]>
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 | |
db051a0d DH |
42 | static inline int pfn_to_block_id(unsigned long pfn) |
43 | { | |
44 | return base_memory_block_id(pfn_to_section_nr(pfn)); | |
45 | } | |
46 | ||
4960e05e RW |
47 | static int memory_subsys_online(struct device *dev); |
48 | static int memory_subsys_offline(struct device *dev); | |
49 | ||
10fbcf4c | 50 | static struct bus_type memory_subsys = { |
af5ca3f4 | 51 | .name = MEMORY_CLASS_NAME, |
10fbcf4c | 52 | .dev_name = MEMORY_CLASS_NAME, |
4960e05e RW |
53 | .online = memory_subsys_online, |
54 | .offline = memory_subsys_offline, | |
3947be19 DH |
55 | }; |
56 | ||
e041c683 | 57 | static BLOCKING_NOTIFIER_HEAD(memory_chain); |
3947be19 | 58 | |
98a38ebd | 59 | int register_memory_notifier(struct notifier_block *nb) |
3947be19 | 60 | { |
2aeebca2 | 61 | return blocking_notifier_chain_register(&memory_chain, nb); |
3947be19 | 62 | } |
3c82c30c | 63 | EXPORT_SYMBOL(register_memory_notifier); |
3947be19 | 64 | |
98a38ebd | 65 | void unregister_memory_notifier(struct notifier_block *nb) |
3947be19 | 66 | { |
2aeebca2 | 67 | blocking_notifier_chain_unregister(&memory_chain, nb); |
3947be19 | 68 | } |
3c82c30c | 69 | EXPORT_SYMBOL(unregister_memory_notifier); |
3947be19 | 70 | |
925cc71e RJ |
71 | static ATOMIC_NOTIFIER_HEAD(memory_isolate_chain); |
72 | ||
73 | int register_memory_isolate_notifier(struct notifier_block *nb) | |
74 | { | |
75 | return atomic_notifier_chain_register(&memory_isolate_chain, nb); | |
76 | } | |
77 | EXPORT_SYMBOL(register_memory_isolate_notifier); | |
78 | ||
79 | void unregister_memory_isolate_notifier(struct notifier_block *nb) | |
80 | { | |
81 | atomic_notifier_chain_unregister(&memory_isolate_chain, nb); | |
82 | } | |
83 | EXPORT_SYMBOL(unregister_memory_isolate_notifier); | |
84 | ||
fa7194eb YI |
85 | static void memory_block_release(struct device *dev) |
86 | { | |
7315f0cc | 87 | struct memory_block *mem = to_memory_block(dev); |
fa7194eb YI |
88 | |
89 | kfree(mem); | |
90 | } | |
91 | ||
0c2c99b1 NF |
92 | unsigned long __weak memory_block_size_bytes(void) |
93 | { | |
94 | return MIN_MEMORY_BLOCK_SIZE; | |
95 | } | |
c221c0b0 | 96 | EXPORT_SYMBOL_GPL(memory_block_size_bytes); |
0c2c99b1 NF |
97 | |
98 | static unsigned long get_memory_block_size(void) | |
99 | { | |
100 | unsigned long block_sz; | |
101 | ||
102 | block_sz = memory_block_size_bytes(); | |
103 | ||
104 | /* Validate blk_sz is a power of 2 and not less than section size */ | |
105 | if ((block_sz & (block_sz - 1)) || (block_sz < MIN_MEMORY_BLOCK_SIZE)) { | |
106 | WARN_ON(1); | |
107 | block_sz = MIN_MEMORY_BLOCK_SIZE; | |
108 | } | |
109 | ||
110 | return block_sz; | |
111 | } | |
112 | ||
3947be19 DH |
113 | /* |
114 | * use this as the physical section index that this memsection | |
115 | * uses. | |
116 | */ | |
117 | ||
3f8e9178 DH |
118 | static ssize_t phys_index_show(struct device *dev, |
119 | struct device_attribute *attr, char *buf) | |
3947be19 | 120 | { |
7315f0cc | 121 | struct memory_block *mem = to_memory_block(dev); |
d3360164 NF |
122 | unsigned long phys_index; |
123 | ||
124 | phys_index = mem->start_section_nr / sections_per_block; | |
125 | return sprintf(buf, "%08lx\n", phys_index); | |
126 | } | |
127 | ||
5c755e9f BP |
128 | /* |
129 | * Show whether the section of memory is likely to be hot-removable | |
130 | */ | |
3f8e9178 DH |
131 | static ssize_t removable_show(struct device *dev, struct device_attribute *attr, |
132 | char *buf) | |
5c755e9f | 133 | { |
0c2c99b1 NF |
134 | unsigned long i, pfn; |
135 | int ret = 1; | |
7315f0cc | 136 | struct memory_block *mem = to_memory_block(dev); |
5c755e9f | 137 | |
8b0662f2 MH |
138 | if (mem->state != MEM_ONLINE) |
139 | goto out; | |
140 | ||
0c2c99b1 | 141 | for (i = 0; i < sections_per_block; i++) { |
21ea9f5a RA |
142 | if (!present_section_nr(mem->start_section_nr + i)) |
143 | continue; | |
d3360164 | 144 | pfn = section_nr_to_pfn(mem->start_section_nr + i); |
0c2c99b1 NF |
145 | ret &= is_mem_section_removable(pfn, PAGES_PER_SECTION); |
146 | } | |
147 | ||
8b0662f2 | 148 | out: |
5c755e9f BP |
149 | return sprintf(buf, "%d\n", ret); |
150 | } | |
151 | ||
3947be19 DH |
152 | /* |
153 | * online, offline, going offline, etc. | |
154 | */ | |
3f8e9178 DH |
155 | static ssize_t state_show(struct device *dev, struct device_attribute *attr, |
156 | char *buf) | |
3947be19 | 157 | { |
7315f0cc | 158 | struct memory_block *mem = to_memory_block(dev); |
3947be19 DH |
159 | ssize_t len = 0; |
160 | ||
161 | /* | |
162 | * We can probably put these states in a nice little array | |
163 | * so that they're not open-coded | |
164 | */ | |
165 | switch (mem->state) { | |
3d3af6af IC |
166 | case MEM_ONLINE: |
167 | len = sprintf(buf, "online\n"); | |
168 | break; | |
169 | case MEM_OFFLINE: | |
170 | len = sprintf(buf, "offline\n"); | |
171 | break; | |
172 | case MEM_GOING_OFFLINE: | |
173 | len = sprintf(buf, "going-offline\n"); | |
174 | break; | |
175 | default: | |
176 | len = sprintf(buf, "ERROR-UNKNOWN-%ld\n", | |
177 | mem->state); | |
178 | WARN_ON(1); | |
179 | break; | |
3947be19 DH |
180 | } |
181 | ||
182 | return len; | |
183 | } | |
184 | ||
7b78d335 | 185 | int memory_notify(unsigned long val, void *v) |
3947be19 | 186 | { |
e041c683 | 187 | return blocking_notifier_call_chain(&memory_chain, val, v); |
3947be19 DH |
188 | } |
189 | ||
925cc71e RJ |
190 | int memory_isolate_notify(unsigned long val, void *v) |
191 | { | |
192 | return atomic_notifier_call_chain(&memory_isolate_chain, val, v); | |
193 | } | |
194 | ||
2bbcb878 | 195 | /* |
b77eab70 PT |
196 | * The probe routines leave the pages uninitialized, just as the bootmem code |
197 | * does. Make sure we do not access them, but instead use only information from | |
198 | * within sections. | |
2bbcb878 | 199 | */ |
b77eab70 | 200 | static bool pages_correctly_probed(unsigned long start_pfn) |
2bbcb878 | 201 | { |
b77eab70 PT |
202 | unsigned long section_nr = pfn_to_section_nr(start_pfn); |
203 | unsigned long section_nr_end = section_nr + sections_per_block; | |
2bbcb878 MG |
204 | unsigned long pfn = start_pfn; |
205 | ||
206 | /* | |
207 | * memmap between sections is not contiguous except with | |
208 | * SPARSEMEM_VMEMMAP. We lookup the page once per section | |
209 | * and assume memmap is contiguous within each section | |
210 | */ | |
b77eab70 | 211 | for (; section_nr < section_nr_end; section_nr++) { |
2bbcb878 MG |
212 | if (WARN_ON_ONCE(!pfn_valid(pfn))) |
213 | return false; | |
2bbcb878 | 214 | |
b77eab70 | 215 | if (!present_section_nr(section_nr)) { |
1ecc07fd | 216 | pr_warn("section %ld pfn[%lx, %lx) not present\n", |
b77eab70 PT |
217 | section_nr, pfn, pfn + PAGES_PER_SECTION); |
218 | return false; | |
219 | } else if (!valid_section_nr(section_nr)) { | |
1ecc07fd | 220 | pr_warn("section %ld pfn[%lx, %lx) no valid memmap\n", |
b77eab70 PT |
221 | section_nr, pfn, pfn + PAGES_PER_SECTION); |
222 | return false; | |
223 | } else if (online_section_nr(section_nr)) { | |
1ecc07fd | 224 | pr_warn("section %ld pfn[%lx, %lx) is already online\n", |
b77eab70 | 225 | section_nr, pfn, pfn + PAGES_PER_SECTION); |
2bbcb878 MG |
226 | return false; |
227 | } | |
b77eab70 | 228 | pfn += PAGES_PER_SECTION; |
2bbcb878 MG |
229 | } |
230 | ||
231 | return true; | |
232 | } | |
233 | ||
3947be19 DH |
234 | /* |
235 | * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is | |
236 | * OK to have direct references to sparsemem variables in here. | |
237 | */ | |
238 | static int | |
063b8a4c BH |
239 | memory_block_action(unsigned long start_section_nr, unsigned long action, |
240 | int online_type) | |
3947be19 | 241 | { |
a16cee10 | 242 | unsigned long start_pfn; |
5409d2cd | 243 | unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block; |
3947be19 | 244 | int ret; |
3947be19 | 245 | |
063b8a4c | 246 | start_pfn = section_nr_to_pfn(start_section_nr); |
de0ed36a | 247 | |
3947be19 | 248 | switch (action) { |
3d3af6af | 249 | case MEM_ONLINE: |
b77eab70 | 250 | if (!pages_correctly_probed(start_pfn)) |
3d3af6af IC |
251 | return -EBUSY; |
252 | ||
253 | ret = online_pages(start_pfn, nr_pages, online_type); | |
254 | break; | |
255 | case MEM_OFFLINE: | |
256 | ret = offline_pages(start_pfn, nr_pages); | |
257 | break; | |
258 | default: | |
259 | WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: " | |
063b8a4c | 260 | "%ld\n", __func__, start_section_nr, action, action); |
3d3af6af | 261 | ret = -EINVAL; |
3947be19 | 262 | } |
3947be19 DH |
263 | |
264 | return ret; | |
265 | } | |
266 | ||
dc18d706 | 267 | static int memory_block_change_state(struct memory_block *mem, |
fa2be40f | 268 | unsigned long to_state, unsigned long from_state_req) |
3947be19 | 269 | { |
de0ed36a | 270 | int ret = 0; |
0c2c99b1 | 271 | |
4960e05e RW |
272 | if (mem->state != from_state_req) |
273 | return -EINVAL; | |
3947be19 | 274 | |
0c2c99b1 NF |
275 | if (to_state == MEM_OFFLINE) |
276 | mem->state = MEM_GOING_OFFLINE; | |
277 | ||
fa2be40f SJ |
278 | ret = memory_block_action(mem->start_section_nr, to_state, |
279 | mem->online_type); | |
280 | ||
b2c064b2 | 281 | mem->state = ret ? from_state_req : to_state; |
fa2be40f | 282 | |
4960e05e RW |
283 | return ret; |
284 | } | |
0c2c99b1 | 285 | |
fa2be40f | 286 | /* The device lock serializes operations on memory_subsys_[online|offline] */ |
4960e05e RW |
287 | static int memory_subsys_online(struct device *dev) |
288 | { | |
7315f0cc | 289 | struct memory_block *mem = to_memory_block(dev); |
4960e05e | 290 | int ret; |
3947be19 | 291 | |
fa2be40f SJ |
292 | if (mem->state == MEM_ONLINE) |
293 | return 0; | |
4960e05e | 294 | |
fa2be40f | 295 | /* |
3f8e9178 | 296 | * If we are called from state_store(), online_type will be |
fa2be40f SJ |
297 | * set >= 0 Otherwise we were called from the device online |
298 | * attribute and need to set the online_type. | |
299 | */ | |
300 | if (mem->online_type < 0) | |
4f7c6b49 | 301 | mem->online_type = MMOP_ONLINE_KEEP; |
4960e05e | 302 | |
fa2be40f | 303 | ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE); |
4960e05e | 304 | |
fa2be40f SJ |
305 | /* clear online_type */ |
306 | mem->online_type = -1; | |
4960e05e | 307 | |
4960e05e RW |
308 | return ret; |
309 | } | |
310 | ||
4960e05e | 311 | static int memory_subsys_offline(struct device *dev) |
e90bdb7f | 312 | { |
7315f0cc | 313 | struct memory_block *mem = to_memory_block(dev); |
e90bdb7f | 314 | |
fa2be40f SJ |
315 | if (mem->state == MEM_OFFLINE) |
316 | return 0; | |
e90bdb7f | 317 | |
26bbe7ef SJ |
318 | /* Can't offline block with non-present sections */ |
319 | if (mem->section_count != sections_per_block) | |
320 | return -EINVAL; | |
321 | ||
fa2be40f | 322 | return memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE); |
e90bdb7f | 323 | } |
4960e05e | 324 | |
3f8e9178 DH |
325 | static ssize_t state_store(struct device *dev, struct device_attribute *attr, |
326 | const char *buf, size_t count) | |
3947be19 | 327 | { |
7315f0cc | 328 | struct memory_block *mem = to_memory_block(dev); |
fa2be40f | 329 | int ret, online_type; |
3947be19 | 330 | |
5e33bc41 RW |
331 | ret = lock_device_hotplug_sysfs(); |
332 | if (ret) | |
333 | return ret; | |
4960e05e | 334 | |
1f6a6cc8 | 335 | if (sysfs_streq(buf, "online_kernel")) |
4f7c6b49 | 336 | online_type = MMOP_ONLINE_KERNEL; |
1f6a6cc8 | 337 | else if (sysfs_streq(buf, "online_movable")) |
4f7c6b49 | 338 | online_type = MMOP_ONLINE_MOVABLE; |
1f6a6cc8 | 339 | else if (sysfs_streq(buf, "online")) |
4f7c6b49 | 340 | online_type = MMOP_ONLINE_KEEP; |
1f6a6cc8 | 341 | else if (sysfs_streq(buf, "offline")) |
4f7c6b49 | 342 | online_type = MMOP_OFFLINE; |
a37f8630 YI |
343 | else { |
344 | ret = -EINVAL; | |
345 | goto err; | |
346 | } | |
fa2be40f SJ |
347 | |
348 | switch (online_type) { | |
4f7c6b49 TC |
349 | case MMOP_ONLINE_KERNEL: |
350 | case MMOP_ONLINE_MOVABLE: | |
351 | case MMOP_ONLINE_KEEP: | |
381eab4a | 352 | /* mem->online_type is protected by device_hotplug_lock */ |
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 | |
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 | */ | |
3f8e9178 | 383 | static ssize_t phys_device_show(struct device *dev, |
10fbcf4c | 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 | ||
3f8e9178 | 404 | static ssize_t valid_zones_show(struct device *dev, |
ed2f2400 ZZ |
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 | * Check the existing zone. Make sure that we do that only on the | |
416 | * online nodes otherwise the page_zone is not reliable | |
417 | */ | |
418 | if (mem->state == MEM_ONLINE) { | |
4e8346d0 MZ |
419 | /* |
420 | * The block contains more than one zone can not be offlined. | |
421 | * This can happen e.g. for ZONE_DMA and ZONE_DMA32 | |
422 | */ | |
423 | if (!test_pages_in_a_zone(start_pfn, start_pfn + nr_pages, | |
424 | &valid_start_pfn, &valid_end_pfn)) | |
425 | return sprintf(buf, "none\n"); | |
426 | start_pfn = valid_start_pfn; | |
f1dd2cd1 MH |
427 | strcat(buf, page_zone(pfn_to_page(start_pfn))->name); |
428 | goto out; | |
ed2f2400 ZZ |
429 | } |
430 | ||
4e8346d0 | 431 | nid = mem->nid; |
e5e68930 MH |
432 | default_zone = zone_for_pfn_range(MMOP_ONLINE_KEEP, nid, start_pfn, nr_pages); |
433 | strcat(buf, default_zone->name); | |
ed2f2400 | 434 | |
e5e68930 MH |
435 | print_allowed_zone(buf, nid, start_pfn, nr_pages, MMOP_ONLINE_KERNEL, |
436 | default_zone); | |
437 | print_allowed_zone(buf, nid, start_pfn, nr_pages, MMOP_ONLINE_MOVABLE, | |
438 | default_zone); | |
f1dd2cd1 | 439 | out: |
a371d9f1 RA |
440 | strcat(buf, "\n"); |
441 | ||
442 | return strlen(buf); | |
ed2f2400 | 443 | } |
3f8e9178 | 444 | static DEVICE_ATTR_RO(valid_zones); |
ed2f2400 ZZ |
445 | #endif |
446 | ||
3f8e9178 DH |
447 | static DEVICE_ATTR_RO(phys_index); |
448 | static DEVICE_ATTR_RW(state); | |
449 | static DEVICE_ATTR_RO(phys_device); | |
450 | static DEVICE_ATTR_RO(removable); | |
3947be19 | 451 | |
3947be19 DH |
452 | /* |
453 | * Block size attribute stuff | |
454 | */ | |
3f8e9178 DH |
455 | static ssize_t block_size_bytes_show(struct device *dev, |
456 | struct device_attribute *attr, char *buf) | |
3947be19 | 457 | { |
0c2c99b1 | 458 | return sprintf(buf, "%lx\n", get_memory_block_size()); |
3947be19 DH |
459 | } |
460 | ||
3f8e9178 | 461 | static DEVICE_ATTR_RO(block_size_bytes); |
3947be19 | 462 | |
31bc3858 VK |
463 | /* |
464 | * Memory auto online policy. | |
465 | */ | |
466 | ||
3f8e9178 DH |
467 | static ssize_t auto_online_blocks_show(struct device *dev, |
468 | struct device_attribute *attr, char *buf) | |
31bc3858 VK |
469 | { |
470 | if (memhp_auto_online) | |
471 | return sprintf(buf, "online\n"); | |
472 | else | |
473 | return sprintf(buf, "offline\n"); | |
474 | } | |
475 | ||
3f8e9178 DH |
476 | static ssize_t auto_online_blocks_store(struct device *dev, |
477 | struct device_attribute *attr, | |
478 | const char *buf, size_t count) | |
31bc3858 VK |
479 | { |
480 | if (sysfs_streq(buf, "online")) | |
481 | memhp_auto_online = true; | |
482 | else if (sysfs_streq(buf, "offline")) | |
483 | memhp_auto_online = false; | |
484 | else | |
485 | return -EINVAL; | |
486 | ||
487 | return count; | |
488 | } | |
489 | ||
3f8e9178 | 490 | static DEVICE_ATTR_RW(auto_online_blocks); |
31bc3858 | 491 | |
3947be19 DH |
492 | /* |
493 | * Some architectures will have custom drivers to do this, and | |
494 | * will not need to do it from userspace. The fake hot-add code | |
495 | * as well as ppc64 will do all of their discovery in userspace | |
496 | * and will require this interface. | |
497 | */ | |
498 | #ifdef CONFIG_ARCH_MEMORY_PROBE | |
3f8e9178 DH |
499 | static ssize_t probe_store(struct device *dev, struct device_attribute *attr, |
500 | const char *buf, size_t count) | |
3947be19 DH |
501 | { |
502 | u64 phys_addr; | |
cb5490a5 | 503 | int nid, ret; |
61b94fea | 504 | unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block; |
3947be19 | 505 | |
b69deb2b ZZ |
506 | ret = kstrtoull(buf, 0, &phys_addr); |
507 | if (ret) | |
508 | return ret; | |
3947be19 | 509 | |
61b94fea AB |
510 | if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1)) |
511 | return -EINVAL; | |
512 | ||
8df1d0e4 DH |
513 | ret = lock_device_hotplug_sysfs(); |
514 | if (ret) | |
37803841 | 515 | return ret; |
8df1d0e4 | 516 | |
cb5490a5 | 517 | nid = memory_add_physaddr_to_nid(phys_addr); |
8df1d0e4 DH |
518 | ret = __add_memory(nid, phys_addr, |
519 | MIN_MEMORY_BLOCK_SIZE * sections_per_block); | |
6add7cd6 | 520 | |
cb5490a5 JA |
521 | if (ret) |
522 | goto out; | |
3947be19 | 523 | |
9f0af69b NK |
524 | ret = count; |
525 | out: | |
8df1d0e4 | 526 | unlock_device_hotplug(); |
9f0af69b | 527 | return ret; |
3947be19 | 528 | } |
3947be19 | 529 | |
3f8e9178 | 530 | static DEVICE_ATTR_WO(probe); |
3947be19 DH |
531 | #endif |
532 | ||
facb6011 AK |
533 | #ifdef CONFIG_MEMORY_FAILURE |
534 | /* | |
535 | * Support for offlining pages of memory | |
536 | */ | |
537 | ||
538 | /* Soft offline a page */ | |
3f8e9178 DH |
539 | static ssize_t soft_offline_page_store(struct device *dev, |
540 | struct device_attribute *attr, | |
541 | const char *buf, size_t count) | |
facb6011 AK |
542 | { |
543 | int ret; | |
544 | u64 pfn; | |
545 | if (!capable(CAP_SYS_ADMIN)) | |
546 | return -EPERM; | |
34da5e67 | 547 | if (kstrtoull(buf, 0, &pfn) < 0) |
facb6011 AK |
548 | return -EINVAL; |
549 | pfn >>= PAGE_SHIFT; | |
550 | if (!pfn_valid(pfn)) | |
551 | return -ENXIO; | |
552 | ret = soft_offline_page(pfn_to_page(pfn), 0); | |
553 | return ret == 0 ? count : ret; | |
554 | } | |
555 | ||
556 | /* Forcibly offline a page, including killing processes. */ | |
3f8e9178 DH |
557 | static ssize_t hard_offline_page_store(struct device *dev, |
558 | struct device_attribute *attr, | |
559 | const char *buf, size_t count) | |
facb6011 AK |
560 | { |
561 | int ret; | |
562 | u64 pfn; | |
563 | if (!capable(CAP_SYS_ADMIN)) | |
564 | return -EPERM; | |
34da5e67 | 565 | if (kstrtoull(buf, 0, &pfn) < 0) |
facb6011 AK |
566 | return -EINVAL; |
567 | pfn >>= PAGE_SHIFT; | |
83b57531 | 568 | ret = memory_failure(pfn, 0); |
facb6011 AK |
569 | return ret ? ret : count; |
570 | } | |
571 | ||
3f8e9178 DH |
572 | static DEVICE_ATTR_WO(soft_offline_page); |
573 | static DEVICE_ATTR_WO(hard_offline_page); | |
facb6011 AK |
574 | #endif |
575 | ||
3947be19 DH |
576 | /* |
577 | * Note that phys_device is optional. It is here to allow for | |
578 | * differentiation between which *physical* devices each | |
579 | * section belongs to... | |
580 | */ | |
bc32df00 HC |
581 | int __weak arch_get_memory_phys_device(unsigned long start_pfn) |
582 | { | |
583 | return 0; | |
584 | } | |
3947be19 | 585 | |
10fbcf4c KS |
586 | /* |
587 | * A reference for the returned object is held and the reference for the | |
588 | * hinted object is released. | |
589 | */ | |
db051a0d DH |
590 | static struct memory_block *find_memory_block_by_id(int block_id, |
591 | struct memory_block *hint) | |
3947be19 | 592 | { |
10fbcf4c KS |
593 | struct device *hintdev = hint ? &hint->dev : NULL; |
594 | struct device *dev; | |
3947be19 | 595 | |
10fbcf4c KS |
596 | dev = subsys_find_device_by_id(&memory_subsys, block_id, hintdev); |
597 | if (hint) | |
598 | put_device(&hint->dev); | |
599 | if (!dev) | |
3947be19 | 600 | return NULL; |
7315f0cc | 601 | return to_memory_block(dev); |
3947be19 DH |
602 | } |
603 | ||
db051a0d DH |
604 | struct memory_block *find_memory_block_hinted(struct mem_section *section, |
605 | struct memory_block *hint) | |
606 | { | |
607 | int block_id = base_memory_block_id(__section_nr(section)); | |
608 | ||
609 | return find_memory_block_by_id(block_id, hint); | |
610 | } | |
611 | ||
98383031 RH |
612 | /* |
613 | * For now, we have a linear search to go find the appropriate | |
614 | * memory_block corresponding to a particular phys_index. If | |
615 | * this gets to be a real problem, we can always use a radix | |
616 | * tree or something here. | |
617 | * | |
10fbcf4c | 618 | * This could be made generic for all device subsystems. |
98383031 RH |
619 | */ |
620 | struct memory_block *find_memory_block(struct mem_section *section) | |
621 | { | |
622 | return find_memory_block_hinted(section, NULL); | |
623 | } | |
624 | ||
96b2c0fc NF |
625 | static struct attribute *memory_memblk_attrs[] = { |
626 | &dev_attr_phys_index.attr, | |
96b2c0fc NF |
627 | &dev_attr_state.attr, |
628 | &dev_attr_phys_device.attr, | |
629 | &dev_attr_removable.attr, | |
ed2f2400 ZZ |
630 | #ifdef CONFIG_MEMORY_HOTREMOVE |
631 | &dev_attr_valid_zones.attr, | |
632 | #endif | |
96b2c0fc NF |
633 | NULL |
634 | }; | |
635 | ||
636 | static struct attribute_group memory_memblk_attr_group = { | |
637 | .attrs = memory_memblk_attrs, | |
638 | }; | |
639 | ||
640 | static const struct attribute_group *memory_memblk_attr_groups[] = { | |
641 | &memory_memblk_attr_group, | |
642 | NULL, | |
643 | }; | |
644 | ||
645 | /* | |
646 | * register_memory - Setup a sysfs device for a memory block | |
647 | */ | |
648 | static | |
649 | int register_memory(struct memory_block *memory) | |
650 | { | |
085aa2de AY |
651 | int ret; |
652 | ||
96b2c0fc NF |
653 | memory->dev.bus = &memory_subsys; |
654 | memory->dev.id = memory->start_section_nr / sections_per_block; | |
655 | memory->dev.release = memory_block_release; | |
656 | memory->dev.groups = memory_memblk_attr_groups; | |
f991fae5 | 657 | memory->dev.offline = memory->state == MEM_OFFLINE; |
96b2c0fc | 658 | |
085aa2de AY |
659 | ret = device_register(&memory->dev); |
660 | if (ret) | |
661 | put_device(&memory->dev); | |
662 | ||
663 | return ret; | |
96b2c0fc NF |
664 | } |
665 | ||
18115825 DH |
666 | static int init_memory_block(struct memory_block **memory, int block_id, |
667 | unsigned long state) | |
e4619c85 | 668 | { |
0c2c99b1 | 669 | struct memory_block *mem; |
e4619c85 NF |
670 | unsigned long start_pfn; |
671 | int ret = 0; | |
672 | ||
db051a0d DH |
673 | mem = find_memory_block_by_id(block_id, NULL); |
674 | if (mem) { | |
675 | put_device(&mem->dev); | |
676 | return -EEXIST; | |
677 | } | |
0c2c99b1 | 678 | mem = kzalloc(sizeof(*mem), GFP_KERNEL); |
e4619c85 NF |
679 | if (!mem) |
680 | return -ENOMEM; | |
681 | ||
18115825 | 682 | mem->start_section_nr = block_id * sections_per_block; |
d3360164 | 683 | mem->end_section_nr = mem->start_section_nr + sections_per_block - 1; |
e4619c85 | 684 | mem->state = state; |
d3360164 | 685 | start_pfn = section_nr_to_pfn(mem->start_section_nr); |
e4619c85 NF |
686 | mem->phys_device = arch_get_memory_phys_device(start_pfn); |
687 | ||
0c2c99b1 | 688 | ret = register_memory(mem); |
0c2c99b1 NF |
689 | |
690 | *memory = mem; | |
691 | return ret; | |
692 | } | |
693 | ||
cb5e39b8 | 694 | static int add_memory_block(int base_section_nr) |
0c2c99b1 | 695 | { |
cb5e39b8 | 696 | struct memory_block *mem; |
18115825 | 697 | int i, ret, section_count = 0; |
0c2c99b1 | 698 | |
cb5e39b8 | 699 | for (i = base_section_nr; |
3b6fd6ff | 700 | i < base_section_nr + sections_per_block; |
18115825 DH |
701 | i++) |
702 | if (present_section_nr(i)) | |
703 | section_count++; | |
e4619c85 | 704 | |
cb5e39b8 SJ |
705 | if (section_count == 0) |
706 | return 0; | |
18115825 DH |
707 | ret = init_memory_block(&mem, base_memory_block_id(base_section_nr), |
708 | MEM_ONLINE); | |
cb5e39b8 SJ |
709 | if (ret) |
710 | return ret; | |
711 | mem->section_count = section_count; | |
712 | return 0; | |
e4619c85 NF |
713 | } |
714 | ||
db051a0d DH |
715 | static void unregister_memory(struct memory_block *memory) |
716 | { | |
717 | if (WARN_ON_ONCE(memory->dev.bus != &memory_subsys)) | |
718 | return; | |
719 | ||
720 | /* drop the ref. we got via find_memory_block() */ | |
721 | put_device(&memory->dev); | |
722 | device_unregister(&memory->dev); | |
723 | } | |
724 | ||
4edd7cef | 725 | /* |
db051a0d DH |
726 | * Create memory block devices for the given memory area. Start and size |
727 | * have to be aligned to memory block granularity. Memory block devices | |
728 | * will be initialized as offline. | |
4edd7cef | 729 | */ |
db051a0d | 730 | int create_memory_block_devices(unsigned long start, unsigned long size) |
4edd7cef | 731 | { |
db051a0d DH |
732 | const int start_block_id = pfn_to_block_id(PFN_DOWN(start)); |
733 | int end_block_id = pfn_to_block_id(PFN_DOWN(start + size)); | |
d7f80530 | 734 | struct memory_block *mem; |
db051a0d DH |
735 | unsigned long block_id; |
736 | int ret = 0; | |
b1eaef3d | 737 | |
db051a0d DH |
738 | if (WARN_ON_ONCE(!IS_ALIGNED(start, memory_block_size_bytes()) || |
739 | !IS_ALIGNED(size, memory_block_size_bytes()))) | |
740 | return -EINVAL; | |
b1eaef3d | 741 | |
db051a0d DH |
742 | mutex_lock(&mem_sysfs_mutex); |
743 | for (block_id = start_block_id; block_id != end_block_id; block_id++) { | |
18115825 | 744 | ret = init_memory_block(&mem, block_id, MEM_OFFLINE); |
d7f80530 | 745 | if (ret) |
db051a0d DH |
746 | break; |
747 | mem->section_count = sections_per_block; | |
748 | } | |
749 | if (ret) { | |
750 | end_block_id = block_id; | |
751 | for (block_id = start_block_id; block_id != end_block_id; | |
752 | block_id++) { | |
753 | mem = find_memory_block_by_id(block_id, NULL); | |
754 | mem->section_count = 0; | |
755 | unregister_memory(mem); | |
756 | } | |
d7f80530 | 757 | } |
d7f80530 | 758 | mutex_unlock(&mem_sysfs_mutex); |
b1eaef3d | 759 | return ret; |
4edd7cef DR |
760 | } |
761 | ||
4c4b7f9b DH |
762 | /* |
763 | * Remove memory block devices for the given memory area. Start and size | |
764 | * have to be aligned to memory block granularity. Memory block devices | |
765 | * have to be offline. | |
766 | */ | |
767 | void remove_memory_block_devices(unsigned long start, unsigned long size) | |
3947be19 | 768 | { |
4c4b7f9b DH |
769 | const int start_block_id = pfn_to_block_id(PFN_DOWN(start)); |
770 | const int end_block_id = pfn_to_block_id(PFN_DOWN(start + size)); | |
3947be19 | 771 | struct memory_block *mem; |
4c4b7f9b | 772 | int block_id; |
3947be19 | 773 | |
4c4b7f9b DH |
774 | if (WARN_ON_ONCE(!IS_ALIGNED(start, memory_block_size_bytes()) || |
775 | !IS_ALIGNED(size, memory_block_size_bytes()))) | |
cb7b3a36 DH |
776 | return; |
777 | ||
2938ffbd | 778 | mutex_lock(&mem_sysfs_mutex); |
4c4b7f9b DH |
779 | for (block_id = start_block_id; block_id != end_block_id; block_id++) { |
780 | mem = find_memory_block_by_id(block_id, NULL); | |
781 | if (WARN_ON_ONCE(!mem)) | |
782 | continue; | |
783 | mem->section_count = 0; | |
784 | unregister_memory_block_under_nodes(mem); | |
0c2c99b1 | 785 | unregister_memory(mem); |
4c4b7f9b | 786 | } |
2938ffbd | 787 | mutex_unlock(&mem_sysfs_mutex); |
3947be19 DH |
788 | } |
789 | ||
6677e3ea YI |
790 | /* return true if the memory block is offlined, otherwise, return false */ |
791 | bool is_memblock_offlined(struct memory_block *mem) | |
792 | { | |
793 | return mem->state == MEM_OFFLINE; | |
794 | } | |
795 | ||
96b2c0fc NF |
796 | static struct attribute *memory_root_attrs[] = { |
797 | #ifdef CONFIG_ARCH_MEMORY_PROBE | |
798 | &dev_attr_probe.attr, | |
799 | #endif | |
800 | ||
801 | #ifdef CONFIG_MEMORY_FAILURE | |
802 | &dev_attr_soft_offline_page.attr, | |
803 | &dev_attr_hard_offline_page.attr, | |
804 | #endif | |
805 | ||
806 | &dev_attr_block_size_bytes.attr, | |
31bc3858 | 807 | &dev_attr_auto_online_blocks.attr, |
96b2c0fc NF |
808 | NULL |
809 | }; | |
810 | ||
811 | static struct attribute_group memory_root_attr_group = { | |
812 | .attrs = memory_root_attrs, | |
813 | }; | |
814 | ||
815 | static const struct attribute_group *memory_root_attr_groups[] = { | |
816 | &memory_root_attr_group, | |
817 | NULL, | |
818 | }; | |
819 | ||
3947be19 DH |
820 | /* |
821 | * Initialize the sysfs support for memory devices... | |
822 | */ | |
823 | int __init memory_dev_init(void) | |
824 | { | |
825 | unsigned int i; | |
826 | int ret; | |
28ec24e2 | 827 | int err; |
0c2c99b1 | 828 | unsigned long block_sz; |
3947be19 | 829 | |
96b2c0fc | 830 | ret = subsys_system_register(&memory_subsys, memory_root_attr_groups); |
28ec24e2 AM |
831 | if (ret) |
832 | goto out; | |
3947be19 | 833 | |
0c2c99b1 NF |
834 | block_sz = get_memory_block_size(); |
835 | sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE; | |
836 | ||
3947be19 DH |
837 | /* |
838 | * Create entries for memory sections that were found | |
839 | * during boot and have been initialized | |
840 | */ | |
b1eaef3d | 841 | mutex_lock(&mem_sysfs_mutex); |
bc8755ba WY |
842 | for (i = 0; i <= __highest_present_section_nr; |
843 | i += sections_per_block) { | |
cb5e39b8 | 844 | err = add_memory_block(i); |
28ec24e2 AM |
845 | if (!ret) |
846 | ret = err; | |
3947be19 | 847 | } |
b1eaef3d | 848 | mutex_unlock(&mem_sysfs_mutex); |
3947be19 | 849 | |
28ec24e2 AM |
850 | out: |
851 | if (ret) | |
2b3a302a | 852 | printk(KERN_ERR "%s() failed: %d\n", __func__, ret); |
3947be19 DH |
853 | return ret; |
854 | } |