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> | |
3947be19 DH |
19 | #include <linux/memory_hotplug.h> |
20 | #include <linux/mm.h> | |
da19cbcf | 21 | #include <linux/mutex.h> |
9f1b16a5 | 22 | #include <linux/stat.h> |
5a0e3ad6 | 23 | #include <linux/slab.h> |
9f1b16a5 | 24 | |
60063497 | 25 | #include <linux/atomic.h> |
3947be19 DH |
26 | #include <asm/uaccess.h> |
27 | ||
2938ffbd NF |
28 | static DEFINE_MUTEX(mem_sysfs_mutex); |
29 | ||
3947be19 | 30 | #define MEMORY_CLASS_NAME "memory" |
0c2c99b1 NF |
31 | |
32 | static int sections_per_block; | |
33 | ||
34 | static inline int base_memory_block_id(int section_nr) | |
35 | { | |
36 | return section_nr / sections_per_block; | |
37 | } | |
3947be19 | 38 | |
4960e05e RW |
39 | static int memory_subsys_online(struct device *dev); |
40 | static int memory_subsys_offline(struct device *dev); | |
41 | ||
10fbcf4c | 42 | static struct bus_type memory_subsys = { |
af5ca3f4 | 43 | .name = MEMORY_CLASS_NAME, |
10fbcf4c | 44 | .dev_name = MEMORY_CLASS_NAME, |
4960e05e RW |
45 | .online = memory_subsys_online, |
46 | .offline = memory_subsys_offline, | |
3947be19 DH |
47 | }; |
48 | ||
e041c683 | 49 | static BLOCKING_NOTIFIER_HEAD(memory_chain); |
3947be19 | 50 | |
98a38ebd | 51 | int register_memory_notifier(struct notifier_block *nb) |
3947be19 | 52 | { |
e041c683 | 53 | return blocking_notifier_chain_register(&memory_chain, nb); |
3947be19 | 54 | } |
3c82c30c | 55 | EXPORT_SYMBOL(register_memory_notifier); |
3947be19 | 56 | |
98a38ebd | 57 | void unregister_memory_notifier(struct notifier_block *nb) |
3947be19 | 58 | { |
e041c683 | 59 | blocking_notifier_chain_unregister(&memory_chain, nb); |
3947be19 | 60 | } |
3c82c30c | 61 | EXPORT_SYMBOL(unregister_memory_notifier); |
3947be19 | 62 | |
925cc71e RJ |
63 | static ATOMIC_NOTIFIER_HEAD(memory_isolate_chain); |
64 | ||
65 | int register_memory_isolate_notifier(struct notifier_block *nb) | |
66 | { | |
67 | return atomic_notifier_chain_register(&memory_isolate_chain, nb); | |
68 | } | |
69 | EXPORT_SYMBOL(register_memory_isolate_notifier); | |
70 | ||
71 | void unregister_memory_isolate_notifier(struct notifier_block *nb) | |
72 | { | |
73 | atomic_notifier_chain_unregister(&memory_isolate_chain, nb); | |
74 | } | |
75 | EXPORT_SYMBOL(unregister_memory_isolate_notifier); | |
76 | ||
fa7194eb YI |
77 | static void memory_block_release(struct device *dev) |
78 | { | |
79 | struct memory_block *mem = container_of(dev, struct memory_block, dev); | |
80 | ||
81 | kfree(mem); | |
82 | } | |
83 | ||
0c2c99b1 NF |
84 | unsigned long __weak memory_block_size_bytes(void) |
85 | { | |
86 | return MIN_MEMORY_BLOCK_SIZE; | |
87 | } | |
88 | ||
89 | static unsigned long get_memory_block_size(void) | |
90 | { | |
91 | unsigned long block_sz; | |
92 | ||
93 | block_sz = memory_block_size_bytes(); | |
94 | ||
95 | /* Validate blk_sz is a power of 2 and not less than section size */ | |
96 | if ((block_sz & (block_sz - 1)) || (block_sz < MIN_MEMORY_BLOCK_SIZE)) { | |
97 | WARN_ON(1); | |
98 | block_sz = MIN_MEMORY_BLOCK_SIZE; | |
99 | } | |
100 | ||
101 | return block_sz; | |
102 | } | |
103 | ||
3947be19 DH |
104 | /* |
105 | * use this as the physical section index that this memsection | |
106 | * uses. | |
107 | */ | |
108 | ||
10fbcf4c KS |
109 | static ssize_t show_mem_start_phys_index(struct device *dev, |
110 | struct device_attribute *attr, char *buf) | |
3947be19 DH |
111 | { |
112 | struct memory_block *mem = | |
10fbcf4c | 113 | container_of(dev, struct memory_block, dev); |
d3360164 NF |
114 | unsigned long phys_index; |
115 | ||
116 | phys_index = mem->start_section_nr / sections_per_block; | |
117 | return sprintf(buf, "%08lx\n", phys_index); | |
118 | } | |
119 | ||
10fbcf4c KS |
120 | static ssize_t show_mem_end_phys_index(struct device *dev, |
121 | struct device_attribute *attr, char *buf) | |
d3360164 NF |
122 | { |
123 | struct memory_block *mem = | |
10fbcf4c | 124 | container_of(dev, struct memory_block, dev); |
d3360164 NF |
125 | unsigned long phys_index; |
126 | ||
127 | phys_index = mem->end_section_nr / sections_per_block; | |
128 | return sprintf(buf, "%08lx\n", phys_index); | |
3947be19 DH |
129 | } |
130 | ||
5c755e9f BP |
131 | /* |
132 | * Show whether the section of memory is likely to be hot-removable | |
133 | */ | |
10fbcf4c KS |
134 | static ssize_t show_mem_removable(struct device *dev, |
135 | struct device_attribute *attr, char *buf) | |
5c755e9f | 136 | { |
0c2c99b1 NF |
137 | unsigned long i, pfn; |
138 | int ret = 1; | |
5c755e9f | 139 | struct memory_block *mem = |
10fbcf4c | 140 | container_of(dev, struct memory_block, dev); |
5c755e9f | 141 | |
0c2c99b1 | 142 | for (i = 0; i < sections_per_block; i++) { |
d3360164 | 143 | pfn = section_nr_to_pfn(mem->start_section_nr + i); |
0c2c99b1 NF |
144 | ret &= is_mem_section_removable(pfn, PAGES_PER_SECTION); |
145 | } | |
146 | ||
5c755e9f BP |
147 | return sprintf(buf, "%d\n", ret); |
148 | } | |
149 | ||
3947be19 DH |
150 | /* |
151 | * online, offline, going offline, etc. | |
152 | */ | |
10fbcf4c KS |
153 | static ssize_t show_mem_state(struct device *dev, |
154 | struct device_attribute *attr, char *buf) | |
3947be19 DH |
155 | { |
156 | struct memory_block *mem = | |
10fbcf4c | 157 | container_of(dev, struct memory_block, dev); |
3947be19 DH |
158 | ssize_t len = 0; |
159 | ||
160 | /* | |
161 | * We can probably put these states in a nice little array | |
162 | * so that they're not open-coded | |
163 | */ | |
164 | switch (mem->state) { | |
165 | case MEM_ONLINE: | |
166 | len = sprintf(buf, "online\n"); | |
167 | break; | |
168 | case MEM_OFFLINE: | |
169 | len = sprintf(buf, "offline\n"); | |
170 | break; | |
171 | case MEM_GOING_OFFLINE: | |
172 | len = sprintf(buf, "going-offline\n"); | |
173 | break; | |
174 | default: | |
175 | len = sprintf(buf, "ERROR-UNKNOWN-%ld\n", | |
176 | mem->state); | |
177 | WARN_ON(1); | |
178 | break; | |
179 | } | |
180 | ||
181 | return len; | |
182 | } | |
183 | ||
7b78d335 | 184 | int memory_notify(unsigned long val, void *v) |
3947be19 | 185 | { |
e041c683 | 186 | return blocking_notifier_call_chain(&memory_chain, val, v); |
3947be19 DH |
187 | } |
188 | ||
925cc71e RJ |
189 | int memory_isolate_notify(unsigned long val, void *v) |
190 | { | |
191 | return atomic_notifier_call_chain(&memory_isolate_chain, val, v); | |
192 | } | |
193 | ||
2bbcb878 MG |
194 | /* |
195 | * The probe routines leave the pages reserved, just as the bootmem code does. | |
196 | * Make sure they're still that way. | |
197 | */ | |
6056d619 | 198 | static bool pages_correctly_reserved(unsigned long start_pfn) |
2bbcb878 MG |
199 | { |
200 | int i, j; | |
201 | struct page *page; | |
202 | unsigned long pfn = start_pfn; | |
203 | ||
204 | /* | |
205 | * memmap between sections is not contiguous except with | |
206 | * SPARSEMEM_VMEMMAP. We lookup the page once per section | |
207 | * and assume memmap is contiguous within each section | |
208 | */ | |
209 | for (i = 0; i < sections_per_block; i++, pfn += PAGES_PER_SECTION) { | |
210 | if (WARN_ON_ONCE(!pfn_valid(pfn))) | |
211 | return false; | |
212 | page = pfn_to_page(pfn); | |
213 | ||
214 | for (j = 0; j < PAGES_PER_SECTION; j++) { | |
215 | if (PageReserved(page + j)) | |
216 | continue; | |
217 | ||
218 | printk(KERN_WARNING "section number %ld page number %d " | |
219 | "not reserved, was it already online?\n", | |
220 | pfn_to_section_nr(pfn), j); | |
221 | ||
222 | return false; | |
223 | } | |
224 | } | |
225 | ||
226 | return true; | |
227 | } | |
228 | ||
3947be19 DH |
229 | /* |
230 | * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is | |
231 | * OK to have direct references to sparsemem variables in here. | |
232 | */ | |
233 | static int | |
511c2aba | 234 | memory_block_action(unsigned long phys_index, unsigned long action, int online_type) |
3947be19 | 235 | { |
a16cee10 | 236 | unsigned long start_pfn; |
5409d2cd | 237 | unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block; |
de0ed36a | 238 | struct page *first_page; |
3947be19 | 239 | int ret; |
3947be19 | 240 | |
de0ed36a | 241 | first_page = pfn_to_page(phys_index << PFN_SECTION_SHIFT); |
a16cee10 | 242 | start_pfn = page_to_pfn(first_page); |
de0ed36a | 243 | |
3947be19 DH |
244 | switch (action) { |
245 | case MEM_ONLINE: | |
6056d619 | 246 | if (!pages_correctly_reserved(start_pfn)) |
2bbcb878 MG |
247 | return -EBUSY; |
248 | ||
511c2aba | 249 | ret = online_pages(start_pfn, nr_pages, online_type); |
3947be19 DH |
250 | break; |
251 | case MEM_OFFLINE: | |
a16cee10 | 252 | ret = offline_pages(start_pfn, nr_pages); |
3947be19 DH |
253 | break; |
254 | default: | |
0c2c99b1 NF |
255 | WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: " |
256 | "%ld\n", __func__, phys_index, action, action); | |
3947be19 DH |
257 | ret = -EINVAL; |
258 | } | |
3947be19 DH |
259 | |
260 | return ret; | |
261 | } | |
262 | ||
fa2be40f SJ |
263 | static int memory_block_change_state(struct memory_block *mem, |
264 | unsigned long to_state, unsigned long from_state_req) | |
3947be19 | 265 | { |
de0ed36a | 266 | int ret = 0; |
0c2c99b1 | 267 | |
4960e05e RW |
268 | if (mem->state != from_state_req) |
269 | return -EINVAL; | |
3947be19 | 270 | |
0c2c99b1 NF |
271 | if (to_state == MEM_OFFLINE) |
272 | mem->state = MEM_GOING_OFFLINE; | |
273 | ||
fa2be40f SJ |
274 | ret = memory_block_action(mem->start_section_nr, to_state, |
275 | mem->online_type); | |
276 | ||
b2c064b2 | 277 | mem->state = ret ? from_state_req : to_state; |
fa2be40f | 278 | |
4960e05e RW |
279 | return ret; |
280 | } | |
0c2c99b1 | 281 | |
fa2be40f | 282 | /* The device lock serializes operations on memory_subsys_[online|offline] */ |
4960e05e RW |
283 | static int memory_subsys_online(struct device *dev) |
284 | { | |
285 | struct memory_block *mem = container_of(dev, struct memory_block, dev); | |
286 | int ret; | |
3947be19 | 287 | |
fa2be40f SJ |
288 | if (mem->state == MEM_ONLINE) |
289 | return 0; | |
290 | ||
291 | /* | |
292 | * If we are called from store_mem_state(), online_type will be | |
293 | * set >= 0 Otherwise we were called from the device online | |
294 | * attribute and need to set the online_type. | |
295 | */ | |
296 | if (mem->online_type < 0) | |
297 | mem->online_type = ONLINE_KEEP; | |
4960e05e | 298 | |
fa2be40f SJ |
299 | ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE); |
300 | ||
301 | /* clear online_type */ | |
302 | mem->online_type = -1; | |
4960e05e | 303 | |
4960e05e RW |
304 | return ret; |
305 | } | |
306 | ||
307 | static int memory_subsys_offline(struct device *dev) | |
308 | { | |
309 | struct memory_block *mem = container_of(dev, struct memory_block, dev); | |
4960e05e | 310 | |
fa2be40f SJ |
311 | if (mem->state == MEM_OFFLINE) |
312 | return 0; | |
4960e05e | 313 | |
fa2be40f | 314 | return memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE); |
4960e05e RW |
315 | } |
316 | ||
3947be19 | 317 | static ssize_t |
10fbcf4c KS |
318 | store_mem_state(struct device *dev, |
319 | struct device_attribute *attr, const char *buf, size_t count) | |
3947be19 DH |
320 | { |
321 | struct memory_block *mem; | |
fa2be40f | 322 | int ret, online_type; |
3947be19 | 323 | |
10fbcf4c | 324 | mem = container_of(dev, struct memory_block, dev); |
3947be19 | 325 | |
4960e05e RW |
326 | lock_device_hotplug(); |
327 | ||
fa2be40f SJ |
328 | if (!strncmp(buf, "online_kernel", min_t(int, count, 13))) |
329 | online_type = ONLINE_KERNEL; | |
330 | else if (!strncmp(buf, "online_movable", min_t(int, count, 14))) | |
331 | online_type = ONLINE_MOVABLE; | |
332 | else if (!strncmp(buf, "online", min_t(int, count, 6))) | |
333 | online_type = ONLINE_KEEP; | |
334 | else if (!strncmp(buf, "offline", min_t(int, count, 7))) | |
335 | online_type = -1; | |
336 | else | |
337 | return -EINVAL; | |
338 | ||
339 | switch (online_type) { | |
340 | case ONLINE_KERNEL: | |
341 | case ONLINE_MOVABLE: | |
342 | case ONLINE_KEEP: | |
343 | /* | |
344 | * mem->online_type is not protected so there can be a | |
345 | * race here. However, when racing online, the first | |
346 | * will succeed and the second will just return as the | |
347 | * block will already be online. The online type | |
348 | * could be either one, but that is expected. | |
349 | */ | |
350 | mem->online_type = online_type; | |
351 | ret = device_online(&mem->dev); | |
352 | break; | |
353 | case -1: | |
354 | ret = device_offline(&mem->dev); | |
355 | break; | |
356 | default: | |
357 | ret = -EINVAL; /* should never happen */ | |
4960e05e | 358 | } |
4960e05e RW |
359 | |
360 | unlock_device_hotplug(); | |
0c2c99b1 | 361 | |
3947be19 DH |
362 | if (ret) |
363 | return ret; | |
364 | return count; | |
365 | } | |
366 | ||
367 | /* | |
368 | * phys_device is a bad name for this. What I really want | |
369 | * is a way to differentiate between memory ranges that | |
370 | * are part of physical devices that constitute | |
371 | * a complete removable unit or fru. | |
372 | * i.e. do these ranges belong to the same physical device, | |
373 | * s.t. if I offline all of these sections I can then | |
374 | * remove the physical device? | |
375 | */ | |
10fbcf4c KS |
376 | static ssize_t show_phys_device(struct device *dev, |
377 | struct device_attribute *attr, char *buf) | |
3947be19 DH |
378 | { |
379 | struct memory_block *mem = | |
10fbcf4c | 380 | container_of(dev, struct memory_block, dev); |
3947be19 DH |
381 | return sprintf(buf, "%d\n", mem->phys_device); |
382 | } | |
383 | ||
10fbcf4c KS |
384 | static DEVICE_ATTR(phys_index, 0444, show_mem_start_phys_index, NULL); |
385 | static DEVICE_ATTR(end_phys_index, 0444, show_mem_end_phys_index, NULL); | |
386 | static DEVICE_ATTR(state, 0644, show_mem_state, store_mem_state); | |
387 | static DEVICE_ATTR(phys_device, 0444, show_phys_device, NULL); | |
388 | static DEVICE_ATTR(removable, 0444, show_mem_removable, NULL); | |
3947be19 | 389 | |
3947be19 DH |
390 | /* |
391 | * Block size attribute stuff | |
392 | */ | |
393 | static ssize_t | |
10fbcf4c | 394 | print_block_size(struct device *dev, struct device_attribute *attr, |
8564a6c1 | 395 | char *buf) |
3947be19 | 396 | { |
0c2c99b1 | 397 | return sprintf(buf, "%lx\n", get_memory_block_size()); |
3947be19 DH |
398 | } |
399 | ||
10fbcf4c | 400 | static DEVICE_ATTR(block_size_bytes, 0444, print_block_size, NULL); |
3947be19 | 401 | |
3947be19 DH |
402 | /* |
403 | * Some architectures will have custom drivers to do this, and | |
404 | * will not need to do it from userspace. The fake hot-add code | |
405 | * as well as ppc64 will do all of their discovery in userspace | |
406 | * and will require this interface. | |
407 | */ | |
408 | #ifdef CONFIG_ARCH_MEMORY_PROBE | |
409 | static ssize_t | |
10fbcf4c | 410 | memory_probe_store(struct device *dev, struct device_attribute *attr, |
28812fe1 | 411 | const char *buf, size_t count) |
3947be19 DH |
412 | { |
413 | u64 phys_addr; | |
bc02af93 | 414 | int nid; |
6add7cd6 | 415 | int i, ret; |
61b94fea | 416 | unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block; |
3947be19 DH |
417 | |
418 | phys_addr = simple_strtoull(buf, NULL, 0); | |
419 | ||
61b94fea AB |
420 | if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1)) |
421 | return -EINVAL; | |
422 | ||
6add7cd6 NF |
423 | for (i = 0; i < sections_per_block; i++) { |
424 | nid = memory_add_physaddr_to_nid(phys_addr); | |
425 | ret = add_memory(nid, phys_addr, | |
426 | PAGES_PER_SECTION << PAGE_SHIFT); | |
427 | if (ret) | |
9f0af69b | 428 | goto out; |
6add7cd6 NF |
429 | |
430 | phys_addr += MIN_MEMORY_BLOCK_SIZE; | |
431 | } | |
3947be19 | 432 | |
9f0af69b NK |
433 | ret = count; |
434 | out: | |
435 | return ret; | |
3947be19 | 436 | } |
3947be19 | 437 | |
96b2c0fc | 438 | static DEVICE_ATTR(probe, S_IWUSR, NULL, memory_probe_store); |
3947be19 DH |
439 | #endif |
440 | ||
facb6011 AK |
441 | #ifdef CONFIG_MEMORY_FAILURE |
442 | /* | |
443 | * Support for offlining pages of memory | |
444 | */ | |
445 | ||
446 | /* Soft offline a page */ | |
447 | static ssize_t | |
10fbcf4c KS |
448 | store_soft_offline_page(struct device *dev, |
449 | struct device_attribute *attr, | |
28812fe1 | 450 | const char *buf, size_t count) |
facb6011 AK |
451 | { |
452 | int ret; | |
453 | u64 pfn; | |
454 | if (!capable(CAP_SYS_ADMIN)) | |
455 | return -EPERM; | |
34da5e67 | 456 | if (kstrtoull(buf, 0, &pfn) < 0) |
facb6011 AK |
457 | return -EINVAL; |
458 | pfn >>= PAGE_SHIFT; | |
459 | if (!pfn_valid(pfn)) | |
460 | return -ENXIO; | |
461 | ret = soft_offline_page(pfn_to_page(pfn), 0); | |
462 | return ret == 0 ? count : ret; | |
463 | } | |
464 | ||
465 | /* Forcibly offline a page, including killing processes. */ | |
466 | static ssize_t | |
10fbcf4c KS |
467 | store_hard_offline_page(struct device *dev, |
468 | struct device_attribute *attr, | |
28812fe1 | 469 | const char *buf, size_t count) |
facb6011 AK |
470 | { |
471 | int ret; | |
472 | u64 pfn; | |
473 | if (!capable(CAP_SYS_ADMIN)) | |
474 | return -EPERM; | |
34da5e67 | 475 | if (kstrtoull(buf, 0, &pfn) < 0) |
facb6011 AK |
476 | return -EINVAL; |
477 | pfn >>= PAGE_SHIFT; | |
cd42f4a3 | 478 | ret = memory_failure(pfn, 0, 0); |
facb6011 AK |
479 | return ret ? ret : count; |
480 | } | |
481 | ||
74fef7a8 FB |
482 | static DEVICE_ATTR(soft_offline_page, S_IWUSR, NULL, store_soft_offline_page); |
483 | static DEVICE_ATTR(hard_offline_page, S_IWUSR, NULL, store_hard_offline_page); | |
facb6011 AK |
484 | #endif |
485 | ||
3947be19 DH |
486 | /* |
487 | * Note that phys_device is optional. It is here to allow for | |
488 | * differentiation between which *physical* devices each | |
489 | * section belongs to... | |
490 | */ | |
bc32df00 HC |
491 | int __weak arch_get_memory_phys_device(unsigned long start_pfn) |
492 | { | |
493 | return 0; | |
494 | } | |
3947be19 | 495 | |
10fbcf4c KS |
496 | /* |
497 | * A reference for the returned object is held and the reference for the | |
498 | * hinted object is released. | |
499 | */ | |
98383031 RH |
500 | struct memory_block *find_memory_block_hinted(struct mem_section *section, |
501 | struct memory_block *hint) | |
3947be19 | 502 | { |
0c2c99b1 | 503 | int block_id = base_memory_block_id(__section_nr(section)); |
10fbcf4c KS |
504 | struct device *hintdev = hint ? &hint->dev : NULL; |
505 | struct device *dev; | |
3947be19 | 506 | |
10fbcf4c KS |
507 | dev = subsys_find_device_by_id(&memory_subsys, block_id, hintdev); |
508 | if (hint) | |
509 | put_device(&hint->dev); | |
510 | if (!dev) | |
3947be19 | 511 | return NULL; |
10fbcf4c | 512 | return container_of(dev, struct memory_block, dev); |
3947be19 DH |
513 | } |
514 | ||
98383031 RH |
515 | /* |
516 | * For now, we have a linear search to go find the appropriate | |
517 | * memory_block corresponding to a particular phys_index. If | |
518 | * this gets to be a real problem, we can always use a radix | |
519 | * tree or something here. | |
520 | * | |
10fbcf4c | 521 | * This could be made generic for all device subsystems. |
98383031 RH |
522 | */ |
523 | struct memory_block *find_memory_block(struct mem_section *section) | |
524 | { | |
525 | return find_memory_block_hinted(section, NULL); | |
526 | } | |
527 | ||
96b2c0fc NF |
528 | static struct attribute *memory_memblk_attrs[] = { |
529 | &dev_attr_phys_index.attr, | |
530 | &dev_attr_end_phys_index.attr, | |
531 | &dev_attr_state.attr, | |
532 | &dev_attr_phys_device.attr, | |
533 | &dev_attr_removable.attr, | |
534 | NULL | |
535 | }; | |
536 | ||
537 | static struct attribute_group memory_memblk_attr_group = { | |
538 | .attrs = memory_memblk_attrs, | |
539 | }; | |
540 | ||
541 | static const struct attribute_group *memory_memblk_attr_groups[] = { | |
542 | &memory_memblk_attr_group, | |
543 | NULL, | |
544 | }; | |
545 | ||
546 | /* | |
547 | * register_memory - Setup a sysfs device for a memory block | |
548 | */ | |
549 | static | |
550 | int register_memory(struct memory_block *memory) | |
551 | { | |
96b2c0fc NF |
552 | memory->dev.bus = &memory_subsys; |
553 | memory->dev.id = memory->start_section_nr / sections_per_block; | |
554 | memory->dev.release = memory_block_release; | |
555 | memory->dev.groups = memory_memblk_attr_groups; | |
f991fae5 | 556 | memory->dev.offline = memory->state == MEM_OFFLINE; |
96b2c0fc | 557 | |
879f1bec | 558 | return device_register(&memory->dev); |
96b2c0fc NF |
559 | } |
560 | ||
0c2c99b1 NF |
561 | static int init_memory_block(struct memory_block **memory, |
562 | struct mem_section *section, unsigned long state) | |
e4619c85 | 563 | { |
0c2c99b1 | 564 | struct memory_block *mem; |
e4619c85 | 565 | unsigned long start_pfn; |
0c2c99b1 | 566 | int scn_nr; |
e4619c85 NF |
567 | int ret = 0; |
568 | ||
0c2c99b1 | 569 | mem = kzalloc(sizeof(*mem), GFP_KERNEL); |
e4619c85 NF |
570 | if (!mem) |
571 | return -ENOMEM; | |
572 | ||
0c2c99b1 | 573 | scn_nr = __section_nr(section); |
d3360164 NF |
574 | mem->start_section_nr = |
575 | base_memory_block_id(scn_nr) * sections_per_block; | |
576 | mem->end_section_nr = mem->start_section_nr + sections_per_block - 1; | |
e4619c85 | 577 | mem->state = state; |
07681215 | 578 | mem->section_count++; |
d3360164 | 579 | start_pfn = section_nr_to_pfn(mem->start_section_nr); |
e4619c85 NF |
580 | mem->phys_device = arch_get_memory_phys_device(start_pfn); |
581 | ||
0c2c99b1 | 582 | ret = register_memory(mem); |
0c2c99b1 NF |
583 | |
584 | *memory = mem; | |
585 | return ret; | |
586 | } | |
587 | ||
cb5e39b8 | 588 | static int add_memory_block(int base_section_nr) |
0c2c99b1 | 589 | { |
cb5e39b8 SJ |
590 | struct memory_block *mem; |
591 | int i, ret, section_count = 0, section_nr; | |
321bf4ed | 592 | |
cb5e39b8 SJ |
593 | for (i = base_section_nr; |
594 | (i < base_section_nr + sections_per_block) && i < NR_MEM_SECTIONS; | |
595 | i++) { | |
596 | if (!present_section_nr(i)) | |
597 | continue; | |
598 | if (section_count == 0) | |
599 | section_nr = i; | |
600 | section_count++; | |
e4619c85 NF |
601 | } |
602 | ||
cb5e39b8 SJ |
603 | if (section_count == 0) |
604 | return 0; | |
605 | ret = init_memory_block(&mem, __nr_to_section(section_nr), MEM_ONLINE); | |
606 | if (ret) | |
607 | return ret; | |
608 | mem->section_count = section_count; | |
609 | return 0; | |
e4619c85 NF |
610 | } |
611 | ||
cb5e39b8 | 612 | |
4edd7cef DR |
613 | /* |
614 | * need an interface for the VM to add new memory regions, | |
615 | * but without onlining it. | |
616 | */ | |
617 | int register_new_memory(int nid, struct mem_section *section) | |
618 | { | |
d7f80530 SJ |
619 | int ret = 0; |
620 | struct memory_block *mem; | |
b1eaef3d SJ |
621 | |
622 | mutex_lock(&mem_sysfs_mutex); | |
b1eaef3d | 623 | |
d7f80530 SJ |
624 | mem = find_memory_block(section); |
625 | if (mem) { | |
626 | mem->section_count++; | |
627 | put_device(&mem->dev); | |
628 | } else { | |
629 | ret = init_memory_block(&mem, section, MEM_OFFLINE); | |
630 | if (ret) | |
631 | goto out; | |
632 | } | |
633 | ||
634 | if (mem->section_count == sections_per_block) | |
635 | ret = register_mem_sect_under_node(mem, nid); | |
636 | out: | |
637 | mutex_unlock(&mem_sysfs_mutex); | |
b1eaef3d | 638 | return ret; |
4edd7cef DR |
639 | } |
640 | ||
641 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
642 | static void | |
643 | unregister_memory(struct memory_block *memory) | |
644 | { | |
645 | BUG_ON(memory->dev.bus != &memory_subsys); | |
646 | ||
647 | /* drop the ref. we got in remove_memory_block() */ | |
df2b717c | 648 | put_device(&memory->dev); |
4edd7cef DR |
649 | device_unregister(&memory->dev); |
650 | } | |
651 | ||
652 | static int remove_memory_block(unsigned long node_id, | |
653 | struct mem_section *section, int phys_device) | |
3947be19 DH |
654 | { |
655 | struct memory_block *mem; | |
656 | ||
2938ffbd | 657 | mutex_lock(&mem_sysfs_mutex); |
3947be19 | 658 | mem = find_memory_block(section); |
d3360164 | 659 | unregister_mem_sect_under_nodes(mem, __section_nr(section)); |
07681215 NF |
660 | |
661 | mem->section_count--; | |
96b2c0fc | 662 | if (mem->section_count == 0) |
0c2c99b1 | 663 | unregister_memory(mem); |
96b2c0fc | 664 | else |
df2b717c | 665 | put_device(&mem->dev); |
3947be19 | 666 | |
2938ffbd | 667 | mutex_unlock(&mem_sysfs_mutex); |
3947be19 DH |
668 | return 0; |
669 | } | |
670 | ||
3947be19 DH |
671 | int unregister_memory_section(struct mem_section *section) |
672 | { | |
540557b9 | 673 | if (!present_section(section)) |
3947be19 DH |
674 | return -EINVAL; |
675 | ||
676 | return remove_memory_block(0, section, 0); | |
677 | } | |
4edd7cef | 678 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
3947be19 | 679 | |
6677e3ea YI |
680 | /* return true if the memory block is offlined, otherwise, return false */ |
681 | bool is_memblock_offlined(struct memory_block *mem) | |
682 | { | |
683 | return mem->state == MEM_OFFLINE; | |
684 | } | |
685 | ||
96b2c0fc NF |
686 | static struct attribute *memory_root_attrs[] = { |
687 | #ifdef CONFIG_ARCH_MEMORY_PROBE | |
688 | &dev_attr_probe.attr, | |
689 | #endif | |
690 | ||
691 | #ifdef CONFIG_MEMORY_FAILURE | |
692 | &dev_attr_soft_offline_page.attr, | |
693 | &dev_attr_hard_offline_page.attr, | |
694 | #endif | |
695 | ||
696 | &dev_attr_block_size_bytes.attr, | |
697 | NULL | |
698 | }; | |
699 | ||
700 | static struct attribute_group memory_root_attr_group = { | |
701 | .attrs = memory_root_attrs, | |
702 | }; | |
703 | ||
704 | static const struct attribute_group *memory_root_attr_groups[] = { | |
705 | &memory_root_attr_group, | |
706 | NULL, | |
707 | }; | |
708 | ||
3947be19 DH |
709 | /* |
710 | * Initialize the sysfs support for memory devices... | |
711 | */ | |
712 | int __init memory_dev_init(void) | |
713 | { | |
714 | unsigned int i; | |
715 | int ret; | |
28ec24e2 | 716 | int err; |
0c2c99b1 | 717 | unsigned long block_sz; |
3947be19 | 718 | |
96b2c0fc | 719 | ret = subsys_system_register(&memory_subsys, memory_root_attr_groups); |
28ec24e2 AM |
720 | if (ret) |
721 | goto out; | |
3947be19 | 722 | |
0c2c99b1 NF |
723 | block_sz = get_memory_block_size(); |
724 | sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE; | |
725 | ||
3947be19 DH |
726 | /* |
727 | * Create entries for memory sections that were found | |
728 | * during boot and have been initialized | |
729 | */ | |
b1eaef3d | 730 | mutex_lock(&mem_sysfs_mutex); |
cb5e39b8 SJ |
731 | for (i = 0; i < NR_MEM_SECTIONS; i += sections_per_block) { |
732 | err = add_memory_block(i); | |
28ec24e2 AM |
733 | if (!ret) |
734 | ret = err; | |
3947be19 | 735 | } |
b1eaef3d | 736 | mutex_unlock(&mem_sysfs_mutex); |
3947be19 | 737 | |
28ec24e2 AM |
738 | out: |
739 | if (ret) | |
2b3a302a | 740 | printk(KERN_ERR "%s() failed: %d\n", __func__, ret); |
3947be19 DH |
741 | return ret; |
742 | } |