]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
3947be19 | 2 | /* |
10fbcf4c | 3 | * Memory subsystem support |
3947be19 DH |
4 | * |
5 | * Written by Matt Tolentino <matthew.e.tolentino@intel.com> | |
6 | * Dave Hansen <haveblue@us.ibm.com> | |
7 | * | |
8 | * This file provides the necessary infrastructure to represent | |
9 | * a SPARSEMEM-memory-model system's physical memory in /sysfs. | |
10 | * All arch-independent code that assumes MEMORY_HOTPLUG requires | |
11 | * SPARSEMEM should be contained here, or in mm/memory_hotplug.c. | |
12 | */ | |
13 | ||
3947be19 DH |
14 | #include <linux/module.h> |
15 | #include <linux/init.h> | |
3947be19 | 16 | #include <linux/topology.h> |
c59ede7b | 17 | #include <linux/capability.h> |
3947be19 DH |
18 | #include <linux/device.h> |
19 | #include <linux/memory.h> | |
3947be19 DH |
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> |
7c0f6ba6 | 27 | #include <linux/uaccess.h> |
3947be19 | 28 | |
2938ffbd NF |
29 | static DEFINE_MUTEX(mem_sysfs_mutex); |
30 | ||
3947be19 | 31 | #define MEMORY_CLASS_NAME "memory" |
0c2c99b1 | 32 | |
7315f0cc GZ |
33 | #define to_memory_block(dev) container_of(dev, struct memory_block, dev) |
34 | ||
0c2c99b1 NF |
35 | static int sections_per_block; |
36 | ||
37 | static inline int base_memory_block_id(int section_nr) | |
38 | { | |
39 | return section_nr / sections_per_block; | |
40 | } | |
3947be19 | 41 | |
4960e05e RW |
42 | static int memory_subsys_online(struct device *dev); |
43 | static int memory_subsys_offline(struct device *dev); | |
44 | ||
10fbcf4c | 45 | static struct bus_type memory_subsys = { |
af5ca3f4 | 46 | .name = MEMORY_CLASS_NAME, |
10fbcf4c | 47 | .dev_name = MEMORY_CLASS_NAME, |
4960e05e RW |
48 | .online = memory_subsys_online, |
49 | .offline = memory_subsys_offline, | |
3947be19 DH |
50 | }; |
51 | ||
e041c683 | 52 | static BLOCKING_NOTIFIER_HEAD(memory_chain); |
3947be19 | 53 | |
98a38ebd | 54 | int register_memory_notifier(struct notifier_block *nb) |
3947be19 | 55 | { |
2aeebca2 | 56 | return blocking_notifier_chain_register(&memory_chain, nb); |
3947be19 | 57 | } |
3c82c30c | 58 | EXPORT_SYMBOL(register_memory_notifier); |
3947be19 | 59 | |
98a38ebd | 60 | void unregister_memory_notifier(struct notifier_block *nb) |
3947be19 | 61 | { |
2aeebca2 | 62 | blocking_notifier_chain_unregister(&memory_chain, nb); |
3947be19 | 63 | } |
3c82c30c | 64 | EXPORT_SYMBOL(unregister_memory_notifier); |
3947be19 | 65 | |
925cc71e RJ |
66 | static ATOMIC_NOTIFIER_HEAD(memory_isolate_chain); |
67 | ||
68 | int register_memory_isolate_notifier(struct notifier_block *nb) | |
69 | { | |
70 | return atomic_notifier_chain_register(&memory_isolate_chain, nb); | |
71 | } | |
72 | EXPORT_SYMBOL(register_memory_isolate_notifier); | |
73 | ||
74 | void unregister_memory_isolate_notifier(struct notifier_block *nb) | |
75 | { | |
76 | atomic_notifier_chain_unregister(&memory_isolate_chain, nb); | |
77 | } | |
78 | EXPORT_SYMBOL(unregister_memory_isolate_notifier); | |
79 | ||
fa7194eb YI |
80 | static void memory_block_release(struct device *dev) |
81 | { | |
7315f0cc | 82 | struct memory_block *mem = to_memory_block(dev); |
fa7194eb YI |
83 | |
84 | kfree(mem); | |
85 | } | |
86 | ||
0c2c99b1 NF |
87 | unsigned long __weak memory_block_size_bytes(void) |
88 | { | |
89 | return MIN_MEMORY_BLOCK_SIZE; | |
90 | } | |
91 | ||
92 | static unsigned long get_memory_block_size(void) | |
93 | { | |
94 | unsigned long block_sz; | |
95 | ||
96 | block_sz = memory_block_size_bytes(); | |
97 | ||
98 | /* Validate blk_sz is a power of 2 and not less than section size */ | |
99 | if ((block_sz & (block_sz - 1)) || (block_sz < MIN_MEMORY_BLOCK_SIZE)) { | |
100 | WARN_ON(1); | |
101 | block_sz = MIN_MEMORY_BLOCK_SIZE; | |
102 | } | |
103 | ||
104 | return block_sz; | |
105 | } | |
106 | ||
3947be19 DH |
107 | /* |
108 | * use this as the physical section index that this memsection | |
109 | * uses. | |
110 | */ | |
111 | ||
10fbcf4c KS |
112 | static ssize_t show_mem_start_phys_index(struct device *dev, |
113 | struct device_attribute *attr, char *buf) | |
3947be19 | 114 | { |
7315f0cc | 115 | struct memory_block *mem = to_memory_block(dev); |
d3360164 NF |
116 | unsigned long phys_index; |
117 | ||
118 | phys_index = mem->start_section_nr / sections_per_block; | |
119 | return sprintf(buf, "%08lx\n", phys_index); | |
120 | } | |
121 | ||
5c755e9f BP |
122 | /* |
123 | * Show whether the section of memory is likely to be hot-removable | |
124 | */ | |
10fbcf4c KS |
125 | static ssize_t show_mem_removable(struct device *dev, |
126 | struct device_attribute *attr, char *buf) | |
5c755e9f | 127 | { |
0c2c99b1 NF |
128 | unsigned long i, pfn; |
129 | int ret = 1; | |
7315f0cc | 130 | struct memory_block *mem = to_memory_block(dev); |
5c755e9f | 131 | |
8b0662f2 MH |
132 | if (mem->state != MEM_ONLINE) |
133 | goto out; | |
134 | ||
0c2c99b1 | 135 | for (i = 0; i < sections_per_block; i++) { |
21ea9f5a RA |
136 | if (!present_section_nr(mem->start_section_nr + i)) |
137 | continue; | |
d3360164 | 138 | pfn = section_nr_to_pfn(mem->start_section_nr + i); |
0c2c99b1 NF |
139 | ret &= is_mem_section_removable(pfn, PAGES_PER_SECTION); |
140 | } | |
141 | ||
8b0662f2 | 142 | out: |
5c755e9f BP |
143 | return sprintf(buf, "%d\n", ret); |
144 | } | |
145 | ||
3947be19 DH |
146 | /* |
147 | * online, offline, going offline, etc. | |
148 | */ | |
10fbcf4c KS |
149 | static ssize_t show_mem_state(struct device *dev, |
150 | struct device_attribute *attr, char *buf) | |
3947be19 | 151 | { |
7315f0cc | 152 | struct memory_block *mem = to_memory_block(dev); |
3947be19 DH |
153 | ssize_t len = 0; |
154 | ||
155 | /* | |
156 | * We can probably put these states in a nice little array | |
157 | * so that they're not open-coded | |
158 | */ | |
159 | switch (mem->state) { | |
3d3af6af IC |
160 | case MEM_ONLINE: |
161 | len = sprintf(buf, "online\n"); | |
162 | break; | |
163 | case MEM_OFFLINE: | |
164 | len = sprintf(buf, "offline\n"); | |
165 | break; | |
166 | case MEM_GOING_OFFLINE: | |
167 | len = sprintf(buf, "going-offline\n"); | |
168 | break; | |
169 | default: | |
170 | len = sprintf(buf, "ERROR-UNKNOWN-%ld\n", | |
171 | mem->state); | |
172 | WARN_ON(1); | |
173 | break; | |
3947be19 DH |
174 | } |
175 | ||
176 | return len; | |
177 | } | |
178 | ||
7b78d335 | 179 | int memory_notify(unsigned long val, void *v) |
3947be19 | 180 | { |
e041c683 | 181 | return blocking_notifier_call_chain(&memory_chain, val, v); |
3947be19 DH |
182 | } |
183 | ||
925cc71e RJ |
184 | int memory_isolate_notify(unsigned long val, void *v) |
185 | { | |
186 | return atomic_notifier_call_chain(&memory_isolate_chain, val, v); | |
187 | } | |
188 | ||
2bbcb878 MG |
189 | /* |
190 | * The probe routines leave the pages reserved, just as the bootmem code does. | |
191 | * Make sure they're still that way. | |
192 | */ | |
6056d619 | 193 | static bool pages_correctly_reserved(unsigned long start_pfn) |
2bbcb878 MG |
194 | { |
195 | int i, j; | |
196 | struct page *page; | |
197 | unsigned long pfn = start_pfn; | |
198 | ||
199 | /* | |
200 | * memmap between sections is not contiguous except with | |
201 | * SPARSEMEM_VMEMMAP. We lookup the page once per section | |
202 | * and assume memmap is contiguous within each section | |
203 | */ | |
204 | for (i = 0; i < sections_per_block; i++, pfn += PAGES_PER_SECTION) { | |
205 | if (WARN_ON_ONCE(!pfn_valid(pfn))) | |
206 | return false; | |
207 | page = pfn_to_page(pfn); | |
208 | ||
209 | for (j = 0; j < PAGES_PER_SECTION; j++) { | |
210 | if (PageReserved(page + j)) | |
211 | continue; | |
212 | ||
213 | printk(KERN_WARNING "section number %ld page number %d " | |
214 | "not reserved, was it already online?\n", | |
215 | pfn_to_section_nr(pfn), j); | |
216 | ||
217 | return false; | |
218 | } | |
219 | } | |
220 | ||
221 | return true; | |
222 | } | |
223 | ||
3947be19 DH |
224 | /* |
225 | * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is | |
226 | * OK to have direct references to sparsemem variables in here. | |
30467e0b | 227 | * Must already be protected by mem_hotplug_begin(). |
3947be19 DH |
228 | */ |
229 | static int | |
511c2aba | 230 | memory_block_action(unsigned long phys_index, unsigned long action, int online_type) |
3947be19 | 231 | { |
a16cee10 | 232 | unsigned long start_pfn; |
5409d2cd | 233 | unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block; |
3947be19 | 234 | int ret; |
3947be19 | 235 | |
19c07d5e | 236 | start_pfn = section_nr_to_pfn(phys_index); |
de0ed36a | 237 | |
3947be19 | 238 | switch (action) { |
3d3af6af IC |
239 | case MEM_ONLINE: |
240 | if (!pages_correctly_reserved(start_pfn)) | |
241 | return -EBUSY; | |
242 | ||
243 | ret = online_pages(start_pfn, nr_pages, online_type); | |
244 | break; | |
245 | case MEM_OFFLINE: | |
246 | ret = offline_pages(start_pfn, nr_pages); | |
247 | break; | |
248 | default: | |
249 | WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: " | |
250 | "%ld\n", __func__, phys_index, action, action); | |
251 | ret = -EINVAL; | |
3947be19 | 252 | } |
3947be19 DH |
253 | |
254 | return ret; | |
255 | } | |
256 | ||
dc18d706 | 257 | static int memory_block_change_state(struct memory_block *mem, |
fa2be40f | 258 | unsigned long to_state, unsigned long from_state_req) |
3947be19 | 259 | { |
de0ed36a | 260 | int ret = 0; |
0c2c99b1 | 261 | |
4960e05e RW |
262 | if (mem->state != from_state_req) |
263 | return -EINVAL; | |
3947be19 | 264 | |
0c2c99b1 NF |
265 | if (to_state == MEM_OFFLINE) |
266 | mem->state = MEM_GOING_OFFLINE; | |
267 | ||
fa2be40f SJ |
268 | ret = memory_block_action(mem->start_section_nr, to_state, |
269 | mem->online_type); | |
270 | ||
b2c064b2 | 271 | mem->state = ret ? from_state_req : to_state; |
fa2be40f | 272 | |
4960e05e RW |
273 | return ret; |
274 | } | |
0c2c99b1 | 275 | |
fa2be40f | 276 | /* The device lock serializes operations on memory_subsys_[online|offline] */ |
4960e05e RW |
277 | static int memory_subsys_online(struct device *dev) |
278 | { | |
7315f0cc | 279 | struct memory_block *mem = to_memory_block(dev); |
4960e05e | 280 | int ret; |
3947be19 | 281 | |
fa2be40f SJ |
282 | if (mem->state == MEM_ONLINE) |
283 | return 0; | |
4960e05e | 284 | |
fa2be40f SJ |
285 | /* |
286 | * If we are called from store_mem_state(), online_type will be | |
287 | * set >= 0 Otherwise we were called from the device online | |
288 | * attribute and need to set the online_type. | |
289 | */ | |
290 | if (mem->online_type < 0) | |
4f7c6b49 | 291 | mem->online_type = MMOP_ONLINE_KEEP; |
4960e05e | 292 | |
30467e0b | 293 | /* Already under protection of mem_hotplug_begin() */ |
fa2be40f | 294 | ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE); |
4960e05e | 295 | |
fa2be40f SJ |
296 | /* clear online_type */ |
297 | mem->online_type = -1; | |
4960e05e | 298 | |
4960e05e RW |
299 | return ret; |
300 | } | |
301 | ||
4960e05e | 302 | static int memory_subsys_offline(struct device *dev) |
e90bdb7f | 303 | { |
7315f0cc | 304 | struct memory_block *mem = to_memory_block(dev); |
e90bdb7f | 305 | |
fa2be40f SJ |
306 | if (mem->state == MEM_OFFLINE) |
307 | return 0; | |
e90bdb7f | 308 | |
26bbe7ef SJ |
309 | /* Can't offline block with non-present sections */ |
310 | if (mem->section_count != sections_per_block) | |
311 | return -EINVAL; | |
312 | ||
fa2be40f | 313 | return memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE); |
e90bdb7f | 314 | } |
4960e05e | 315 | |
3947be19 | 316 | static ssize_t |
10fbcf4c KS |
317 | store_mem_state(struct device *dev, |
318 | struct device_attribute *attr, const char *buf, size_t count) | |
3947be19 | 319 | { |
7315f0cc | 320 | struct memory_block *mem = to_memory_block(dev); |
fa2be40f | 321 | int ret, online_type; |
3947be19 | 322 | |
5e33bc41 RW |
323 | ret = lock_device_hotplug_sysfs(); |
324 | if (ret) | |
325 | return ret; | |
4960e05e | 326 | |
1f6a6cc8 | 327 | if (sysfs_streq(buf, "online_kernel")) |
4f7c6b49 | 328 | online_type = MMOP_ONLINE_KERNEL; |
1f6a6cc8 | 329 | else if (sysfs_streq(buf, "online_movable")) |
4f7c6b49 | 330 | online_type = MMOP_ONLINE_MOVABLE; |
1f6a6cc8 | 331 | else if (sysfs_streq(buf, "online")) |
4f7c6b49 | 332 | online_type = MMOP_ONLINE_KEEP; |
1f6a6cc8 | 333 | else if (sysfs_streq(buf, "offline")) |
4f7c6b49 | 334 | online_type = MMOP_OFFLINE; |
a37f8630 YI |
335 | else { |
336 | ret = -EINVAL; | |
337 | goto err; | |
338 | } | |
fa2be40f | 339 | |
30467e0b DR |
340 | /* |
341 | * Memory hotplug needs to hold mem_hotplug_begin() for probe to find | |
342 | * the correct memory block to online before doing device_online(dev), | |
343 | * which will take dev->mutex. Take the lock early to prevent an | |
344 | * inversion, memory_subsys_online() callbacks will be implemented by | |
345 | * assuming it's already protected. | |
346 | */ | |
347 | mem_hotplug_begin(); | |
348 | ||
fa2be40f | 349 | switch (online_type) { |
4f7c6b49 TC |
350 | case MMOP_ONLINE_KERNEL: |
351 | case MMOP_ONLINE_MOVABLE: | |
352 | case MMOP_ONLINE_KEEP: | |
fa2be40f SJ |
353 | mem->online_type = online_type; |
354 | ret = device_online(&mem->dev); | |
355 | break; | |
4f7c6b49 | 356 | case MMOP_OFFLINE: |
fa2be40f SJ |
357 | ret = device_offline(&mem->dev); |
358 | break; | |
359 | default: | |
360 | ret = -EINVAL; /* should never happen */ | |
4960e05e | 361 | } |
4960e05e | 362 | |
30467e0b | 363 | mem_hotplug_done(); |
a37f8630 | 364 | err: |
4960e05e | 365 | unlock_device_hotplug(); |
0c2c99b1 | 366 | |
d66ba15b | 367 | if (ret < 0) |
3947be19 | 368 | return ret; |
d66ba15b RA |
369 | if (ret) |
370 | return -EINVAL; | |
371 | ||
3947be19 DH |
372 | return count; |
373 | } | |
374 | ||
375 | /* | |
376 | * phys_device is a bad name for this. What I really want | |
377 | * is a way to differentiate between memory ranges that | |
378 | * are part of physical devices that constitute | |
379 | * a complete removable unit or fru. | |
380 | * i.e. do these ranges belong to the same physical device, | |
381 | * s.t. if I offline all of these sections I can then | |
382 | * remove the physical device? | |
383 | */ | |
10fbcf4c KS |
384 | static ssize_t show_phys_device(struct device *dev, |
385 | struct device_attribute *attr, char *buf) | |
3947be19 | 386 | { |
7315f0cc | 387 | struct memory_block *mem = to_memory_block(dev); |
3947be19 DH |
388 | return sprintf(buf, "%d\n", mem->phys_device); |
389 | } | |
390 | ||
ed2f2400 | 391 | #ifdef CONFIG_MEMORY_HOTREMOVE |
e5e68930 MH |
392 | static void print_allowed_zone(char *buf, int nid, unsigned long start_pfn, |
393 | unsigned long nr_pages, int online_type, | |
394 | struct zone *default_zone) | |
395 | { | |
396 | struct zone *zone; | |
397 | ||
e5e68930 MH |
398 | zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages); |
399 | if (zone != default_zone) { | |
400 | strcat(buf, " "); | |
401 | strcat(buf, zone->name); | |
402 | } | |
403 | } | |
404 | ||
ed2f2400 ZZ |
405 | static ssize_t show_valid_zones(struct device *dev, |
406 | struct device_attribute *attr, char *buf) | |
407 | { | |
408 | struct memory_block *mem = to_memory_block(dev); | |
f1dd2cd1 | 409 | unsigned long start_pfn = section_nr_to_pfn(mem->start_section_nr); |
ed2f2400 | 410 | unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block; |
f1dd2cd1 | 411 | unsigned long valid_start_pfn, valid_end_pfn; |
e5e68930 | 412 | struct zone *default_zone; |
f1dd2cd1 | 413 | int nid; |
ed2f2400 | 414 | |
f1dd2cd1 MH |
415 | /* |
416 | * The block contains more than one zone can not be offlined. | |
417 | * This can happen e.g. for ZONE_DMA and ZONE_DMA32 | |
418 | */ | |
419 | if (!test_pages_in_a_zone(start_pfn, start_pfn + nr_pages, &valid_start_pfn, &valid_end_pfn)) | |
ed2f2400 ZZ |
420 | return sprintf(buf, "none\n"); |
421 | ||
f1dd2cd1 MH |
422 | start_pfn = valid_start_pfn; |
423 | nr_pages = valid_end_pfn - start_pfn; | |
ed2f2400 | 424 | |
f1dd2cd1 MH |
425 | /* |
426 | * Check the existing zone. Make sure that we do that only on the | |
427 | * online nodes otherwise the page_zone is not reliable | |
428 | */ | |
429 | if (mem->state == MEM_ONLINE) { | |
430 | strcat(buf, page_zone(pfn_to_page(start_pfn))->name); | |
431 | goto out; | |
ed2f2400 ZZ |
432 | } |
433 | ||
f1dd2cd1 | 434 | nid = pfn_to_nid(start_pfn); |
e5e68930 MH |
435 | default_zone = zone_for_pfn_range(MMOP_ONLINE_KEEP, nid, start_pfn, nr_pages); |
436 | strcat(buf, default_zone->name); | |
ed2f2400 | 437 | |
e5e68930 MH |
438 | print_allowed_zone(buf, nid, start_pfn, nr_pages, MMOP_ONLINE_KERNEL, |
439 | default_zone); | |
440 | print_allowed_zone(buf, nid, start_pfn, nr_pages, MMOP_ONLINE_MOVABLE, | |
441 | default_zone); | |
f1dd2cd1 | 442 | out: |
a371d9f1 RA |
443 | strcat(buf, "\n"); |
444 | ||
445 | return strlen(buf); | |
ed2f2400 ZZ |
446 | } |
447 | static DEVICE_ATTR(valid_zones, 0444, show_valid_zones, NULL); | |
448 | #endif | |
449 | ||
10fbcf4c | 450 | static DEVICE_ATTR(phys_index, 0444, show_mem_start_phys_index, NULL); |
10fbcf4c KS |
451 | static DEVICE_ATTR(state, 0644, show_mem_state, store_mem_state); |
452 | static DEVICE_ATTR(phys_device, 0444, show_phys_device, NULL); | |
453 | static DEVICE_ATTR(removable, 0444, show_mem_removable, NULL); | |
3947be19 | 454 | |
3947be19 DH |
455 | /* |
456 | * Block size attribute stuff | |
457 | */ | |
458 | static ssize_t | |
10fbcf4c | 459 | print_block_size(struct device *dev, struct device_attribute *attr, |
8564a6c1 | 460 | char *buf) |
3947be19 | 461 | { |
0c2c99b1 | 462 | return sprintf(buf, "%lx\n", get_memory_block_size()); |
3947be19 DH |
463 | } |
464 | ||
10fbcf4c | 465 | static DEVICE_ATTR(block_size_bytes, 0444, print_block_size, NULL); |
3947be19 | 466 | |
31bc3858 VK |
467 | /* |
468 | * Memory auto online policy. | |
469 | */ | |
470 | ||
471 | static ssize_t | |
472 | show_auto_online_blocks(struct device *dev, struct device_attribute *attr, | |
473 | char *buf) | |
474 | { | |
475 | if (memhp_auto_online) | |
476 | return sprintf(buf, "online\n"); | |
477 | else | |
478 | return sprintf(buf, "offline\n"); | |
479 | } | |
480 | ||
481 | static ssize_t | |
482 | store_auto_online_blocks(struct device *dev, struct device_attribute *attr, | |
483 | const char *buf, size_t count) | |
484 | { | |
485 | if (sysfs_streq(buf, "online")) | |
486 | memhp_auto_online = true; | |
487 | else if (sysfs_streq(buf, "offline")) | |
488 | memhp_auto_online = false; | |
489 | else | |
490 | return -EINVAL; | |
491 | ||
492 | return count; | |
493 | } | |
494 | ||
495 | static DEVICE_ATTR(auto_online_blocks, 0644, show_auto_online_blocks, | |
496 | store_auto_online_blocks); | |
497 | ||
3947be19 DH |
498 | /* |
499 | * Some architectures will have custom drivers to do this, and | |
500 | * will not need to do it from userspace. The fake hot-add code | |
501 | * as well as ppc64 will do all of their discovery in userspace | |
502 | * and will require this interface. | |
503 | */ | |
504 | #ifdef CONFIG_ARCH_MEMORY_PROBE | |
505 | static ssize_t | |
10fbcf4c | 506 | memory_probe_store(struct device *dev, struct device_attribute *attr, |
28812fe1 | 507 | const char *buf, size_t count) |
3947be19 DH |
508 | { |
509 | u64 phys_addr; | |
cb5490a5 | 510 | int nid, ret; |
61b94fea | 511 | unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block; |
3947be19 | 512 | |
b69deb2b ZZ |
513 | ret = kstrtoull(buf, 0, &phys_addr); |
514 | if (ret) | |
515 | return ret; | |
3947be19 | 516 | |
61b94fea AB |
517 | if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1)) |
518 | return -EINVAL; | |
519 | ||
de29654b DH |
520 | ret = lock_device_hotplug_sysfs(); |
521 | if (ret) | |
b69edb14 | 522 | return ret; |
de29654b | 523 | |
cb5490a5 | 524 | nid = memory_add_physaddr_to_nid(phys_addr); |
de29654b DH |
525 | ret = __add_memory(nid, phys_addr, |
526 | MIN_MEMORY_BLOCK_SIZE * sections_per_block); | |
6add7cd6 | 527 | |
cb5490a5 JA |
528 | if (ret) |
529 | goto out; | |
3947be19 | 530 | |
9f0af69b NK |
531 | ret = count; |
532 | out: | |
de29654b | 533 | unlock_device_hotplug(); |
9f0af69b | 534 | return ret; |
3947be19 | 535 | } |
3947be19 | 536 | |
96b2c0fc | 537 | static DEVICE_ATTR(probe, S_IWUSR, NULL, memory_probe_store); |
3947be19 DH |
538 | #endif |
539 | ||
facb6011 AK |
540 | #ifdef CONFIG_MEMORY_FAILURE |
541 | /* | |
542 | * Support for offlining pages of memory | |
543 | */ | |
544 | ||
545 | /* Soft offline a page */ | |
546 | static ssize_t | |
10fbcf4c KS |
547 | store_soft_offline_page(struct device *dev, |
548 | struct device_attribute *attr, | |
28812fe1 | 549 | const char *buf, size_t count) |
facb6011 AK |
550 | { |
551 | int ret; | |
552 | u64 pfn; | |
553 | if (!capable(CAP_SYS_ADMIN)) | |
554 | return -EPERM; | |
34da5e67 | 555 | if (kstrtoull(buf, 0, &pfn) < 0) |
facb6011 AK |
556 | return -EINVAL; |
557 | pfn >>= PAGE_SHIFT; | |
558 | if (!pfn_valid(pfn)) | |
559 | return -ENXIO; | |
2cba4b08 DH |
560 | /* Only online pages can be soft-offlined (esp., not ZONE_DEVICE). */ |
561 | if (!pfn_to_online_page(pfn)) | |
562 | return -EIO; | |
facb6011 AK |
563 | ret = soft_offline_page(pfn_to_page(pfn), 0); |
564 | return ret == 0 ? count : ret; | |
565 | } | |
566 | ||
567 | /* Forcibly offline a page, including killing processes. */ | |
568 | static ssize_t | |
10fbcf4c KS |
569 | store_hard_offline_page(struct device *dev, |
570 | struct device_attribute *attr, | |
28812fe1 | 571 | const char *buf, size_t count) |
facb6011 AK |
572 | { |
573 | int ret; | |
574 | u64 pfn; | |
575 | if (!capable(CAP_SYS_ADMIN)) | |
576 | return -EPERM; | |
34da5e67 | 577 | if (kstrtoull(buf, 0, &pfn) < 0) |
facb6011 AK |
578 | return -EINVAL; |
579 | pfn >>= PAGE_SHIFT; | |
cd42f4a3 | 580 | ret = memory_failure(pfn, 0, 0); |
facb6011 AK |
581 | return ret ? ret : count; |
582 | } | |
583 | ||
74fef7a8 FB |
584 | static DEVICE_ATTR(soft_offline_page, S_IWUSR, NULL, store_soft_offline_page); |
585 | static DEVICE_ATTR(hard_offline_page, S_IWUSR, NULL, store_hard_offline_page); | |
facb6011 AK |
586 | #endif |
587 | ||
3947be19 DH |
588 | /* |
589 | * Note that phys_device is optional. It is here to allow for | |
590 | * differentiation between which *physical* devices each | |
591 | * section belongs to... | |
592 | */ | |
bc32df00 HC |
593 | int __weak arch_get_memory_phys_device(unsigned long start_pfn) |
594 | { | |
595 | return 0; | |
596 | } | |
3947be19 | 597 | |
10fbcf4c KS |
598 | /* |
599 | * A reference for the returned object is held and the reference for the | |
600 | * hinted object is released. | |
601 | */ | |
98383031 RH |
602 | struct memory_block *find_memory_block_hinted(struct mem_section *section, |
603 | struct memory_block *hint) | |
3947be19 | 604 | { |
0c2c99b1 | 605 | int block_id = base_memory_block_id(__section_nr(section)); |
10fbcf4c KS |
606 | struct device *hintdev = hint ? &hint->dev : NULL; |
607 | struct device *dev; | |
3947be19 | 608 | |
10fbcf4c KS |
609 | dev = subsys_find_device_by_id(&memory_subsys, block_id, hintdev); |
610 | if (hint) | |
611 | put_device(&hint->dev); | |
612 | if (!dev) | |
3947be19 | 613 | return NULL; |
7315f0cc | 614 | return to_memory_block(dev); |
3947be19 DH |
615 | } |
616 | ||
98383031 RH |
617 | /* |
618 | * For now, we have a linear search to go find the appropriate | |
619 | * memory_block corresponding to a particular phys_index. If | |
620 | * this gets to be a real problem, we can always use a radix | |
621 | * tree or something here. | |
622 | * | |
10fbcf4c | 623 | * This could be made generic for all device subsystems. |
98383031 RH |
624 | */ |
625 | struct memory_block *find_memory_block(struct mem_section *section) | |
626 | { | |
627 | return find_memory_block_hinted(section, NULL); | |
628 | } | |
629 | ||
96b2c0fc NF |
630 | static struct attribute *memory_memblk_attrs[] = { |
631 | &dev_attr_phys_index.attr, | |
96b2c0fc NF |
632 | &dev_attr_state.attr, |
633 | &dev_attr_phys_device.attr, | |
634 | &dev_attr_removable.attr, | |
ed2f2400 ZZ |
635 | #ifdef CONFIG_MEMORY_HOTREMOVE |
636 | &dev_attr_valid_zones.attr, | |
637 | #endif | |
96b2c0fc NF |
638 | NULL |
639 | }; | |
640 | ||
641 | static struct attribute_group memory_memblk_attr_group = { | |
642 | .attrs = memory_memblk_attrs, | |
643 | }; | |
644 | ||
645 | static const struct attribute_group *memory_memblk_attr_groups[] = { | |
646 | &memory_memblk_attr_group, | |
647 | NULL, | |
648 | }; | |
649 | ||
650 | /* | |
651 | * register_memory - Setup a sysfs device for a memory block | |
652 | */ | |
653 | static | |
654 | int register_memory(struct memory_block *memory) | |
655 | { | |
96b2c0fc NF |
656 | memory->dev.bus = &memory_subsys; |
657 | memory->dev.id = memory->start_section_nr / sections_per_block; | |
658 | memory->dev.release = memory_block_release; | |
659 | memory->dev.groups = memory_memblk_attr_groups; | |
f991fae5 | 660 | memory->dev.offline = memory->state == MEM_OFFLINE; |
96b2c0fc | 661 | |
879f1bec | 662 | return device_register(&memory->dev); |
96b2c0fc NF |
663 | } |
664 | ||
0c2c99b1 NF |
665 | static int init_memory_block(struct memory_block **memory, |
666 | struct mem_section *section, unsigned long state) | |
e4619c85 | 667 | { |
0c2c99b1 | 668 | struct memory_block *mem; |
e4619c85 | 669 | unsigned long start_pfn; |
0c2c99b1 | 670 | int scn_nr; |
e4619c85 NF |
671 | int ret = 0; |
672 | ||
0c2c99b1 | 673 | mem = kzalloc(sizeof(*mem), GFP_KERNEL); |
e4619c85 NF |
674 | if (!mem) |
675 | return -ENOMEM; | |
676 | ||
0c2c99b1 | 677 | scn_nr = __section_nr(section); |
d3360164 NF |
678 | mem->start_section_nr = |
679 | base_memory_block_id(scn_nr) * sections_per_block; | |
680 | mem->end_section_nr = mem->start_section_nr + sections_per_block - 1; | |
e4619c85 | 681 | mem->state = state; |
d3360164 | 682 | start_pfn = section_nr_to_pfn(mem->start_section_nr); |
e4619c85 NF |
683 | mem->phys_device = arch_get_memory_phys_device(start_pfn); |
684 | ||
0c2c99b1 | 685 | ret = register_memory(mem); |
0c2c99b1 NF |
686 | |
687 | *memory = mem; | |
688 | return ret; | |
689 | } | |
690 | ||
cb5e39b8 | 691 | static int add_memory_block(int base_section_nr) |
0c2c99b1 | 692 | { |
cb5e39b8 SJ |
693 | struct memory_block *mem; |
694 | int i, ret, section_count = 0, section_nr; | |
0c2c99b1 | 695 | |
cb5e39b8 SJ |
696 | for (i = base_section_nr; |
697 | (i < base_section_nr + sections_per_block) && i < NR_MEM_SECTIONS; | |
698 | i++) { | |
699 | if (!present_section_nr(i)) | |
700 | continue; | |
701 | if (section_count == 0) | |
702 | section_nr = i; | |
703 | section_count++; | |
e4619c85 NF |
704 | } |
705 | ||
cb5e39b8 SJ |
706 | if (section_count == 0) |
707 | return 0; | |
708 | ret = init_memory_block(&mem, __nr_to_section(section_nr), MEM_ONLINE); | |
709 | if (ret) | |
710 | return ret; | |
711 | mem->section_count = section_count; | |
712 | return 0; | |
e4619c85 NF |
713 | } |
714 | ||
4edd7cef DR |
715 | /* |
716 | * need an interface for the VM to add new memory regions, | |
717 | * but without onlining it. | |
718 | */ | |
719 | int register_new_memory(int nid, struct mem_section *section) | |
720 | { | |
d7f80530 SJ |
721 | int ret = 0; |
722 | struct memory_block *mem; | |
b1eaef3d SJ |
723 | |
724 | mutex_lock(&mem_sysfs_mutex); | |
b1eaef3d | 725 | |
d7f80530 SJ |
726 | mem = find_memory_block(section); |
727 | if (mem) { | |
728 | mem->section_count++; | |
729 | put_device(&mem->dev); | |
730 | } else { | |
731 | ret = init_memory_block(&mem, section, MEM_OFFLINE); | |
732 | if (ret) | |
733 | goto out; | |
56c6b5d3 | 734 | mem->section_count++; |
d7f80530 SJ |
735 | } |
736 | ||
737 | if (mem->section_count == sections_per_block) | |
738 | ret = register_mem_sect_under_node(mem, nid); | |
739 | out: | |
740 | mutex_unlock(&mem_sysfs_mutex); | |
b1eaef3d | 741 | return ret; |
4edd7cef DR |
742 | } |
743 | ||
744 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
745 | static void | |
746 | unregister_memory(struct memory_block *memory) | |
747 | { | |
748 | BUG_ON(memory->dev.bus != &memory_subsys); | |
749 | ||
750 | /* drop the ref. we got in remove_memory_block() */ | |
df2b717c | 751 | put_device(&memory->dev); |
4edd7cef DR |
752 | device_unregister(&memory->dev); |
753 | } | |
754 | ||
cc292b0b | 755 | static int remove_memory_section(unsigned long node_id, |
4edd7cef | 756 | struct mem_section *section, int phys_device) |
3947be19 DH |
757 | { |
758 | struct memory_block *mem; | |
759 | ||
2938ffbd | 760 | mutex_lock(&mem_sysfs_mutex); |
1b862aec MH |
761 | |
762 | /* | |
763 | * Some users of the memory hotplug do not want/need memblock to | |
764 | * track all sections. Skip over those. | |
765 | */ | |
3947be19 | 766 | mem = find_memory_block(section); |
1b862aec MH |
767 | if (!mem) |
768 | goto out_unlock; | |
769 | ||
d3360164 | 770 | unregister_mem_sect_under_nodes(mem, __section_nr(section)); |
07681215 NF |
771 | |
772 | mem->section_count--; | |
96b2c0fc | 773 | if (mem->section_count == 0) |
0c2c99b1 | 774 | unregister_memory(mem); |
96b2c0fc | 775 | else |
df2b717c | 776 | put_device(&mem->dev); |
3947be19 | 777 | |
1b862aec | 778 | out_unlock: |
2938ffbd | 779 | mutex_unlock(&mem_sysfs_mutex); |
3947be19 DH |
780 | return 0; |
781 | } | |
782 | ||
3947be19 DH |
783 | int unregister_memory_section(struct mem_section *section) |
784 | { | |
540557b9 | 785 | if (!present_section(section)) |
3947be19 DH |
786 | return -EINVAL; |
787 | ||
cc292b0b | 788 | return remove_memory_section(0, section, 0); |
3947be19 | 789 | } |
4edd7cef | 790 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
3947be19 | 791 | |
6677e3ea YI |
792 | /* return true if the memory block is offlined, otherwise, return false */ |
793 | bool is_memblock_offlined(struct memory_block *mem) | |
794 | { | |
795 | return mem->state == MEM_OFFLINE; | |
796 | } | |
797 | ||
96b2c0fc NF |
798 | static struct attribute *memory_root_attrs[] = { |
799 | #ifdef CONFIG_ARCH_MEMORY_PROBE | |
800 | &dev_attr_probe.attr, | |
801 | #endif | |
802 | ||
803 | #ifdef CONFIG_MEMORY_FAILURE | |
804 | &dev_attr_soft_offline_page.attr, | |
805 | &dev_attr_hard_offline_page.attr, | |
806 | #endif | |
807 | ||
808 | &dev_attr_block_size_bytes.attr, | |
31bc3858 | 809 | &dev_attr_auto_online_blocks.attr, |
96b2c0fc NF |
810 | NULL |
811 | }; | |
812 | ||
813 | static struct attribute_group memory_root_attr_group = { | |
814 | .attrs = memory_root_attrs, | |
815 | }; | |
816 | ||
817 | static const struct attribute_group *memory_root_attr_groups[] = { | |
818 | &memory_root_attr_group, | |
819 | NULL, | |
820 | }; | |
821 | ||
3947be19 DH |
822 | /* |
823 | * Initialize the sysfs support for memory devices... | |
824 | */ | |
825 | int __init memory_dev_init(void) | |
826 | { | |
827 | unsigned int i; | |
828 | int ret; | |
28ec24e2 | 829 | int err; |
0c2c99b1 | 830 | unsigned long block_sz; |
3947be19 | 831 | |
96b2c0fc | 832 | ret = subsys_system_register(&memory_subsys, memory_root_attr_groups); |
28ec24e2 AM |
833 | if (ret) |
834 | goto out; | |
3947be19 | 835 | |
0c2c99b1 NF |
836 | block_sz = get_memory_block_size(); |
837 | sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE; | |
838 | ||
3947be19 DH |
839 | /* |
840 | * Create entries for memory sections that were found | |
841 | * during boot and have been initialized | |
842 | */ | |
b1eaef3d | 843 | mutex_lock(&mem_sysfs_mutex); |
cb5e39b8 | 844 | for (i = 0; i < NR_MEM_SECTIONS; i += sections_per_block) { |
c4e1be9e DH |
845 | /* Don't iterate over sections we know are !present: */ |
846 | if (i > __highest_present_section_nr) | |
847 | break; | |
848 | ||
cb5e39b8 | 849 | err = add_memory_block(i); |
28ec24e2 AM |
850 | if (!ret) |
851 | ret = err; | |
3947be19 | 852 | } |
b1eaef3d | 853 | mutex_unlock(&mem_sysfs_mutex); |
3947be19 | 854 | |
28ec24e2 AM |
855 | out: |
856 | if (ret) | |
2b3a302a | 857 | printk(KERN_ERR "%s() failed: %d\n", __func__, ret); |
3947be19 DH |
858 | return ret; |
859 | } |