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 | ||
3f8e9178 DH |
112 | static ssize_t phys_index_show(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 | */ | |
3f8e9178 DH |
125 | static ssize_t removable_show(struct device *dev, struct device_attribute *attr, |
126 | 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 | */ | |
3f8e9178 DH |
149 | static ssize_t state_show(struct device *dev, struct device_attribute *attr, |
150 | 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 | 189 | /* |
b77eab70 PT |
190 | * The probe routines leave the pages uninitialized, just as the bootmem code |
191 | * does. Make sure we do not access them, but instead use only information from | |
192 | * within sections. | |
2bbcb878 | 193 | */ |
b77eab70 | 194 | static bool pages_correctly_probed(unsigned long start_pfn) |
2bbcb878 | 195 | { |
b77eab70 PT |
196 | unsigned long section_nr = pfn_to_section_nr(start_pfn); |
197 | unsigned long section_nr_end = section_nr + sections_per_block; | |
2bbcb878 MG |
198 | unsigned long pfn = start_pfn; |
199 | ||
200 | /* | |
201 | * memmap between sections is not contiguous except with | |
202 | * SPARSEMEM_VMEMMAP. We lookup the page once per section | |
203 | * and assume memmap is contiguous within each section | |
204 | */ | |
b77eab70 | 205 | for (; section_nr < section_nr_end; section_nr++) { |
2bbcb878 MG |
206 | if (WARN_ON_ONCE(!pfn_valid(pfn))) |
207 | return false; | |
2bbcb878 | 208 | |
b77eab70 | 209 | if (!present_section_nr(section_nr)) { |
1ecc07fd | 210 | pr_warn("section %ld pfn[%lx, %lx) not present\n", |
b77eab70 PT |
211 | section_nr, pfn, pfn + PAGES_PER_SECTION); |
212 | return false; | |
213 | } else if (!valid_section_nr(section_nr)) { | |
1ecc07fd | 214 | pr_warn("section %ld pfn[%lx, %lx) no valid memmap\n", |
b77eab70 PT |
215 | section_nr, pfn, pfn + PAGES_PER_SECTION); |
216 | return false; | |
217 | } else if (online_section_nr(section_nr)) { | |
1ecc07fd | 218 | pr_warn("section %ld pfn[%lx, %lx) is already online\n", |
b77eab70 | 219 | section_nr, pfn, pfn + PAGES_PER_SECTION); |
2bbcb878 MG |
220 | return false; |
221 | } | |
b77eab70 | 222 | pfn += PAGES_PER_SECTION; |
2bbcb878 MG |
223 | } |
224 | ||
225 | return true; | |
226 | } | |
227 | ||
3947be19 DH |
228 | /* |
229 | * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is | |
230 | * OK to have direct references to sparsemem variables in here. | |
231 | */ | |
232 | static int | |
511c2aba | 233 | memory_block_action(unsigned long phys_index, unsigned long action, int online_type) |
3947be19 | 234 | { |
a16cee10 | 235 | unsigned long start_pfn; |
5409d2cd | 236 | unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block; |
3947be19 | 237 | int ret; |
3947be19 | 238 | |
19c07d5e | 239 | start_pfn = section_nr_to_pfn(phys_index); |
de0ed36a | 240 | |
3947be19 | 241 | switch (action) { |
3d3af6af | 242 | case MEM_ONLINE: |
b77eab70 | 243 | if (!pages_correctly_probed(start_pfn)) |
3d3af6af IC |
244 | return -EBUSY; |
245 | ||
246 | ret = online_pages(start_pfn, nr_pages, online_type); | |
247 | break; | |
248 | case MEM_OFFLINE: | |
249 | ret = offline_pages(start_pfn, nr_pages); | |
250 | break; | |
251 | default: | |
252 | WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: " | |
253 | "%ld\n", __func__, phys_index, action, action); | |
254 | ret = -EINVAL; | |
3947be19 | 255 | } |
3947be19 DH |
256 | |
257 | return ret; | |
258 | } | |
259 | ||
dc18d706 | 260 | static int memory_block_change_state(struct memory_block *mem, |
fa2be40f | 261 | unsigned long to_state, unsigned long from_state_req) |
3947be19 | 262 | { |
de0ed36a | 263 | int ret = 0; |
0c2c99b1 | 264 | |
4960e05e RW |
265 | if (mem->state != from_state_req) |
266 | return -EINVAL; | |
3947be19 | 267 | |
0c2c99b1 NF |
268 | if (to_state == MEM_OFFLINE) |
269 | mem->state = MEM_GOING_OFFLINE; | |
270 | ||
fa2be40f SJ |
271 | ret = memory_block_action(mem->start_section_nr, to_state, |
272 | mem->online_type); | |
273 | ||
b2c064b2 | 274 | mem->state = ret ? from_state_req : to_state; |
fa2be40f | 275 | |
4960e05e RW |
276 | return ret; |
277 | } | |
0c2c99b1 | 278 | |
fa2be40f | 279 | /* The device lock serializes operations on memory_subsys_[online|offline] */ |
4960e05e RW |
280 | static int memory_subsys_online(struct device *dev) |
281 | { | |
7315f0cc | 282 | struct memory_block *mem = to_memory_block(dev); |
4960e05e | 283 | int ret; |
3947be19 | 284 | |
fa2be40f SJ |
285 | if (mem->state == MEM_ONLINE) |
286 | return 0; | |
4960e05e | 287 | |
fa2be40f | 288 | /* |
3f8e9178 | 289 | * If we are called from state_store(), online_type will be |
fa2be40f SJ |
290 | * set >= 0 Otherwise we were called from the device online |
291 | * attribute and need to set the online_type. | |
292 | */ | |
293 | if (mem->online_type < 0) | |
4f7c6b49 | 294 | mem->online_type = MMOP_ONLINE_KEEP; |
4960e05e | 295 | |
fa2be40f | 296 | ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE); |
4960e05e | 297 | |
fa2be40f SJ |
298 | /* clear online_type */ |
299 | mem->online_type = -1; | |
4960e05e | 300 | |
4960e05e RW |
301 | return ret; |
302 | } | |
303 | ||
4960e05e | 304 | static int memory_subsys_offline(struct device *dev) |
e90bdb7f | 305 | { |
7315f0cc | 306 | struct memory_block *mem = to_memory_block(dev); |
e90bdb7f | 307 | |
fa2be40f SJ |
308 | if (mem->state == MEM_OFFLINE) |
309 | return 0; | |
e90bdb7f | 310 | |
26bbe7ef SJ |
311 | /* Can't offline block with non-present sections */ |
312 | if (mem->section_count != sections_per_block) | |
313 | return -EINVAL; | |
314 | ||
fa2be40f | 315 | return memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE); |
e90bdb7f | 316 | } |
4960e05e | 317 | |
3f8e9178 DH |
318 | static ssize_t state_store(struct device *dev, struct device_attribute *attr, |
319 | const char *buf, size_t count) | |
3947be19 | 320 | { |
7315f0cc | 321 | struct memory_block *mem = to_memory_block(dev); |
fa2be40f | 322 | int ret, online_type; |
3947be19 | 323 | |
5e33bc41 RW |
324 | ret = lock_device_hotplug_sysfs(); |
325 | if (ret) | |
326 | return ret; | |
4960e05e | 327 | |
1f6a6cc8 | 328 | if (sysfs_streq(buf, "online_kernel")) |
4f7c6b49 | 329 | online_type = MMOP_ONLINE_KERNEL; |
1f6a6cc8 | 330 | else if (sysfs_streq(buf, "online_movable")) |
4f7c6b49 | 331 | online_type = MMOP_ONLINE_MOVABLE; |
1f6a6cc8 | 332 | else if (sysfs_streq(buf, "online")) |
4f7c6b49 | 333 | online_type = MMOP_ONLINE_KEEP; |
1f6a6cc8 | 334 | else if (sysfs_streq(buf, "offline")) |
4f7c6b49 | 335 | online_type = MMOP_OFFLINE; |
a37f8630 YI |
336 | else { |
337 | ret = -EINVAL; | |
338 | goto err; | |
339 | } | |
fa2be40f SJ |
340 | |
341 | switch (online_type) { | |
4f7c6b49 TC |
342 | case MMOP_ONLINE_KERNEL: |
343 | case MMOP_ONLINE_MOVABLE: | |
344 | case MMOP_ONLINE_KEEP: | |
381eab4a | 345 | /* mem->online_type is protected by device_hotplug_lock */ |
fa2be40f SJ |
346 | mem->online_type = online_type; |
347 | ret = device_online(&mem->dev); | |
348 | break; | |
4f7c6b49 | 349 | case MMOP_OFFLINE: |
fa2be40f SJ |
350 | ret = device_offline(&mem->dev); |
351 | break; | |
352 | default: | |
353 | ret = -EINVAL; /* should never happen */ | |
4960e05e | 354 | } |
4960e05e | 355 | |
a37f8630 | 356 | err: |
4960e05e | 357 | unlock_device_hotplug(); |
0c2c99b1 | 358 | |
d66ba15b | 359 | if (ret < 0) |
3947be19 | 360 | return ret; |
d66ba15b RA |
361 | if (ret) |
362 | return -EINVAL; | |
363 | ||
3947be19 DH |
364 | return count; |
365 | } | |
366 | ||
367 | /* | |
368 | * phys_device is a bad name for this. What I really want | |
369 | * is a way to differentiate between memory ranges that | |
370 | * are part of physical devices that constitute | |
371 | * a complete removable unit or fru. | |
372 | * i.e. do these ranges belong to the same physical device, | |
373 | * s.t. if I offline all of these sections I can then | |
374 | * remove the physical device? | |
375 | */ | |
3f8e9178 | 376 | static ssize_t phys_device_show(struct device *dev, |
10fbcf4c | 377 | struct device_attribute *attr, char *buf) |
3947be19 | 378 | { |
7315f0cc | 379 | struct memory_block *mem = to_memory_block(dev); |
3947be19 DH |
380 | return sprintf(buf, "%d\n", mem->phys_device); |
381 | } | |
382 | ||
ed2f2400 | 383 | #ifdef CONFIG_MEMORY_HOTREMOVE |
e5e68930 MH |
384 | static void print_allowed_zone(char *buf, int nid, unsigned long start_pfn, |
385 | unsigned long nr_pages, int online_type, | |
386 | struct zone *default_zone) | |
387 | { | |
388 | struct zone *zone; | |
389 | ||
e5e68930 MH |
390 | zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages); |
391 | if (zone != default_zone) { | |
392 | strcat(buf, " "); | |
393 | strcat(buf, zone->name); | |
394 | } | |
395 | } | |
396 | ||
3f8e9178 | 397 | static ssize_t valid_zones_show(struct device *dev, |
ed2f2400 ZZ |
398 | struct device_attribute *attr, char *buf) |
399 | { | |
400 | struct memory_block *mem = to_memory_block(dev); | |
f1dd2cd1 | 401 | unsigned long start_pfn = section_nr_to_pfn(mem->start_section_nr); |
ed2f2400 | 402 | unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block; |
f1dd2cd1 | 403 | unsigned long valid_start_pfn, valid_end_pfn; |
e5e68930 | 404 | struct zone *default_zone; |
f1dd2cd1 | 405 | int nid; |
ed2f2400 | 406 | |
f1dd2cd1 MH |
407 | /* |
408 | * Check the existing zone. Make sure that we do that only on the | |
409 | * online nodes otherwise the page_zone is not reliable | |
410 | */ | |
411 | if (mem->state == MEM_ONLINE) { | |
4e8346d0 MZ |
412 | /* |
413 | * The block contains more than one zone can not be offlined. | |
414 | * This can happen e.g. for ZONE_DMA and ZONE_DMA32 | |
415 | */ | |
416 | if (!test_pages_in_a_zone(start_pfn, start_pfn + nr_pages, | |
417 | &valid_start_pfn, &valid_end_pfn)) | |
418 | return sprintf(buf, "none\n"); | |
419 | start_pfn = valid_start_pfn; | |
f1dd2cd1 MH |
420 | strcat(buf, page_zone(pfn_to_page(start_pfn))->name); |
421 | goto out; | |
ed2f2400 ZZ |
422 | } |
423 | ||
4e8346d0 | 424 | nid = mem->nid; |
e5e68930 MH |
425 | default_zone = zone_for_pfn_range(MMOP_ONLINE_KEEP, nid, start_pfn, nr_pages); |
426 | strcat(buf, default_zone->name); | |
ed2f2400 | 427 | |
e5e68930 MH |
428 | print_allowed_zone(buf, nid, start_pfn, nr_pages, MMOP_ONLINE_KERNEL, |
429 | default_zone); | |
430 | print_allowed_zone(buf, nid, start_pfn, nr_pages, MMOP_ONLINE_MOVABLE, | |
431 | default_zone); | |
f1dd2cd1 | 432 | out: |
a371d9f1 RA |
433 | strcat(buf, "\n"); |
434 | ||
435 | return strlen(buf); | |
ed2f2400 | 436 | } |
3f8e9178 | 437 | static DEVICE_ATTR_RO(valid_zones); |
ed2f2400 ZZ |
438 | #endif |
439 | ||
3f8e9178 DH |
440 | static DEVICE_ATTR_RO(phys_index); |
441 | static DEVICE_ATTR_RW(state); | |
442 | static DEVICE_ATTR_RO(phys_device); | |
443 | static DEVICE_ATTR_RO(removable); | |
3947be19 | 444 | |
3947be19 DH |
445 | /* |
446 | * Block size attribute stuff | |
447 | */ | |
3f8e9178 DH |
448 | static ssize_t block_size_bytes_show(struct device *dev, |
449 | struct device_attribute *attr, char *buf) | |
3947be19 | 450 | { |
0c2c99b1 | 451 | return sprintf(buf, "%lx\n", get_memory_block_size()); |
3947be19 DH |
452 | } |
453 | ||
3f8e9178 | 454 | static DEVICE_ATTR_RO(block_size_bytes); |
3947be19 | 455 | |
31bc3858 VK |
456 | /* |
457 | * Memory auto online policy. | |
458 | */ | |
459 | ||
3f8e9178 DH |
460 | static ssize_t auto_online_blocks_show(struct device *dev, |
461 | struct device_attribute *attr, char *buf) | |
31bc3858 VK |
462 | { |
463 | if (memhp_auto_online) | |
464 | return sprintf(buf, "online\n"); | |
465 | else | |
466 | return sprintf(buf, "offline\n"); | |
467 | } | |
468 | ||
3f8e9178 DH |
469 | static ssize_t auto_online_blocks_store(struct device *dev, |
470 | struct device_attribute *attr, | |
471 | const char *buf, size_t count) | |
31bc3858 VK |
472 | { |
473 | if (sysfs_streq(buf, "online")) | |
474 | memhp_auto_online = true; | |
475 | else if (sysfs_streq(buf, "offline")) | |
476 | memhp_auto_online = false; | |
477 | else | |
478 | return -EINVAL; | |
479 | ||
480 | return count; | |
481 | } | |
482 | ||
3f8e9178 | 483 | static DEVICE_ATTR_RW(auto_online_blocks); |
31bc3858 | 484 | |
3947be19 DH |
485 | /* |
486 | * Some architectures will have custom drivers to do this, and | |
487 | * will not need to do it from userspace. The fake hot-add code | |
488 | * as well as ppc64 will do all of their discovery in userspace | |
489 | * and will require this interface. | |
490 | */ | |
491 | #ifdef CONFIG_ARCH_MEMORY_PROBE | |
3f8e9178 DH |
492 | static ssize_t probe_store(struct device *dev, struct device_attribute *attr, |
493 | const char *buf, size_t count) | |
3947be19 DH |
494 | { |
495 | u64 phys_addr; | |
cb5490a5 | 496 | int nid, ret; |
61b94fea | 497 | unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block; |
3947be19 | 498 | |
b69deb2b ZZ |
499 | ret = kstrtoull(buf, 0, &phys_addr); |
500 | if (ret) | |
501 | return ret; | |
3947be19 | 502 | |
61b94fea AB |
503 | if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1)) |
504 | return -EINVAL; | |
505 | ||
8df1d0e4 DH |
506 | ret = lock_device_hotplug_sysfs(); |
507 | if (ret) | |
508 | goto out; | |
509 | ||
cb5490a5 | 510 | nid = memory_add_physaddr_to_nid(phys_addr); |
8df1d0e4 DH |
511 | ret = __add_memory(nid, phys_addr, |
512 | MIN_MEMORY_BLOCK_SIZE * sections_per_block); | |
6add7cd6 | 513 | |
cb5490a5 JA |
514 | if (ret) |
515 | goto out; | |
3947be19 | 516 | |
9f0af69b NK |
517 | ret = count; |
518 | out: | |
8df1d0e4 | 519 | unlock_device_hotplug(); |
9f0af69b | 520 | return ret; |
3947be19 | 521 | } |
3947be19 | 522 | |
3f8e9178 | 523 | static DEVICE_ATTR_WO(probe); |
3947be19 DH |
524 | #endif |
525 | ||
facb6011 AK |
526 | #ifdef CONFIG_MEMORY_FAILURE |
527 | /* | |
528 | * Support for offlining pages of memory | |
529 | */ | |
530 | ||
531 | /* Soft offline a page */ | |
3f8e9178 DH |
532 | static ssize_t soft_offline_page_store(struct device *dev, |
533 | struct device_attribute *attr, | |
534 | const char *buf, size_t count) | |
facb6011 AK |
535 | { |
536 | int ret; | |
537 | u64 pfn; | |
538 | if (!capable(CAP_SYS_ADMIN)) | |
539 | return -EPERM; | |
34da5e67 | 540 | if (kstrtoull(buf, 0, &pfn) < 0) |
facb6011 AK |
541 | return -EINVAL; |
542 | pfn >>= PAGE_SHIFT; | |
543 | if (!pfn_valid(pfn)) | |
544 | return -ENXIO; | |
545 | ret = soft_offline_page(pfn_to_page(pfn), 0); | |
546 | return ret == 0 ? count : ret; | |
547 | } | |
548 | ||
549 | /* Forcibly offline a page, including killing processes. */ | |
3f8e9178 DH |
550 | static ssize_t hard_offline_page_store(struct device *dev, |
551 | struct device_attribute *attr, | |
552 | const char *buf, size_t count) | |
facb6011 AK |
553 | { |
554 | int ret; | |
555 | u64 pfn; | |
556 | if (!capable(CAP_SYS_ADMIN)) | |
557 | return -EPERM; | |
34da5e67 | 558 | if (kstrtoull(buf, 0, &pfn) < 0) |
facb6011 AK |
559 | return -EINVAL; |
560 | pfn >>= PAGE_SHIFT; | |
83b57531 | 561 | ret = memory_failure(pfn, 0); |
facb6011 AK |
562 | return ret ? ret : count; |
563 | } | |
564 | ||
3f8e9178 DH |
565 | static DEVICE_ATTR_WO(soft_offline_page); |
566 | static DEVICE_ATTR_WO(hard_offline_page); | |
facb6011 AK |
567 | #endif |
568 | ||
3947be19 DH |
569 | /* |
570 | * Note that phys_device is optional. It is here to allow for | |
571 | * differentiation between which *physical* devices each | |
572 | * section belongs to... | |
573 | */ | |
bc32df00 HC |
574 | int __weak arch_get_memory_phys_device(unsigned long start_pfn) |
575 | { | |
576 | return 0; | |
577 | } | |
3947be19 | 578 | |
10fbcf4c KS |
579 | /* |
580 | * A reference for the returned object is held and the reference for the | |
581 | * hinted object is released. | |
582 | */ | |
98383031 RH |
583 | struct memory_block *find_memory_block_hinted(struct mem_section *section, |
584 | struct memory_block *hint) | |
3947be19 | 585 | { |
0c2c99b1 | 586 | int block_id = base_memory_block_id(__section_nr(section)); |
10fbcf4c KS |
587 | struct device *hintdev = hint ? &hint->dev : NULL; |
588 | struct device *dev; | |
3947be19 | 589 | |
10fbcf4c KS |
590 | dev = subsys_find_device_by_id(&memory_subsys, block_id, hintdev); |
591 | if (hint) | |
592 | put_device(&hint->dev); | |
593 | if (!dev) | |
3947be19 | 594 | return NULL; |
7315f0cc | 595 | return to_memory_block(dev); |
3947be19 DH |
596 | } |
597 | ||
98383031 RH |
598 | /* |
599 | * For now, we have a linear search to go find the appropriate | |
600 | * memory_block corresponding to a particular phys_index. If | |
601 | * this gets to be a real problem, we can always use a radix | |
602 | * tree or something here. | |
603 | * | |
10fbcf4c | 604 | * This could be made generic for all device subsystems. |
98383031 RH |
605 | */ |
606 | struct memory_block *find_memory_block(struct mem_section *section) | |
607 | { | |
608 | return find_memory_block_hinted(section, NULL); | |
609 | } | |
610 | ||
96b2c0fc NF |
611 | static struct attribute *memory_memblk_attrs[] = { |
612 | &dev_attr_phys_index.attr, | |
96b2c0fc NF |
613 | &dev_attr_state.attr, |
614 | &dev_attr_phys_device.attr, | |
615 | &dev_attr_removable.attr, | |
ed2f2400 ZZ |
616 | #ifdef CONFIG_MEMORY_HOTREMOVE |
617 | &dev_attr_valid_zones.attr, | |
618 | #endif | |
96b2c0fc NF |
619 | NULL |
620 | }; | |
621 | ||
622 | static struct attribute_group memory_memblk_attr_group = { | |
623 | .attrs = memory_memblk_attrs, | |
624 | }; | |
625 | ||
626 | static const struct attribute_group *memory_memblk_attr_groups[] = { | |
627 | &memory_memblk_attr_group, | |
628 | NULL, | |
629 | }; | |
630 | ||
631 | /* | |
632 | * register_memory - Setup a sysfs device for a memory block | |
633 | */ | |
634 | static | |
635 | int register_memory(struct memory_block *memory) | |
636 | { | |
085aa2de AY |
637 | int ret; |
638 | ||
96b2c0fc NF |
639 | memory->dev.bus = &memory_subsys; |
640 | memory->dev.id = memory->start_section_nr / sections_per_block; | |
641 | memory->dev.release = memory_block_release; | |
642 | memory->dev.groups = memory_memblk_attr_groups; | |
f991fae5 | 643 | memory->dev.offline = memory->state == MEM_OFFLINE; |
96b2c0fc | 644 | |
085aa2de AY |
645 | ret = device_register(&memory->dev); |
646 | if (ret) | |
647 | put_device(&memory->dev); | |
648 | ||
649 | return ret; | |
96b2c0fc NF |
650 | } |
651 | ||
0c2c99b1 NF |
652 | static int init_memory_block(struct memory_block **memory, |
653 | struct mem_section *section, unsigned long state) | |
e4619c85 | 654 | { |
0c2c99b1 | 655 | struct memory_block *mem; |
e4619c85 | 656 | unsigned long start_pfn; |
0c2c99b1 | 657 | int scn_nr; |
e4619c85 NF |
658 | int ret = 0; |
659 | ||
0c2c99b1 | 660 | mem = kzalloc(sizeof(*mem), GFP_KERNEL); |
e4619c85 NF |
661 | if (!mem) |
662 | return -ENOMEM; | |
663 | ||
0c2c99b1 | 664 | scn_nr = __section_nr(section); |
d3360164 NF |
665 | mem->start_section_nr = |
666 | base_memory_block_id(scn_nr) * sections_per_block; | |
667 | mem->end_section_nr = mem->start_section_nr + sections_per_block - 1; | |
e4619c85 | 668 | mem->state = state; |
d3360164 | 669 | start_pfn = section_nr_to_pfn(mem->start_section_nr); |
e4619c85 NF |
670 | mem->phys_device = arch_get_memory_phys_device(start_pfn); |
671 | ||
0c2c99b1 | 672 | ret = register_memory(mem); |
0c2c99b1 NF |
673 | |
674 | *memory = mem; | |
675 | return ret; | |
676 | } | |
677 | ||
cb5e39b8 | 678 | static int add_memory_block(int base_section_nr) |
0c2c99b1 | 679 | { |
cb5e39b8 SJ |
680 | struct memory_block *mem; |
681 | int i, ret, section_count = 0, section_nr; | |
0c2c99b1 | 682 | |
cb5e39b8 | 683 | for (i = base_section_nr; |
3b6fd6ff | 684 | i < base_section_nr + sections_per_block; |
cb5e39b8 SJ |
685 | i++) { |
686 | if (!present_section_nr(i)) | |
687 | continue; | |
688 | if (section_count == 0) | |
689 | section_nr = i; | |
690 | section_count++; | |
e4619c85 NF |
691 | } |
692 | ||
cb5e39b8 SJ |
693 | if (section_count == 0) |
694 | return 0; | |
695 | ret = init_memory_block(&mem, __nr_to_section(section_nr), MEM_ONLINE); | |
696 | if (ret) | |
697 | return ret; | |
698 | mem->section_count = section_count; | |
699 | return 0; | |
e4619c85 NF |
700 | } |
701 | ||
4edd7cef DR |
702 | /* |
703 | * need an interface for the VM to add new memory regions, | |
704 | * but without onlining it. | |
705 | */ | |
fc44f7f9 | 706 | int hotplug_memory_register(int nid, struct mem_section *section) |
4edd7cef | 707 | { |
d7f80530 SJ |
708 | int ret = 0; |
709 | struct memory_block *mem; | |
b1eaef3d SJ |
710 | |
711 | mutex_lock(&mem_sysfs_mutex); | |
b1eaef3d | 712 | |
d7f80530 SJ |
713 | mem = find_memory_block(section); |
714 | if (mem) { | |
715 | mem->section_count++; | |
716 | put_device(&mem->dev); | |
717 | } else { | |
718 | ret = init_memory_block(&mem, section, MEM_OFFLINE); | |
719 | if (ret) | |
720 | goto out; | |
56c6b5d3 | 721 | mem->section_count++; |
d7f80530 SJ |
722 | } |
723 | ||
d7f80530 SJ |
724 | out: |
725 | mutex_unlock(&mem_sysfs_mutex); | |
b1eaef3d | 726 | return ret; |
4edd7cef DR |
727 | } |
728 | ||
729 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
730 | static void | |
731 | unregister_memory(struct memory_block *memory) | |
732 | { | |
733 | BUG_ON(memory->dev.bus != &memory_subsys); | |
734 | ||
16df1456 | 735 | /* drop the ref. we got in remove_memory_section() */ |
df2b717c | 736 | put_device(&memory->dev); |
4edd7cef DR |
737 | device_unregister(&memory->dev); |
738 | } | |
739 | ||
cc292b0b | 740 | static int remove_memory_section(unsigned long node_id, |
4edd7cef | 741 | struct mem_section *section, int phys_device) |
3947be19 DH |
742 | { |
743 | struct memory_block *mem; | |
744 | ||
2938ffbd | 745 | mutex_lock(&mem_sysfs_mutex); |
1b862aec MH |
746 | |
747 | /* | |
748 | * Some users of the memory hotplug do not want/need memblock to | |
749 | * track all sections. Skip over those. | |
750 | */ | |
3947be19 | 751 | mem = find_memory_block(section); |
1b862aec MH |
752 | if (!mem) |
753 | goto out_unlock; | |
754 | ||
d3360164 | 755 | unregister_mem_sect_under_nodes(mem, __section_nr(section)); |
07681215 NF |
756 | |
757 | mem->section_count--; | |
96b2c0fc | 758 | if (mem->section_count == 0) |
0c2c99b1 | 759 | unregister_memory(mem); |
96b2c0fc | 760 | else |
df2b717c | 761 | put_device(&mem->dev); |
3947be19 | 762 | |
1b862aec | 763 | out_unlock: |
2938ffbd | 764 | mutex_unlock(&mem_sysfs_mutex); |
3947be19 DH |
765 | return 0; |
766 | } | |
767 | ||
3947be19 DH |
768 | int unregister_memory_section(struct mem_section *section) |
769 | { | |
540557b9 | 770 | if (!present_section(section)) |
3947be19 DH |
771 | return -EINVAL; |
772 | ||
cc292b0b | 773 | return remove_memory_section(0, section, 0); |
3947be19 | 774 | } |
4edd7cef | 775 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
3947be19 | 776 | |
6677e3ea YI |
777 | /* return true if the memory block is offlined, otherwise, return false */ |
778 | bool is_memblock_offlined(struct memory_block *mem) | |
779 | { | |
780 | return mem->state == MEM_OFFLINE; | |
781 | } | |
782 | ||
96b2c0fc NF |
783 | static struct attribute *memory_root_attrs[] = { |
784 | #ifdef CONFIG_ARCH_MEMORY_PROBE | |
785 | &dev_attr_probe.attr, | |
786 | #endif | |
787 | ||
788 | #ifdef CONFIG_MEMORY_FAILURE | |
789 | &dev_attr_soft_offline_page.attr, | |
790 | &dev_attr_hard_offline_page.attr, | |
791 | #endif | |
792 | ||
793 | &dev_attr_block_size_bytes.attr, | |
31bc3858 | 794 | &dev_attr_auto_online_blocks.attr, |
96b2c0fc NF |
795 | NULL |
796 | }; | |
797 | ||
798 | static struct attribute_group memory_root_attr_group = { | |
799 | .attrs = memory_root_attrs, | |
800 | }; | |
801 | ||
802 | static const struct attribute_group *memory_root_attr_groups[] = { | |
803 | &memory_root_attr_group, | |
804 | NULL, | |
805 | }; | |
806 | ||
3947be19 DH |
807 | /* |
808 | * Initialize the sysfs support for memory devices... | |
809 | */ | |
810 | int __init memory_dev_init(void) | |
811 | { | |
812 | unsigned int i; | |
813 | int ret; | |
28ec24e2 | 814 | int err; |
0c2c99b1 | 815 | unsigned long block_sz; |
3947be19 | 816 | |
96b2c0fc | 817 | ret = subsys_system_register(&memory_subsys, memory_root_attr_groups); |
28ec24e2 AM |
818 | if (ret) |
819 | goto out; | |
3947be19 | 820 | |
0c2c99b1 NF |
821 | block_sz = get_memory_block_size(); |
822 | sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE; | |
823 | ||
3947be19 DH |
824 | /* |
825 | * Create entries for memory sections that were found | |
826 | * during boot and have been initialized | |
827 | */ | |
b1eaef3d | 828 | mutex_lock(&mem_sysfs_mutex); |
bc8755ba WY |
829 | for (i = 0; i <= __highest_present_section_nr; |
830 | i += sections_per_block) { | |
cb5e39b8 | 831 | err = add_memory_block(i); |
28ec24e2 AM |
832 | if (!ret) |
833 | ret = err; | |
3947be19 | 834 | } |
b1eaef3d | 835 | mutex_unlock(&mem_sysfs_mutex); |
3947be19 | 836 | |
28ec24e2 AM |
837 | out: |
838 | if (ret) | |
2b3a302a | 839 | printk(KERN_ERR "%s() failed: %d\n", __func__, ret); |
3947be19 DH |
840 | return ret; |
841 | } |