]>
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 ZZ |
390 | #ifdef CONFIG_MEMORY_HOTREMOVE |
391 | static ssize_t show_valid_zones(struct device *dev, | |
392 | struct device_attribute *attr, char *buf) | |
393 | { | |
394 | struct memory_block *mem = to_memory_block(dev); | |
4fe85d5a CIK |
395 | unsigned long start_pfn, end_pfn; |
396 | unsigned long valid_start, valid_end, valid_pages; | |
ed2f2400 | 397 | unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block; |
4fe85d5a CIK |
398 | struct zone *zone; |
399 | int zone_shift = 0; | |
ed2f2400 | 400 | |
4fe85d5a CIK |
401 | start_pfn = section_nr_to_pfn(mem->start_section_nr); |
402 | end_pfn = start_pfn + nr_pages; | |
403 | ||
404 | /* The block contains more than one zone can not be offlined. */ | |
405 | if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, &valid_end)) | |
ed2f2400 ZZ |
406 | return sprintf(buf, "none\n"); |
407 | ||
4fe85d5a CIK |
408 | zone = page_zone(pfn_to_page(valid_start)); |
409 | valid_pages = valid_end - valid_start; | |
ed2f2400 | 410 | |
4fe85d5a CIK |
411 | /* MMOP_ONLINE_KEEP */ |
412 | sprintf(buf, "%s", zone->name); | |
ed2f2400 | 413 | |
4fe85d5a CIK |
414 | /* MMOP_ONLINE_KERNEL */ |
415 | zone_can_shift(valid_start, valid_pages, ZONE_NORMAL, &zone_shift); | |
416 | if (zone_shift) { | |
417 | strcat(buf, " "); | |
418 | strcat(buf, (zone + zone_shift)->name); | |
ed2f2400 ZZ |
419 | } |
420 | ||
4fe85d5a CIK |
421 | /* MMOP_ONLINE_MOVABLE */ |
422 | zone_can_shift(valid_start, valid_pages, ZONE_MOVABLE, &zone_shift); | |
423 | if (zone_shift) { | |
424 | strcat(buf, " "); | |
425 | strcat(buf, (zone + zone_shift)->name); | |
f1dd2cd1 | 426 | } |
4fe85d5a | 427 | |
a371d9f1 RA |
428 | strcat(buf, "\n"); |
429 | ||
430 | return strlen(buf); | |
ed2f2400 ZZ |
431 | } |
432 | static DEVICE_ATTR(valid_zones, 0444, show_valid_zones, NULL); | |
433 | #endif | |
434 | ||
10fbcf4c | 435 | static DEVICE_ATTR(phys_index, 0444, show_mem_start_phys_index, NULL); |
10fbcf4c KS |
436 | static DEVICE_ATTR(state, 0644, show_mem_state, store_mem_state); |
437 | static DEVICE_ATTR(phys_device, 0444, show_phys_device, NULL); | |
438 | static DEVICE_ATTR(removable, 0444, show_mem_removable, NULL); | |
3947be19 | 439 | |
3947be19 DH |
440 | /* |
441 | * Block size attribute stuff | |
442 | */ | |
443 | static ssize_t | |
10fbcf4c | 444 | print_block_size(struct device *dev, struct device_attribute *attr, |
8564a6c1 | 445 | char *buf) |
3947be19 | 446 | { |
0c2c99b1 | 447 | return sprintf(buf, "%lx\n", get_memory_block_size()); |
3947be19 DH |
448 | } |
449 | ||
10fbcf4c | 450 | static DEVICE_ATTR(block_size_bytes, 0444, print_block_size, NULL); |
3947be19 | 451 | |
31bc3858 VK |
452 | /* |
453 | * Memory auto online policy. | |
454 | */ | |
455 | ||
456 | static ssize_t | |
457 | show_auto_online_blocks(struct device *dev, struct device_attribute *attr, | |
458 | char *buf) | |
459 | { | |
460 | if (memhp_auto_online) | |
461 | return sprintf(buf, "online\n"); | |
462 | else | |
463 | return sprintf(buf, "offline\n"); | |
464 | } | |
465 | ||
466 | static ssize_t | |
467 | store_auto_online_blocks(struct device *dev, struct device_attribute *attr, | |
468 | const char *buf, size_t count) | |
469 | { | |
470 | if (sysfs_streq(buf, "online")) | |
471 | memhp_auto_online = true; | |
472 | else if (sysfs_streq(buf, "offline")) | |
473 | memhp_auto_online = false; | |
474 | else | |
475 | return -EINVAL; | |
476 | ||
477 | return count; | |
478 | } | |
479 | ||
480 | static DEVICE_ATTR(auto_online_blocks, 0644, show_auto_online_blocks, | |
481 | store_auto_online_blocks); | |
482 | ||
3947be19 DH |
483 | /* |
484 | * Some architectures will have custom drivers to do this, and | |
485 | * will not need to do it from userspace. The fake hot-add code | |
486 | * as well as ppc64 will do all of their discovery in userspace | |
487 | * and will require this interface. | |
488 | */ | |
489 | #ifdef CONFIG_ARCH_MEMORY_PROBE | |
490 | static ssize_t | |
10fbcf4c | 491 | memory_probe_store(struct device *dev, struct device_attribute *attr, |
28812fe1 | 492 | const char *buf, size_t count) |
3947be19 DH |
493 | { |
494 | u64 phys_addr; | |
cb5490a5 | 495 | int nid, ret; |
61b94fea | 496 | unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block; |
3947be19 | 497 | |
b69deb2b ZZ |
498 | ret = kstrtoull(buf, 0, &phys_addr); |
499 | if (ret) | |
500 | return ret; | |
3947be19 | 501 | |
61b94fea AB |
502 | if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1)) |
503 | return -EINVAL; | |
504 | ||
cb5490a5 JA |
505 | nid = memory_add_physaddr_to_nid(phys_addr); |
506 | ret = add_memory(nid, phys_addr, | |
507 | MIN_MEMORY_BLOCK_SIZE * sections_per_block); | |
6add7cd6 | 508 | |
cb5490a5 JA |
509 | if (ret) |
510 | goto out; | |
3947be19 | 511 | |
9f0af69b NK |
512 | ret = count; |
513 | out: | |
514 | return ret; | |
3947be19 | 515 | } |
3947be19 | 516 | |
96b2c0fc | 517 | static DEVICE_ATTR(probe, S_IWUSR, NULL, memory_probe_store); |
3947be19 DH |
518 | #endif |
519 | ||
facb6011 AK |
520 | #ifdef CONFIG_MEMORY_FAILURE |
521 | /* | |
522 | * Support for offlining pages of memory | |
523 | */ | |
524 | ||
525 | /* Soft offline a page */ | |
526 | static ssize_t | |
10fbcf4c KS |
527 | store_soft_offline_page(struct device *dev, |
528 | struct device_attribute *attr, | |
28812fe1 | 529 | const char *buf, size_t count) |
facb6011 AK |
530 | { |
531 | int ret; | |
532 | u64 pfn; | |
533 | if (!capable(CAP_SYS_ADMIN)) | |
534 | return -EPERM; | |
34da5e67 | 535 | if (kstrtoull(buf, 0, &pfn) < 0) |
facb6011 AK |
536 | return -EINVAL; |
537 | pfn >>= PAGE_SHIFT; | |
538 | if (!pfn_valid(pfn)) | |
539 | return -ENXIO; | |
540 | ret = soft_offline_page(pfn_to_page(pfn), 0); | |
541 | return ret == 0 ? count : ret; | |
542 | } | |
543 | ||
544 | /* Forcibly offline a page, including killing processes. */ | |
545 | static ssize_t | |
10fbcf4c KS |
546 | store_hard_offline_page(struct device *dev, |
547 | struct device_attribute *attr, | |
28812fe1 | 548 | const char *buf, size_t count) |
facb6011 AK |
549 | { |
550 | int ret; | |
551 | u64 pfn; | |
552 | if (!capable(CAP_SYS_ADMIN)) | |
553 | return -EPERM; | |
34da5e67 | 554 | if (kstrtoull(buf, 0, &pfn) < 0) |
facb6011 AK |
555 | return -EINVAL; |
556 | pfn >>= PAGE_SHIFT; | |
cd42f4a3 | 557 | ret = memory_failure(pfn, 0, 0); |
facb6011 AK |
558 | return ret ? ret : count; |
559 | } | |
560 | ||
74fef7a8 FB |
561 | static DEVICE_ATTR(soft_offline_page, S_IWUSR, NULL, store_soft_offline_page); |
562 | static DEVICE_ATTR(hard_offline_page, S_IWUSR, NULL, store_hard_offline_page); | |
facb6011 AK |
563 | #endif |
564 | ||
3947be19 DH |
565 | /* |
566 | * Note that phys_device is optional. It is here to allow for | |
567 | * differentiation between which *physical* devices each | |
568 | * section belongs to... | |
569 | */ | |
bc32df00 HC |
570 | int __weak arch_get_memory_phys_device(unsigned long start_pfn) |
571 | { | |
572 | return 0; | |
573 | } | |
3947be19 | 574 | |
10fbcf4c KS |
575 | /* |
576 | * A reference for the returned object is held and the reference for the | |
577 | * hinted object is released. | |
578 | */ | |
98383031 RH |
579 | struct memory_block *find_memory_block_hinted(struct mem_section *section, |
580 | struct memory_block *hint) | |
3947be19 | 581 | { |
0c2c99b1 | 582 | int block_id = base_memory_block_id(__section_nr(section)); |
10fbcf4c KS |
583 | struct device *hintdev = hint ? &hint->dev : NULL; |
584 | struct device *dev; | |
3947be19 | 585 | |
10fbcf4c KS |
586 | dev = subsys_find_device_by_id(&memory_subsys, block_id, hintdev); |
587 | if (hint) | |
588 | put_device(&hint->dev); | |
589 | if (!dev) | |
3947be19 | 590 | return NULL; |
7315f0cc | 591 | return to_memory_block(dev); |
3947be19 DH |
592 | } |
593 | ||
98383031 RH |
594 | /* |
595 | * For now, we have a linear search to go find the appropriate | |
596 | * memory_block corresponding to a particular phys_index. If | |
597 | * this gets to be a real problem, we can always use a radix | |
598 | * tree or something here. | |
599 | * | |
10fbcf4c | 600 | * This could be made generic for all device subsystems. |
98383031 RH |
601 | */ |
602 | struct memory_block *find_memory_block(struct mem_section *section) | |
603 | { | |
604 | return find_memory_block_hinted(section, NULL); | |
605 | } | |
606 | ||
96b2c0fc NF |
607 | static struct attribute *memory_memblk_attrs[] = { |
608 | &dev_attr_phys_index.attr, | |
96b2c0fc NF |
609 | &dev_attr_state.attr, |
610 | &dev_attr_phys_device.attr, | |
611 | &dev_attr_removable.attr, | |
ed2f2400 ZZ |
612 | #ifdef CONFIG_MEMORY_HOTREMOVE |
613 | &dev_attr_valid_zones.attr, | |
614 | #endif | |
96b2c0fc NF |
615 | NULL |
616 | }; | |
617 | ||
618 | static struct attribute_group memory_memblk_attr_group = { | |
619 | .attrs = memory_memblk_attrs, | |
620 | }; | |
621 | ||
622 | static const struct attribute_group *memory_memblk_attr_groups[] = { | |
623 | &memory_memblk_attr_group, | |
624 | NULL, | |
625 | }; | |
626 | ||
627 | /* | |
628 | * register_memory - Setup a sysfs device for a memory block | |
629 | */ | |
630 | static | |
631 | int register_memory(struct memory_block *memory) | |
632 | { | |
96b2c0fc NF |
633 | memory->dev.bus = &memory_subsys; |
634 | memory->dev.id = memory->start_section_nr / sections_per_block; | |
635 | memory->dev.release = memory_block_release; | |
636 | memory->dev.groups = memory_memblk_attr_groups; | |
f991fae5 | 637 | memory->dev.offline = memory->state == MEM_OFFLINE; |
96b2c0fc | 638 | |
879f1bec | 639 | return device_register(&memory->dev); |
96b2c0fc NF |
640 | } |
641 | ||
0c2c99b1 NF |
642 | static int init_memory_block(struct memory_block **memory, |
643 | struct mem_section *section, unsigned long state) | |
e4619c85 | 644 | { |
0c2c99b1 | 645 | struct memory_block *mem; |
e4619c85 | 646 | unsigned long start_pfn; |
0c2c99b1 | 647 | int scn_nr; |
e4619c85 NF |
648 | int ret = 0; |
649 | ||
0c2c99b1 | 650 | mem = kzalloc(sizeof(*mem), GFP_KERNEL); |
e4619c85 NF |
651 | if (!mem) |
652 | return -ENOMEM; | |
653 | ||
0c2c99b1 | 654 | scn_nr = __section_nr(section); |
d3360164 NF |
655 | mem->start_section_nr = |
656 | base_memory_block_id(scn_nr) * sections_per_block; | |
657 | mem->end_section_nr = mem->start_section_nr + sections_per_block - 1; | |
e4619c85 | 658 | mem->state = state; |
d3360164 | 659 | start_pfn = section_nr_to_pfn(mem->start_section_nr); |
e4619c85 NF |
660 | mem->phys_device = arch_get_memory_phys_device(start_pfn); |
661 | ||
0c2c99b1 | 662 | ret = register_memory(mem); |
0c2c99b1 NF |
663 | |
664 | *memory = mem; | |
665 | return ret; | |
666 | } | |
667 | ||
cb5e39b8 | 668 | static int add_memory_block(int base_section_nr) |
0c2c99b1 | 669 | { |
cb5e39b8 SJ |
670 | struct memory_block *mem; |
671 | int i, ret, section_count = 0, section_nr; | |
0c2c99b1 | 672 | |
cb5e39b8 SJ |
673 | for (i = base_section_nr; |
674 | (i < base_section_nr + sections_per_block) && i < NR_MEM_SECTIONS; | |
675 | i++) { | |
676 | if (!present_section_nr(i)) | |
677 | continue; | |
678 | if (section_count == 0) | |
679 | section_nr = i; | |
680 | section_count++; | |
e4619c85 NF |
681 | } |
682 | ||
cb5e39b8 SJ |
683 | if (section_count == 0) |
684 | return 0; | |
685 | ret = init_memory_block(&mem, __nr_to_section(section_nr), MEM_ONLINE); | |
686 | if (ret) | |
687 | return ret; | |
688 | mem->section_count = section_count; | |
689 | return 0; | |
e4619c85 NF |
690 | } |
691 | ||
4edd7cef DR |
692 | /* |
693 | * need an interface for the VM to add new memory regions, | |
694 | * but without onlining it. | |
695 | */ | |
696 | int register_new_memory(int nid, struct mem_section *section) | |
697 | { | |
d7f80530 SJ |
698 | int ret = 0; |
699 | struct memory_block *mem; | |
b1eaef3d SJ |
700 | |
701 | mutex_lock(&mem_sysfs_mutex); | |
b1eaef3d | 702 | |
d7f80530 SJ |
703 | mem = find_memory_block(section); |
704 | if (mem) { | |
705 | mem->section_count++; | |
706 | put_device(&mem->dev); | |
707 | } else { | |
708 | ret = init_memory_block(&mem, section, MEM_OFFLINE); | |
709 | if (ret) | |
710 | goto out; | |
56c6b5d3 | 711 | mem->section_count++; |
d7f80530 SJ |
712 | } |
713 | ||
714 | if (mem->section_count == sections_per_block) | |
715 | ret = register_mem_sect_under_node(mem, nid); | |
716 | out: | |
717 | mutex_unlock(&mem_sysfs_mutex); | |
b1eaef3d | 718 | return ret; |
4edd7cef DR |
719 | } |
720 | ||
721 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
722 | static void | |
723 | unregister_memory(struct memory_block *memory) | |
724 | { | |
725 | BUG_ON(memory->dev.bus != &memory_subsys); | |
726 | ||
727 | /* drop the ref. we got in remove_memory_block() */ | |
df2b717c | 728 | put_device(&memory->dev); |
4edd7cef DR |
729 | device_unregister(&memory->dev); |
730 | } | |
731 | ||
cc292b0b | 732 | static int remove_memory_section(unsigned long node_id, |
4edd7cef | 733 | struct mem_section *section, int phys_device) |
3947be19 DH |
734 | { |
735 | struct memory_block *mem; | |
736 | ||
2938ffbd | 737 | mutex_lock(&mem_sysfs_mutex); |
1b862aec MH |
738 | |
739 | /* | |
740 | * Some users of the memory hotplug do not want/need memblock to | |
741 | * track all sections. Skip over those. | |
742 | */ | |
3947be19 | 743 | mem = find_memory_block(section); |
1b862aec MH |
744 | if (!mem) |
745 | goto out_unlock; | |
746 | ||
d3360164 | 747 | unregister_mem_sect_under_nodes(mem, __section_nr(section)); |
07681215 NF |
748 | |
749 | mem->section_count--; | |
96b2c0fc | 750 | if (mem->section_count == 0) |
0c2c99b1 | 751 | unregister_memory(mem); |
96b2c0fc | 752 | else |
df2b717c | 753 | put_device(&mem->dev); |
3947be19 | 754 | |
1b862aec | 755 | out_unlock: |
2938ffbd | 756 | mutex_unlock(&mem_sysfs_mutex); |
3947be19 DH |
757 | return 0; |
758 | } | |
759 | ||
3947be19 DH |
760 | int unregister_memory_section(struct mem_section *section) |
761 | { | |
540557b9 | 762 | if (!present_section(section)) |
3947be19 DH |
763 | return -EINVAL; |
764 | ||
cc292b0b | 765 | return remove_memory_section(0, section, 0); |
3947be19 | 766 | } |
4edd7cef | 767 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
3947be19 | 768 | |
6677e3ea YI |
769 | /* return true if the memory block is offlined, otherwise, return false */ |
770 | bool is_memblock_offlined(struct memory_block *mem) | |
771 | { | |
772 | return mem->state == MEM_OFFLINE; | |
773 | } | |
774 | ||
96b2c0fc NF |
775 | static struct attribute *memory_root_attrs[] = { |
776 | #ifdef CONFIG_ARCH_MEMORY_PROBE | |
777 | &dev_attr_probe.attr, | |
778 | #endif | |
779 | ||
780 | #ifdef CONFIG_MEMORY_FAILURE | |
781 | &dev_attr_soft_offline_page.attr, | |
782 | &dev_attr_hard_offline_page.attr, | |
783 | #endif | |
784 | ||
785 | &dev_attr_block_size_bytes.attr, | |
31bc3858 | 786 | &dev_attr_auto_online_blocks.attr, |
96b2c0fc NF |
787 | NULL |
788 | }; | |
789 | ||
790 | static struct attribute_group memory_root_attr_group = { | |
791 | .attrs = memory_root_attrs, | |
792 | }; | |
793 | ||
794 | static const struct attribute_group *memory_root_attr_groups[] = { | |
795 | &memory_root_attr_group, | |
796 | NULL, | |
797 | }; | |
798 | ||
3947be19 DH |
799 | /* |
800 | * Initialize the sysfs support for memory devices... | |
801 | */ | |
802 | int __init memory_dev_init(void) | |
803 | { | |
804 | unsigned int i; | |
805 | int ret; | |
28ec24e2 | 806 | int err; |
0c2c99b1 | 807 | unsigned long block_sz; |
3947be19 | 808 | |
96b2c0fc | 809 | ret = subsys_system_register(&memory_subsys, memory_root_attr_groups); |
28ec24e2 AM |
810 | if (ret) |
811 | goto out; | |
3947be19 | 812 | |
0c2c99b1 NF |
813 | block_sz = get_memory_block_size(); |
814 | sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE; | |
815 | ||
3947be19 DH |
816 | /* |
817 | * Create entries for memory sections that were found | |
818 | * during boot and have been initialized | |
819 | */ | |
b1eaef3d | 820 | mutex_lock(&mem_sysfs_mutex); |
cb5e39b8 | 821 | for (i = 0; i < NR_MEM_SECTIONS; i += sections_per_block) { |
c4e1be9e DH |
822 | /* Don't iterate over sections we know are !present: */ |
823 | if (i > __highest_present_section_nr) | |
824 | break; | |
825 | ||
cb5e39b8 | 826 | err = add_memory_block(i); |
28ec24e2 AM |
827 | if (!ret) |
828 | ret = err; | |
3947be19 | 829 | } |
b1eaef3d | 830 | mutex_unlock(&mem_sysfs_mutex); |
3947be19 | 831 | |
28ec24e2 AM |
832 | out: |
833 | if (ret) | |
2b3a302a | 834 | printk(KERN_ERR "%s() failed: %d\n", __func__, ret); |
3947be19 DH |
835 | return ret; |
836 | } |