPERM; | |
461 | if (strict_strtoull(buf, 0, &pfn) < 0) | |
462 | return -EINVAL; | |
463 | pfn >>= PAGE_SHIFT; | |
464 | if (!pfn_valid(pfn)) | |
465 | return -ENXIO; | |
466 | ret = soft_offline_page(pfn_to_page(pfn), 0); | |
467 | return ret == 0 ? count : ret; | |
468 | } | |
469 | ||
470 | /* Forcibly offline a page, including killing processes. */ | |
471 | static ssize_t | |
10fbcf4c KS |
472 | store_hard_offline_page(struct device *dev, |
473 | struct device_attribute *attr, | |
28812fe1 | 474 | const char *buf, size_t count) |
facb6011 AK |
475 | { |
476 | int ret; | |
477 | u64 pfn; | |
478 | if (!capable(CAP_SYS_ADMIN)) | |
479 | return -EPERM; | |
480 | if (strict_strtoull(buf, 0, &pfn) < 0) | |
481 | return -EINVAL; | |
482 | pfn >>= PAGE_SHIFT; | |
cd42f4a3 | 483 | ret = memory_failure(pfn, 0, 0); |
facb6011 AK |
484 | return ret ? ret : count; |
485 | } | |
486 | ||
74fef7a8 FB |
487 | static DEVICE_ATTR(soft_offline_page, S_IWUSR, NULL, store_soft_offline_page); |
488 | static DEVICE_ATTR(hard_offline_page, S_IWUSR, NULL, store_hard_offline_page); | |
facb6011 AK |
489 | |
490 | static __init int memory_fail_init(void) | |
491 | { | |
492 | int err; | |
493 | ||
10fbcf4c KS |
494 | err = device_create_file(memory_subsys.dev_root, |
495 | &dev_attr_soft_offline_page); | |
facb6011 | 496 | if (!err) |
10fbcf4c KS |
497 | err = device_create_file(memory_subsys.dev_root, |
498 | &dev_attr_hard_offline_page); | |
facb6011 AK |
499 | return err; |
500 | } | |
501 | #else | |
502 | static inline int memory_fail_init(void) | |
503 | { | |
504 | return 0; | |
505 | } | |
506 | #endif | |
507 | ||
3947be19 DH |
508 | /* |
509 | * Note that phys_device is optional. It is here to allow for | |
510 | * differentiation between which *physical* devices each | |
511 | * section belongs to... | |
512 | */ | |
bc32df00 HC |
513 | int __weak arch_get_memory_phys_device(unsigned long start_pfn) |
514 | { | |
515 | return 0; | |
516 | } | |
3947be19 | 517 | |
10fbcf4c KS |
518 | /* |
519 | * A reference for the returned object is held and the reference for the | |
520 | * hinted object is released. | |
521 | */ | |
98383031 RH |
522 | struct memory_block *find_memory_block_hinted(struct mem_section *section, |
523 | struct memory_block *hint) | |
3947be19 | 524 | { |
0c2c99b1 | 525 | int block_id = base_memory_block_id(__section_nr(section)); |
10fbcf4c KS |
526 | struct device *hintdev = hint ? &hint->dev : NULL; |
527 | struct device *dev; | |
3947be19 | 528 | |
10fbcf4c KS |
529 | dev = subsys_find_device_by_id(&memory_subsys, block_id, hintdev); |
530 | if (hint) | |
531 | put_device(&hint->dev); | |
532 | if (!dev) | |
3947be19 | 533 | return NULL; |
10fbcf4c | 534 | return container_of(dev, struct memory_block, dev); |
3947be19 DH |
535 | } |
536 | ||
98383031 RH |
537 | /* |
538 | * For now, we have a linear search to go find the appropriate | |
539 | * memory_block corresponding to a particular phys_index. If | |
540 | * this gets to be a real problem, we can always use a radix | |
541 | * tree or something here. | |
542 | * | |
10fbcf4c | 543 | * This could be made generic for all device subsystems. |
98383031 RH |
544 | */ |
545 | struct memory_block *find_memory_block(struct mem_section *section) | |
546 | { | |
547 | return find_memory_block_hinted(section, NULL); | |
548 | } | |
549 | ||
0c2c99b1 NF |
550 | static int init_memory_block(struct memory_block **memory, |
551 | struct mem_section *section, unsigned long state) | |
e4619c85 | 552 | { |
0c2c99b1 | 553 | struct memory_block *mem; |
e4619c85 | 554 | unsigned long start_pfn; |
0c2c99b1 | 555 | int scn_nr; |
e4619c85 NF |
556 | int ret = 0; |
557 | ||
0c2c99b1 | 558 | mem = kzalloc(sizeof(*mem), GFP_KERNEL); |
e4619c85 NF |
559 | if (!mem) |
560 | return -ENOMEM; | |
561 | ||
0c2c99b1 | 562 | scn_nr = __section_nr(section); |
d3360164 NF |
563 | mem->start_section_nr = |
564 | base_memory_block_id(scn_nr) * sections_per_block; | |
565 | mem->end_section_nr = mem->start_section_nr + sections_per_block - 1; | |
e4619c85 | 566 | mem->state = state; |
07681215 | 567 | mem->section_count++; |
e4619c85 | 568 | mutex_init(&mem->state_mutex); |
d3360164 | 569 | start_pfn = section_nr_to_pfn(mem->start_section_nr); |
e4619c85 NF |
570 | mem->phys_device = arch_get_memory_phys_device(start_pfn); |
571 | ||
0c2c99b1 | 572 | ret = register_memory(mem); |
e4619c85 NF |
573 | if (!ret) |
574 | ret = mem_create_simple_file(mem, phys_index); | |
d3360164 NF |
575 | if (!ret) |
576 | ret = mem_create_simple_file(mem, end_phys_index); | |
e4619c85 NF |
577 | if (!ret) |
578 | ret = mem_create_simple_file(mem, state); | |
579 | if (!ret) | |
580 | ret = mem_create_simple_file(mem, phys_device); | |
581 | if (!ret) | |
582 | ret = mem_create_simple_file(mem, removable); | |
0c2c99b1 NF |
583 | |
584 | *memory = mem; | |
585 | return ret; | |
586 | } | |
587 | ||
588 | static int add_memory_section(int nid, struct mem_section *section, | |
321bf4ed | 589 | struct memory_block **mem_p, |
0c2c99b1 NF |
590 | unsigned long state, enum mem_add_context context) |
591 | { | |
321bf4ed YL |
592 | struct memory_block *mem = NULL; |
593 | int scn_nr = __section_nr(section); | |
0c2c99b1 NF |
594 | int ret = 0; |
595 | ||
596 | mutex_lock(&mem_sysfs_mutex); | |
597 | ||
321bf4ed YL |
598 | if (context == BOOT) { |
599 | /* same memory block ? */ | |
600 | if (mem_p && *mem_p) | |
601 | if (scn_nr >= (*mem_p)->start_section_nr && | |
602 | scn_nr <= (*mem_p)->end_section_nr) { | |
603 | mem = *mem_p; | |
604 | kobject_get(&mem->dev.kobj); | |
605 | } | |
606 | } else | |
607 | mem = find_memory_block(section); | |
608 | ||
0c2c99b1 NF |
609 | if (mem) { |
610 | mem->section_count++; | |
10fbcf4c | 611 | kobject_put(&mem->dev.kobj); |
321bf4ed | 612 | } else { |
0c2c99b1 | 613 | ret = init_memory_block(&mem, section, state); |
321bf4ed YL |
614 | /* store memory_block pointer for next loop */ |
615 | if (!ret && context == BOOT) | |
616 | if (mem_p) | |
617 | *mem_p = mem; | |
618 | } | |
0c2c99b1 | 619 | |
e4619c85 | 620 | if (!ret) { |
0c2c99b1 NF |
621 | if (context == HOTPLUG && |
622 | mem->section_count == sections_per_block) | |
e4619c85 NF |
623 | ret = register_mem_sect_under_node(mem, nid); |
624 | } | |
625 | ||
2938ffbd | 626 | mutex_unlock(&mem_sysfs_mutex); |
e4619c85 NF |
627 | return ret; |
628 | } | |
629 | ||
4edd7cef DR |
630 | /* |
631 | * need an interface for the VM to add new memory regions, | |
632 | * but without onlining it. | |
633 | */ | |
634 | int register_new_memory(int nid, struct mem_section *section) | |
635 | { | |
636 | return add_memory_section(nid, section, NULL, MEM_OFFLINE, HOTPLUG); | |
637 | } | |
638 | ||
639 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
640 | static void | |
641 | unregister_memory(struct memory_block *memory) | |
642 | { | |
643 | BUG_ON(memory->dev.bus != &memory_subsys); | |
644 | ||
645 | /* drop the ref. we got in remove_memory_block() */ | |
646 | kobject_put(&memory->dev.kobj); | |
647 | device_unregister(&memory->dev); | |
648 | } | |
649 | ||
650 | static int remove_memory_block(unsigned long node_id, | |
651 | struct mem_section *section, int phys_device) | |
3947be19 DH |
652 | { |
653 | struct memory_block *mem; | |
654 | ||
2938ffbd | 655 | mutex_lock(&mem_sysfs_mutex); |
3947be19 | 656 | mem = find_memory_block(section); |
d3360164 | 657 | unregister_mem_sect_under_nodes(mem, __section_nr(section)); |
07681215 NF |
658 | |
659 | mem->section_count--; | |
660 | if (mem->section_count == 0) { | |
07681215 | 661 | mem_remove_simple_file(mem, phys_index); |
d3360164 | 662 | mem_remove_simple_file(mem, end_phys_index); |
07681215 NF |
663 | mem_remove_simple_file(mem, state); |
664 | mem_remove_simple_file(mem, phys_device); | |
665 | mem_remove_simple_file(mem, removable); | |
0c2c99b1 | 666 | unregister_memory(mem); |
0c2c99b1 | 667 | } else |
10fbcf4c | 668 | kobject_put(&mem->dev.kobj); |
3947be19 | 669 | |
2938ffbd | 670 | mutex_unlock(&mem_sysfs_mutex); |
3947be19 DH |
671 | return 0; |
672 | } | |
673 | ||
3947be19 DH |
674 | int unregister_memory_section(struct mem_section *section) |
675 | { | |
540557b9 | 676 | if (!present_section(section)) |
3947be19 DH |
677 | return -EINVAL; |
678 | ||
679 | return remove_memory_block(0, section, 0); | |
680 | } | |
4edd7cef | 681 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
3947be19 | 682 | |
e90bdb7f WC |
683 | /* |
684 | * offline one memory block. If the memory block has been offlined, do nothing. | |
685 | */ | |
686 | int offline_memory_block(struct memory_block *mem) | |
687 | { | |
688 | int ret = 0; | |
689 | ||
690 | mutex_lock(&mem->state_mutex); | |
691 | if (mem->state != MEM_OFFLINE) | |
511c2aba | 692 | ret = __memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE, -1); |
e90bdb7f WC |
693 | mutex_unlock(&mem->state_mutex); |
694 | ||
695 | return ret; | |
696 | } | |
697 | ||
6677e3ea YI |
698 | /* return true if the memory block is offlined, otherwise, return false */ |
699 | bool is_memblock_offlined(struct memory_block *mem) | |
700 | { | |
701 | return mem->state == MEM_OFFLINE; | |
702 | } | |
703 | ||
3947be19 DH |
704 | /* |
705 | * Initialize the sysfs support for memory devices... | |
706 | */ | |
707 | int __init memory_dev_init(void) | |
708 | { | |
709 | unsigned int i; | |
710 | int ret; | |
28ec24e2 | 711 | int err; |
0c2c99b1 | 712 | unsigned long block_sz; |
321bf4ed | 713 | struct memory_block *mem = NULL; |
3947be19 | 714 | |
10fbcf4c | 715 | ret = subsys_system_register(&memory_subsys, NULL); |
28ec24e2 AM |
716 | if (ret) |
717 | goto out; | |
3947be19 | 718 | |
0c2c99b1 NF |
719 | block_sz = get_memory_block_size(); |
720 | sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE; | |
721 | ||
3947be19 DH |
722 | /* |
723 | * Create entries for memory sections that were found | |
724 | * during boot and have been initialized | |
725 | */ | |
726 | for (i = 0; i < NR_MEM_SECTIONS; i++) { | |
540557b9 | 727 | if (!present_section_nr(i)) |
3947be19 | 728 | continue; |
321bf4ed YL |
729 | /* don't need to reuse memory_block if only one per block */ |
730 | err = add_memory_section(0, __nr_to_section(i), | |
731 | (sections_per_block == 1) ? NULL : &mem, | |
732 | MEM_ONLINE, | |
0c2c99b1 | 733 | BOOT); |
28ec24e2 AM |
734 | if (!ret) |
735 | ret = err; | |
3947be19 DH |
736 | } |
737 | ||
28ec24e2 | 738 | err = memory_probe_init(); |
facb6011 AK |
739 | if (!ret) |
740 | ret = err; | |
741 | err = memory_fail_init(); | |
28ec24e2 AM |
742 | if (!ret) |
743 | ret = err; | |
744 | err = block_size_init(); | |
745 | if (!ret) | |
746 | ret = err; | |
747 | out: | |
748 | if (ret) | |
2b3a302a | 749 | printk(KERN_ERR "%s() failed: %d\n", __func__, ret); |
3947be19 DH |
750 | return ret; |
751 | } |