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> |
7c0f6ba6 | 26 | #include <linux/uaccess.h> |
3947be19 | 27 | |
2938ffbd NF |
28 | static DEFINE_MUTEX(mem_sysfs_mutex); |
29 | ||
3947be19 | 30 | #define MEMORY_CLASS_NAME "memory" |
0c2c99b1 | 31 | |
7315f0cc GZ |
32 | #define to_memory_block(dev) container_of(dev, struct memory_block, dev) |
33 | ||
0c2c99b1 NF |
34 | static int sections_per_block; |
35 | ||
36 | static inline int base_memory_block_id(int section_nr) | |
37 | { | |
38 | return section_nr / sections_per_block; | |
39 | } | |
3947be19 | 40 | |
4960e05e RW |
41 | static int memory_subsys_online(struct device *dev); |
42 | static int memory_subsys_offline(struct device *dev); | |
43 | ||
10fbcf4c | 44 | static struct bus_type memory_subsys = { |
af5ca3f4 | 45 | .name = MEMORY_CLASS_NAME, |
10fbcf4c | 46 | .dev_name = MEMORY_CLASS_NAME, |
4960e05e RW |
47 | .online = memory_subsys_online, |
48 | .offline = memory_subsys_offline, | |
3947be19 DH |
49 | }; |
50 | ||
e041c683 | 51 | static BLOCKING_NOTIFIER_HEAD(memory_chain); |
3947be19 | 52 | |
98a38ebd | 53 | int register_memory_notifier(struct notifier_block *nb) |
3947be19 | 54 | { |
2aeebca2 | 55 | return blocking_notifier_chain_register(&memory_chain, nb); |
3947be19 | 56 | } |
3c82c30c | 57 | EXPORT_SYMBOL(register_memory_notifier); |
3947be19 | 58 | |
98a38ebd | 59 | void unregister_memory_notifier(struct notifier_block *nb) |
3947be19 | 60 | { |
2aeebca2 | 61 | blocking_notifier_chain_unregister(&memory_chain, nb); |
3947be19 | 62 | } |
3c82c30c | 63 | EXPORT_SYMBOL(unregister_memory_notifier); |
3947be19 | 64 | |
925cc71e RJ |
65 | static ATOMIC_NOTIFIER_HEAD(memory_isolate_chain); |
66 | ||
67 | int register_memory_isolate_notifier(struct notifier_block *nb) | |
68 | { | |
69 | return atomic_notifier_chain_register(&memory_isolate_chain, nb); | |
70 | } | |
71 | EXPORT_SYMBOL(register_memory_isolate_notifier); | |
72 | ||
73 | void unregister_memory_isolate_notifier(struct notifier_block *nb) | |
74 | { | |
75 | atomic_notifier_chain_unregister(&memory_isolate_chain, nb); | |
76 | } | |
77 | EXPORT_SYMBOL(unregister_memory_isolate_notifier); | |
78 | ||
fa7194eb YI |
79 | static void memory_block_release(struct device *dev) |
80 | { | |
7315f0cc | 81 | struct memory_block *mem = to_memory_block(dev); |
fa7194eb YI |
82 | |
83 | kfree(mem); | |
84 | } | |
85 | ||
0c2c99b1 NF |
86 | unsigned long __weak memory_block_size_bytes(void) |
87 | { | |
88 | return MIN_MEMORY_BLOCK_SIZE; | |
89 | } | |
90 | ||
91 | static unsigned long get_memory_block_size(void) | |
92 | { | |
93 | unsigned long block_sz; | |
94 | ||
95 | block_sz = memory_block_size_bytes(); | |
96 | ||
97 | /* Validate blk_sz is a power of 2 and not less than section size */ | |
98 | if ((block_sz & (block_sz - 1)) || (block_sz < MIN_MEMORY_BLOCK_SIZE)) { | |
99 | WARN_ON(1); | |
100 | block_sz = MIN_MEMORY_BLOCK_SIZE; | |
101 | } | |
102 | ||
103 | return block_sz; | |
104 | } | |
105 | ||
3947be19 DH |
106 | /* |
107 | * use this as the physical section index that this memsection | |
108 | * uses. | |
109 | */ | |
110 | ||
10fbcf4c KS |
111 | static ssize_t show_mem_start_phys_index(struct device *dev, |
112 | struct device_attribute *attr, char *buf) | |
3947be19 | 113 | { |
7315f0cc | 114 | struct memory_block *mem = to_memory_block(dev); |
d3360164 NF |
115 | unsigned long phys_index; |
116 | ||
117 | phys_index = mem->start_section_nr / sections_per_block; | |
118 | return sprintf(buf, "%08lx\n", phys_index); | |
119 | } | |
120 | ||
5c755e9f BP |
121 | /* |
122 | * Show whether the section of memory is likely to be hot-removable | |
123 | */ | |
10fbcf4c KS |
124 | static ssize_t show_mem_removable(struct device *dev, |
125 | struct device_attribute *attr, char *buf) | |
5c755e9f | 126 | { |
0c2c99b1 NF |
127 | unsigned long i, pfn; |
128 | int ret = 1; | |
7315f0cc | 129 | struct memory_block *mem = to_memory_block(dev); |
5c755e9f | 130 | |
8b0662f2 MH |
131 | if (mem->state != MEM_ONLINE) |
132 | goto out; | |
133 | ||
0c2c99b1 | 134 | for (i = 0; i < sections_per_block; i++) { |
21ea9f5a RA |
135 | if (!present_section_nr(mem->start_section_nr + i)) |
136 | continue; | |
d3360164 | 137 | pfn = section_nr_to_pfn(mem->start_section_nr + i); |
0c2c99b1 NF |
138 | ret &= is_mem_section_removable(pfn, PAGES_PER_SECTION); |
139 | } | |
140 | ||
8b0662f2 | 141 | out: |
5c755e9f BP |
142 | return sprintf(buf, "%d\n", ret); |
143 | } | |
144 | ||
3947be19 DH |
145 | /* |
146 | * online, offline, going offline, etc. | |
147 | */ | |
10fbcf4c KS |
148 | static ssize_t show_mem_state(struct device *dev, |
149 | struct device_attribute *attr, char *buf) | |
3947be19 | 150 | { |
7315f0cc | 151 | struct memory_block *mem = to_memory_block(dev); |
3947be19 DH |
152 | ssize_t len = 0; |
153 | ||
154 | /* | |
155 | * We can probably put these states in a nice little array | |
156 | * so that they're not open-coded | |
157 | */ | |
158 | switch (mem->state) { | |
3d3af6af IC |
159 | case MEM_ONLINE: |
160 | len = sprintf(buf, "online\n"); | |
161 | break; | |
162 | case MEM_OFFLINE: | |
163 | len = sprintf(buf, "offline\n"); | |
164 | break; | |
165 | case MEM_GOING_OFFLINE: | |
166 | len = sprintf(buf, "going-offline\n"); | |
167 | break; | |
168 | default: | |
169 | len = sprintf(buf, "ERROR-UNKNOWN-%ld\n", | |
170 | mem->state); | |
171 | WARN_ON(1); | |
172 | break; | |
3947be19 DH |
173 | } |
174 | ||
175 | return len; | |
176 | } | |
177 | ||
7b78d335 | 178 | int memory_notify(unsigned long val, void *v) |
3947be19 | 179 | { |
e041c683 | 180 | return blocking_notifier_call_chain(&memory_chain, val, v); |
3947be19 DH |
181 | } |
182 | ||
925cc71e RJ |
183 | int memory_isolate_notify(unsigned long val, void *v) |
184 | { | |
185 | return atomic_notifier_call_chain(&memory_isolate_chain, val, v); | |
186 | } | |
187 | ||
2bbcb878 MG |
188 | /* |
189 | * The probe routines leave the pages reserved, just as the bootmem code does. | |
190 | * Make sure they're still that way. | |
191 | */ | |
6056d619 | 192 | static bool pages_correctly_reserved(unsigned long start_pfn) |
2bbcb878 MG |
193 | { |
194 | int i, j; | |
195 | struct page *page; | |
196 | unsigned long pfn = start_pfn; | |
197 | ||
198 | /* | |
199 | * memmap between sections is not contiguous except with | |
200 | * SPARSEMEM_VMEMMAP. We lookup the page once per section | |
201 | * and assume memmap is contiguous within each section | |
202 | */ | |
203 | for (i = 0; i < sections_per_block; i++, pfn += PAGES_PER_SECTION) { | |
204 | if (WARN_ON_ONCE(!pfn_valid(pfn))) | |
205 | return false; | |
206 | page = pfn_to_page(pfn); | |
207 | ||
208 | for (j = 0; j < PAGES_PER_SECTION; j++) { | |
209 | if (PageReserved(page + j)) | |
210 | continue; | |
211 | ||
212 | printk(KERN_WARNING "section number %ld page number %d " | |
213 | "not reserved, was it already online?\n", | |
214 | pfn_to_section_nr(pfn), j); | |
215 | ||
216 | return false; | |
217 | } | |
218 | } | |
219 | ||
220 | return true; | |
221 | } | |
222 | ||
3947be19 DH |
223 | /* |
224 | * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is | |
225 | * OK to have direct references to sparsemem variables in here. | |
30467e0b | 226 | * Must already be protected by mem_hotplug_begin(). |
3947be19 DH |
227 | */ |
228 | static int | |
511c2aba | 229 | memory_block_action(unsigned long phys_index, unsigned long action, int online_type) |
3947be19 | 230 | { |
a16cee10 | 231 | unsigned long start_pfn; |
5409d2cd | 232 | unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block; |
3947be19 | 233 | int ret; |
3947be19 | 234 | |
19c07d5e | 235 | start_pfn = section_nr_to_pfn(phys_index); |
de0ed36a | 236 | |
3947be19 | 237 | switch (action) { |
3d3af6af IC |
238 | case MEM_ONLINE: |
239 | if (!pages_correctly_reserved(start_pfn)) | |
240 | return -EBUSY; | |
241 | ||
242 | ret = online_pages(start_pfn, nr_pages, online_type); | |
243 | break; | |
244 | case MEM_OFFLINE: | |
245 | ret = offline_pages(start_pfn, nr_pages); | |
246 | break; | |
247 | default: | |
248 | WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: " | |
249 | "%ld\n", __func__, phys_index, action, action); | |
250 | ret = -EINVAL; | |
3947be19 | 251 | } |
3947be19 DH |
252 | |
253 | return ret; | |
254 | } | |
255 | ||
dc18d706 | 256 | static int memory_block_change_state(struct memory_block *mem, |
fa2be40f | 257 | unsigned long to_state, unsigned long from_state_req) |
3947be19 | 258 | { |
de0ed36a | 259 | int ret = 0; |
0c2c99b1 | 260 | |
4960e05e RW |
261 | if (mem->state != from_state_req) |
262 | return -EINVAL; | |
3947be19 | 263 | |
0c2c99b1 NF |
264 | if (to_state == MEM_OFFLINE) |
265 | mem->state = MEM_GOING_OFFLINE; | |
266 | ||
fa2be40f SJ |
267 | ret = memory_block_action(mem->start_section_nr, to_state, |
268 | mem->online_type); | |
269 | ||
b2c064b2 | 270 | mem->state = ret ? from_state_req : to_state; |
fa2be40f | 271 | |
4960e05e RW |
272 | return ret; |
273 | } | |
0c2c99b1 | 274 | |
fa2be40f | 275 | /* The device lock serializes operations on memory_subsys_[online|offline] */ |
4960e05e RW |
276 | static int memory_subsys_online(struct device *dev) |
277 | { | |
7315f0cc | 278 | struct memory_block *mem = to_memory_block(dev); |
4960e05e | 279 | int ret; |
3947be19 | 280 | |
fa2be40f SJ |
281 | if (mem->state == MEM_ONLINE) |
282 | return 0; | |
4960e05e | 283 | |
fa2be40f SJ |
284 | /* |
285 | * If we are called from store_mem_state(), online_type will be | |
286 | * set >= 0 Otherwise we were called from the device online | |
287 | * attribute and need to set the online_type. | |
288 | */ | |
289 | if (mem->online_type < 0) | |
4f7c6b49 | 290 | mem->online_type = MMOP_ONLINE_KEEP; |
4960e05e | 291 | |
30467e0b | 292 | /* Already under protection of mem_hotplug_begin() */ |
fa2be40f | 293 | ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE); |
4960e05e | 294 | |
fa2be40f SJ |
295 | /* clear online_type */ |
296 | mem->online_type = -1; | |
4960e05e | 297 | |
4960e05e RW |
298 | return ret; |
299 | } | |
300 | ||
4960e05e | 301 | static int memory_subsys_offline(struct device *dev) |
e90bdb7f | 302 | { |
7315f0cc | 303 | struct memory_block *mem = to_memory_block(dev); |
e90bdb7f | 304 | |
fa2be40f SJ |
305 | if (mem->state == MEM_OFFLINE) |
306 | return 0; | |
e90bdb7f | 307 | |
26bbe7ef SJ |
308 | /* Can't offline block with non-present sections */ |
309 | if (mem->section_count != sections_per_block) | |
310 | return -EINVAL; | |
311 | ||
fa2be40f | 312 | return memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE); |
e90bdb7f | 313 | } |
4960e05e | 314 | |
3947be19 | 315 | static ssize_t |
10fbcf4c KS |
316 | store_mem_state(struct device *dev, |
317 | struct device_attribute *attr, const char *buf, size_t count) | |
3947be19 | 318 | { |
7315f0cc | 319 | struct memory_block *mem = to_memory_block(dev); |
fa2be40f | 320 | int ret, online_type; |
3947be19 | 321 | |
5e33bc41 RW |
322 | ret = lock_device_hotplug_sysfs(); |
323 | if (ret) | |
324 | return ret; | |
4960e05e | 325 | |
1f6a6cc8 | 326 | if (sysfs_streq(buf, "online_kernel")) |
4f7c6b49 | 327 | online_type = MMOP_ONLINE_KERNEL; |
1f6a6cc8 | 328 | else if (sysfs_streq(buf, "online_movable")) |
4f7c6b49 | 329 | online_type = MMOP_ONLINE_MOVABLE; |
1f6a6cc8 | 330 | else if (sysfs_streq(buf, "online")) |
4f7c6b49 | 331 | online_type = MMOP_ONLINE_KEEP; |
1f6a6cc8 | 332 | else if (sysfs_streq(buf, "offline")) |
4f7c6b49 | 333 | online_type = MMOP_OFFLINE; |
a37f8630 YI |
334 | else { |
335 | ret = -EINVAL; | |
336 | goto err; | |
337 | } | |
fa2be40f | 338 | |
30467e0b DR |
339 | /* |
340 | * Memory hotplug needs to hold mem_hotplug_begin() for probe to find | |
341 | * the correct memory block to online before doing device_online(dev), | |
342 | * which will take dev->mutex. Take the lock early to prevent an | |
343 | * inversion, memory_subsys_online() callbacks will be implemented by | |
344 | * assuming it's already protected. | |
345 | */ | |
346 | mem_hotplug_begin(); | |
347 | ||
fa2be40f | 348 | switch (online_type) { |
4f7c6b49 TC |
349 | case MMOP_ONLINE_KERNEL: |
350 | case MMOP_ONLINE_MOVABLE: | |
351 | case MMOP_ONLINE_KEEP: | |
fa2be40f SJ |
352 | mem->online_type = online_type; |
353 | ret = device_online(&mem->dev); | |
354 | break; | |
4f7c6b49 | 355 | case MMOP_OFFLINE: |
fa2be40f SJ |
356 | ret = device_offline(&mem->dev); |
357 | break; | |
358 | default: | |
359 | ret = -EINVAL; /* should never happen */ | |
4960e05e | 360 | } |
4960e05e | 361 | |
30467e0b | 362 | mem_hotplug_done(); |
a37f8630 | 363 | err: |
4960e05e | 364 | unlock_device_hotplug(); |
0c2c99b1 | 365 | |
d66ba15b | 366 | if (ret < 0) |
3947be19 | 367 | return ret; |
d66ba15b RA |
368 | if (ret) |
369 | return -EINVAL; | |
370 | ||
3947be19 DH |
371 | return count; |
372 | } | |
373 | ||
374 | /* | |
375 | * phys_device is a bad name for this. What I really want | |
376 | * is a way to differentiate between memory ranges that | |
377 | * are part of physical devices that constitute | |
378 | * a complete removable unit or fru. | |
379 | * i.e. do these ranges belong to the same physical device, | |
380 | * s.t. if I offline all of these sections I can then | |
381 | * remove the physical device? | |
382 | */ | |
10fbcf4c KS |
383 | static ssize_t show_phys_device(struct device *dev, |
384 | struct device_attribute *attr, char *buf) | |
3947be19 | 385 | { |
7315f0cc | 386 | struct memory_block *mem = to_memory_block(dev); |
3947be19 DH |
387 | return sprintf(buf, "%d\n", mem->phys_device); |
388 | } | |
389 | ||
ed2f2400 | 390 | #ifdef CONFIG_MEMORY_HOTREMOVE |
e5e68930 MH |
391 | static void print_allowed_zone(char *buf, int nid, unsigned long start_pfn, |
392 | unsigned long nr_pages, int online_type, | |
393 | struct zone *default_zone) | |
394 | { | |
395 | struct zone *zone; | |
396 | ||
e5e68930 MH |
397 | zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages); |
398 | if (zone != default_zone) { | |
399 | strcat(buf, " "); | |
400 | strcat(buf, zone->name); | |
401 | } | |
402 | } | |
403 | ||
ed2f2400 ZZ |
404 | static ssize_t show_valid_zones(struct device *dev, |
405 | struct device_attribute *attr, char *buf) | |
406 | { | |
407 | struct memory_block *mem = to_memory_block(dev); | |
f1dd2cd1 | 408 | unsigned long start_pfn = section_nr_to_pfn(mem->start_section_nr); |
ed2f2400 | 409 | unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block; |
f1dd2cd1 | 410 | unsigned long valid_start_pfn, valid_end_pfn; |
e5e68930 | 411 | struct zone *default_zone; |
f1dd2cd1 | 412 | int nid; |
ed2f2400 | 413 | |
f1dd2cd1 MH |
414 | /* |
415 | * The block contains more than one zone can not be offlined. | |
416 | * This can happen e.g. for ZONE_DMA and ZONE_DMA32 | |
417 | */ | |
418 | if (!test_pages_in_a_zone(start_pfn, start_pfn + nr_pages, &valid_start_pfn, &valid_end_pfn)) | |
ed2f2400 ZZ |
419 | return sprintf(buf, "none\n"); |
420 | ||
f1dd2cd1 MH |
421 | start_pfn = valid_start_pfn; |
422 | nr_pages = valid_end_pfn - start_pfn; | |
ed2f2400 | 423 | |
f1dd2cd1 MH |
424 | /* |
425 | * Check the existing zone. Make sure that we do that only on the | |
426 | * online nodes otherwise the page_zone is not reliable | |
427 | */ | |
428 | if (mem->state == MEM_ONLINE) { | |
429 | strcat(buf, page_zone(pfn_to_page(start_pfn))->name); | |
430 | goto out; | |
ed2f2400 ZZ |
431 | } |
432 | ||
f1dd2cd1 | 433 | nid = pfn_to_nid(start_pfn); |
e5e68930 MH |
434 | default_zone = zone_for_pfn_range(MMOP_ONLINE_KEEP, nid, start_pfn, nr_pages); |
435 | strcat(buf, default_zone->name); | |
ed2f2400 | 436 | |
e5e68930 MH |
437 | print_allowed_zone(buf, nid, start_pfn, nr_pages, MMOP_ONLINE_KERNEL, |
438 | default_zone); | |
439 | print_allowed_zone(buf, nid, start_pfn, nr_pages, MMOP_ONLINE_MOVABLE, | |
440 | default_zone); | |
f1dd2cd1 | 441 | out: |
a371d9f1 RA |
442 | strcat(buf, "\n"); |
443 | ||
444 | return strlen(buf); | |
ed2f2400 ZZ |
445 | } |
446 | static DEVICE_ATTR(valid_zones, 0444, show_valid_zones, NULL); | |
447 | #endif | |
448 | ||
10fbcf4c | 449 | static DEVICE_ATTR(phys_index, 0444, show_mem_start_phys_index, NULL); |
10fbcf4c KS |
450 | static DEVICE_ATTR(state, 0644, show_mem_state, store_mem_state); |
451 | static DEVICE_ATTR(phys_device, 0444, show_phys_device, NULL); | |
452 | static DEVICE_ATTR(removable, 0444, show_mem_removable, NULL); | |
3947be19 | 453 | |
3947be19 DH |
454 | /* |
455 | * Block size attribute stuff | |
456 | */ | |
457 | static ssize_t | |
10fbcf4c | 458 | print_block_size(struct device *dev, struct device_attribute *attr, |
8564a6c1 | 459 | char *buf) |
3947be19 | 460 | { |
0c2c99b1 | 461 | return sprintf(buf, "%lx\n", get_memory_block_size()); |
3947be19 DH |
462 | } |
463 | ||
10fbcf4c | 464 | static DEVICE_ATTR(block_size_bytes, 0444, print_block_size, NULL); |
3947be19 | 465 | |
31bc3858 VK |
466 | /* |
467 | * Memory auto online policy. | |
468 | */ | |
469 | ||
470 | static ssize_t | |
471 | show_auto_online_blocks(struct device *dev, struct device_attribute *attr, | |
472 | char *buf) | |
473 | { | |
474 | if (memhp_auto_online) | |
475 | return sprintf(buf, "online\n"); | |
476 | else | |
477 | return sprintf(buf, "offline\n"); | |
478 | } | |
479 | ||
480 | static ssize_t | |
481 | store_auto_online_blocks(struct device *dev, struct device_attribute *attr, | |
482 | const char *buf, size_t count) | |
483 | { | |
484 | if (sysfs_streq(buf, "online")) | |
485 | memhp_auto_online = true; | |
486 | else if (sysfs_streq(buf, "offline")) | |
487 | memhp_auto_online = false; | |
488 | else | |
489 | return -EINVAL; | |
490 | ||
491 | return count; | |
492 | } | |
493 | ||
494 | static DEVICE_ATTR(auto_online_blocks, 0644, show_auto_online_blocks, | |
495 | store_auto_online_blocks); | |
496 | ||
3947be19 DH |
497 | /* |
498 | * Some architectures will have custom drivers to do this, and | |
499 | * will not need to do it from userspace. The fake hot-add code | |
500 | * as well as ppc64 will do all of their discovery in userspace | |
501 | * and will require this interface. | |
502 | */ | |
503 | #ifdef CONFIG_ARCH_MEMORY_PROBE | |
504 | static ssize_t | |
10fbcf4c | 505 | memory_probe_store(struct device *dev, struct device_attribute *attr, |
28812fe1 | 506 | const char *buf, size_t count) |
3947be19 DH |
507 | { |
508 | u64 phys_addr; | |
cb5490a5 | 509 | int nid, ret; |
61b94fea | 510 | unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block; |
3947be19 | 511 | |
b69deb2b ZZ |
512 | ret = kstrtoull(buf, 0, &phys_addr); |
513 | if (ret) | |
514 | return ret; | |
3947be19 | 515 | |
61b94fea AB |
516 | if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1)) |
517 | return -EINVAL; | |
518 | ||
cb5490a5 JA |
519 | nid = memory_add_physaddr_to_nid(phys_addr); |
520 | ret = add_memory(nid, phys_addr, | |
521 | MIN_MEMORY_BLOCK_SIZE * sections_per_block); | |
6add7cd6 | 522 | |
cb5490a5 JA |
523 | if (ret) |
524 | goto out; | |
3947be19 | 525 | |
9f0af69b NK |
526 | ret = count; |
527 | out: | |
528 | return ret; | |
3947be19 | 529 | } |
3947be19 | 530 | |
96b2c0fc | 531 | static DEVICE_ATTR(probe, S_IWUSR, NULL, memory_probe_store); |
3947be19 DH |
532 | #endif |
533 | ||
facb6011 AK |
534 | #ifdef CONFIG_MEMORY_FAILURE |
535 | /* | |
536 | * Support for offlining pages of memory | |
537 | */ | |
538 | ||
539 | /* Soft offline a page */ | |
540 | static ssize_t | |
10fbcf4c KS |
541 | store_soft_offline_page(struct device *dev, |
542 | struct device_attribute *attr, | |
28812fe1 | 543 | const char *buf, size_t count) |
facb6011 AK |
544 | { |
545 | int ret; | |
546 | u64 pfn; | |
547 | if (!capable(CAP_SYS_ADMIN)) | |
548 | return -EPERM; | |
34da5e67 | 549 | if (kstrtoull(buf, 0, &pfn) < 0) |
facb6011 AK |
550 | return -EINVAL; |
551 | pfn >>= PAGE_SHIFT; | |
552 | if (!pfn_valid(pfn)) | |
553 | return -ENXIO; | |
554 | ret = soft_offline_page(pfn_to_page(pfn), 0); | |
555 | return ret == 0 ? count : ret; | |
556 | } | |
557 | ||
558 | /* Forcibly offline a page, including killing processes. */ | |
559 | static ssize_t | |
10fbcf4c KS |
560 | store_hard_offline_page(struct device *dev, |
561 | struct device_attribute *attr, | |
28812fe1 | 562 | const char *buf, size_t count) |
facb6011 AK |
563 | { |
564 | int ret; | |
565 | u64 pfn; | |
566 | if (!capable(CAP_SYS_ADMIN)) | |
567 | return -EPERM; | |
34da5e67 | 568 | if (kstrtoull(buf, 0, &pfn) < 0) |
facb6011 AK |
569 | return -EINVAL; |
570 | pfn >>= PAGE_SHIFT; | |
cd42f4a3 | 571 | ret = memory_failure(pfn, 0, 0); |
facb6011 AK |
572 | return ret ? ret : count; |
573 | } | |
574 | ||
74fef7a8 FB |
575 | static DEVICE_ATTR(soft_offline_page, S_IWUSR, NULL, store_soft_offline_page); |
576 | static DEVICE_ATTR(hard_offline_page, S_IWUSR, NULL, store_hard_offline_page); | |
facb6011 AK |
577 | #endif |
578 | ||
3947be19 DH |
579 | /* |
580 | * Note that phys_device is optional. It is here to allow for | |
581 | * differentiation between which *physical* devices each | |
582 | * section belongs to... | |
583 | */ | |
bc32df00 HC |
584 | int __weak arch_get_memory_phys_device(unsigned long start_pfn) |
585 | { | |
586 | return 0; | |
587 | } | |
3947be19 | 588 | |
10fbcf4c KS |
589 | /* |
590 | * A reference for the returned object is held and the reference for the | |
591 | * hinted object is released. | |
592 | */ | |
98383031 RH |
593 | struct memory_block *find_memory_block_hinted(struct mem_section *section, |
594 | struct memory_block *hint) | |
3947be19 | 595 | { |
0c2c99b1 | 596 | int block_id = base_memory_block_id(__section_nr(section)); |
10fbcf4c KS |
597 | struct device *hintdev = hint ? &hint->dev : NULL; |
598 | struct device *dev; | |
3947be19 | 599 | |
10fbcf4c KS |
600 | dev = subsys_find_device_by_id(&memory_subsys, block_id, hintdev); |
601 | if (hint) | |
602 | put_device(&hint->dev); | |
603 | if (!dev) | |
3947be19 | 604 | return NULL; |
7315f0cc | 605 | return to_memory_block(dev); |
3947be19 DH |
606 | } |
607 | ||
98383031 RH |
608 | /* |
609 | * For now, we have a linear search to go find the appropriate | |
610 | * memory_block corresponding to a particular phys_index. If | |
611 | * this gets to be a real problem, we can always use a radix | |
612 | * tree or something here. | |
613 | * | |
10fbcf4c | 614 | * This could be made generic for all device subsystems. |
98383031 RH |
615 | */ |
616 | struct memory_block *find_memory_block(struct mem_section *section) | |
617 | { | |
618 | return find_memory_block_hinted(section, NULL); | |
619 | } | |
620 | ||
96b2c0fc NF |
621 | static struct attribute *memory_memblk_attrs[] = { |
622 | &dev_attr_phys_index.attr, | |
96b2c0fc NF |
623 | &dev_attr_state.attr, |
624 | &dev_attr_phys_device.attr, | |
625 | &dev_attr_removable.attr, | |
ed2f2400 ZZ |
626 | #ifdef CONFIG_MEMORY_HOTREMOVE |
627 | &dev_attr_valid_zones.attr, | |
628 | #endif | |
96b2c0fc NF |
629 | NULL |
630 | }; | |
631 | ||
632 | static struct attribute_group memory_memblk_attr_group = { | |
633 | .attrs = memory_memblk_attrs, | |
634 | }; | |
635 | ||
636 | static const struct attribute_group *memory_memblk_attr_groups[] = { | |
637 | &memory_memblk_attr_group, | |
638 | NULL, | |
639 | }; | |
640 | ||
641 | /* | |
642 | * register_memory - Setup a sysfs device for a memory block | |
643 | */ | |
644 | static | |
645 | int register_memory(struct memory_block *memory) | |
646 | { | |
96b2c0fc NF |
647 | memory->dev.bus = &memory_subsys; |
648 | memory->dev.id = memory->start_section_nr / sections_per_block; | |
649 | memory->dev.release = memory_block_release; | |
650 | memory->dev.groups = memory_memblk_attr_groups; | |
f991fae5 | 651 | memory->dev.offline = memory->state == MEM_OFFLINE; |
96b2c0fc | 652 | |
879f1bec | 653 | return device_register(&memory->dev); |
96b2c0fc NF |
654 | } |
655 | ||
0c2c99b1 NF |
656 | static int init_memory_block(struct memory_block **memory, |
657 | struct mem_section *section, unsigned long state) | |
e4619c85 | 658 | { |
0c2c99b1 | 659 | struct memory_block *mem; |
e4619c85 | 660 | unsigned long start_pfn; |
0c2c99b1 | 661 | int scn_nr; |
e4619c85 NF |
662 | int ret = 0; |
663 | ||
0c2c99b1 | 664 | mem = kzalloc(sizeof(*mem), GFP_KERNEL); |
e4619c85 NF |
665 | if (!mem) |
666 | return -ENOMEM; | |
667 | ||
0c2c99b1 | 668 | scn_nr = __section_nr(section); |
d3360164 NF |
669 | mem->start_section_nr = |
670 | base_memory_block_id(scn_nr) * sections_per_block; | |
671 | mem->end_section_nr = mem->start_section_nr + sections_per_block - 1; | |
e4619c85 | 672 | mem->state = state; |
d3360164 | 673 | start_pfn = section_nr_to_pfn(mem->start_section_nr); |
e4619c85 NF |
674 | mem->phys_device = arch_get_memory_phys_device(start_pfn); |
675 | ||
0c2c99b1 | 676 | ret = register_memory(mem); |
0c2c99b1 NF |
677 | |
678 | *memory = mem; | |
679 | return ret; | |
680 | } | |
681 | ||
cb5e39b8 | 682 | static int add_memory_block(int base_section_nr) |
0c2c99b1 | 683 | { |
cb5e39b8 SJ |
684 | struct memory_block *mem; |
685 | int i, ret, section_count = 0, section_nr; | |
0c2c99b1 | 686 | |
cb5e39b8 SJ |
687 | for (i = base_section_nr; |
688 | (i < base_section_nr + sections_per_block) && i < NR_MEM_SECTIONS; | |
689 | i++) { | |
690 | if (!present_section_nr(i)) | |
691 | continue; | |
692 | if (section_count == 0) | |
693 | section_nr = i; | |
694 | section_count++; | |
e4619c85 NF |
695 | } |
696 | ||
cb5e39b8 SJ |
697 | if (section_count == 0) |
698 | return 0; | |
699 | ret = init_memory_block(&mem, __nr_to_section(section_nr), MEM_ONLINE); | |
700 | if (ret) | |
701 | return ret; | |
702 | mem->section_count = section_count; | |
703 | return 0; | |
e4619c85 NF |
704 | } |
705 | ||
4edd7cef DR |
706 | /* |
707 | * need an interface for the VM to add new memory regions, | |
708 | * but without onlining it. | |
709 | */ | |
710 | int register_new_memory(int nid, struct mem_section *section) | |
711 | { | |
d7f80530 SJ |
712 | int ret = 0; |
713 | struct memory_block *mem; | |
b1eaef3d SJ |
714 | |
715 | mutex_lock(&mem_sysfs_mutex); | |
b1eaef3d | 716 | |
d7f80530 SJ |
717 | mem = find_memory_block(section); |
718 | if (mem) { | |
719 | mem->section_count++; | |
720 | put_device(&mem->dev); | |
721 | } else { | |
722 | ret = init_memory_block(&mem, section, MEM_OFFLINE); | |
723 | if (ret) | |
724 | goto out; | |
56c6b5d3 | 725 | mem->section_count++; |
d7f80530 SJ |
726 | } |
727 | ||
728 | if (mem->section_count == sections_per_block) | |
729 | ret = register_mem_sect_under_node(mem, nid); | |
730 | out: | |
731 | mutex_unlock(&mem_sysfs_mutex); | |
b1eaef3d | 732 | return ret; |
4edd7cef DR |
733 | } |
734 | ||
735 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
736 | static void | |
737 | unregister_memory(struct memory_block *memory) | |
738 | { | |
739 | BUG_ON(memory->dev.bus != &memory_subsys); | |
740 | ||
741 | /* drop the ref. we got in remove_memory_block() */ | |
df2b717c | 742 | put_device(&memory->dev); |
4edd7cef DR |
743 | device_unregister(&memory->dev); |
744 | } | |
745 | ||
cc292b0b | 746 | static int remove_memory_section(unsigned long node_id, |
4edd7cef | 747 | struct mem_section *section, int phys_device) |
3947be19 DH |
748 | { |
749 | struct memory_block *mem; | |
750 | ||
2938ffbd | 751 | mutex_lock(&mem_sysfs_mutex); |
1b862aec MH |
752 | |
753 | /* | |
754 | * Some users of the memory hotplug do not want/need memblock to | |
755 | * track all sections. Skip over those. | |
756 | */ | |
3947be19 | 757 | mem = find_memory_block(section); |
1b862aec MH |
758 | if (!mem) |
759 | goto out_unlock; | |
760 | ||
d3360164 | 761 | unregister_mem_sect_under_nodes(mem, __section_nr(section)); |
07681215 NF |
762 | |
763 | mem->section_count--; | |
96b2c0fc | 764 | if (mem->section_count == 0) |
0c2c99b1 | 765 | unregister_memory(mem); |
96b2c0fc | 766 | else |
df2b717c | 767 | put_device(&mem->dev); |
3947be19 | 768 | |
1b862aec | 769 | out_unlock: |
2938ffbd | 770 | mutex_unlock(&mem_sysfs_mutex); |
3947be19 DH |
771 | return 0; |
772 | } | |
773 | ||
3947be19 DH |
774 | int unregister_memory_section(struct mem_section *section) |
775 | { | |
540557b9 | 776 | if (!present_section(section)) |
3947be19 DH |
777 | return -EINVAL; |
778 | ||
cc292b0b | 779 | return remove_memory_section(0, section, 0); |
3947be19 | 780 | } |
4edd7cef | 781 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
3947be19 | 782 | |
6677e3ea YI |
783 | /* return true if the memory block is offlined, otherwise, return false */ |
784 | bool is_memblock_offlined(struct memory_block *mem) | |
785 | { | |
786 | return mem->state == MEM_OFFLINE; | |
787 | } | |
788 | ||
96b2c0fc NF |
789 | static struct attribute *memory_root_attrs[] = { |
790 | #ifdef CONFIG_ARCH_MEMORY_PROBE | |
791 | &dev_attr_probe.attr, | |
792 | #endif | |
793 | ||
794 | #ifdef CONFIG_MEMORY_FAILURE | |
795 | &dev_attr_soft_offline_page.attr, | |
796 | &dev_attr_hard_offline_page.attr, | |
797 | #endif | |
798 | ||
799 | &dev_attr_block_size_bytes.attr, | |
31bc3858 | 800 | &dev_attr_auto_online_blocks.attr, |
96b2c0fc NF |
801 | NULL |
802 | }; | |
803 | ||
804 | static struct attribute_group memory_root_attr_group = { | |
805 | .attrs = memory_root_attrs, | |
806 | }; | |
807 | ||
808 | static const struct attribute_group *memory_root_attr_groups[] = { | |
809 | &memory_root_attr_group, | |
810 | NULL, | |
811 | }; | |
812 | ||
3947be19 DH |
813 | /* |
814 | * Initialize the sysfs support for memory devices... | |
815 | */ | |
816 | int __init memory_dev_init(void) | |
817 | { | |
818 | unsigned int i; | |
819 | int ret; | |
28ec24e2 | 820 | int err; |
0c2c99b1 | 821 | unsigned long block_sz; |
3947be19 | 822 | |
96b2c0fc | 823 | ret = subsys_system_register(&memory_subsys, memory_root_attr_groups); |
28ec24e2 AM |
824 | if (ret) |
825 | goto out; | |
3947be19 | 826 | |
0c2c99b1 NF |
827 | block_sz = get_memory_block_size(); |
828 | sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE; | |
829 | ||
3947be19 DH |
830 | /* |
831 | * Create entries for memory sections that were found | |
832 | * during boot and have been initialized | |
833 | */ | |
b1eaef3d | 834 | mutex_lock(&mem_sysfs_mutex); |
cb5e39b8 | 835 | for (i = 0; i < NR_MEM_SECTIONS; i += sections_per_block) { |
c4e1be9e DH |
836 | /* Don't iterate over sections we know are !present: */ |
837 | if (i > __highest_present_section_nr) | |
838 | break; | |
839 | ||
cb5e39b8 | 840 | err = add_memory_block(i); |
28ec24e2 AM |
841 | if (!ret) |
842 | ret = err; | |
3947be19 | 843 | } |
b1eaef3d | 844 | mutex_unlock(&mem_sysfs_mutex); |
3947be19 | 845 | |
28ec24e2 AM |
846 | out: |
847 | if (ret) | |
2b3a302a | 848 | printk(KERN_ERR "%s() failed: %d\n", __func__, ret); |
3947be19 DH |
849 | return ret; |
850 | } |