]>
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> | |
9f1b16a5 | 22 | #include <linux/stat.h> |
5a0e3ad6 | 23 | #include <linux/slab.h> |
4fb6eabf | 24 | #include <linux/xarray.h> |
9f1b16a5 | 25 | |
60063497 | 26 | #include <linux/atomic.h> |
7c0f6ba6 | 27 | #include <linux/uaccess.h> |
3947be19 DH |
28 | |
29 | #define MEMORY_CLASS_NAME "memory" | |
0c2c99b1 | 30 | |
4dc8207b DH |
31 | static const char *const online_type_to_str[] = { |
32 | [MMOP_OFFLINE] = "offline", | |
33 | [MMOP_ONLINE] = "online", | |
34 | [MMOP_ONLINE_KERNEL] = "online_kernel", | |
35 | [MMOP_ONLINE_MOVABLE] = "online_movable", | |
36 | }; | |
37 | ||
5f47adf7 | 38 | int memhp_online_type_from_str(const char *str) |
4dc8207b DH |
39 | { |
40 | int i; | |
41 | ||
42 | for (i = 0; i < ARRAY_SIZE(online_type_to_str); i++) { | |
43 | if (sysfs_streq(str, online_type_to_str[i])) | |
44 | return i; | |
45 | } | |
46 | return -EINVAL; | |
47 | } | |
48 | ||
7315f0cc GZ |
49 | #define to_memory_block(dev) container_of(dev, struct memory_block, dev) |
50 | ||
0c2c99b1 NF |
51 | static int sections_per_block; |
52 | ||
178bdbed | 53 | static inline unsigned long memory_block_id(unsigned long section_nr) |
0c2c99b1 NF |
54 | { |
55 | return section_nr / sections_per_block; | |
56 | } | |
3947be19 | 57 | |
90ec010f | 58 | static inline unsigned long pfn_to_block_id(unsigned long pfn) |
db051a0d | 59 | { |
178bdbed | 60 | return memory_block_id(pfn_to_section_nr(pfn)); |
db051a0d DH |
61 | } |
62 | ||
ea884641 DH |
63 | static inline unsigned long phys_to_block_id(unsigned long phys) |
64 | { | |
65 | return pfn_to_block_id(PFN_DOWN(phys)); | |
66 | } | |
67 | ||
4960e05e RW |
68 | static int memory_subsys_online(struct device *dev); |
69 | static int memory_subsys_offline(struct device *dev); | |
70 | ||
10fbcf4c | 71 | static struct bus_type memory_subsys = { |
af5ca3f4 | 72 | .name = MEMORY_CLASS_NAME, |
10fbcf4c | 73 | .dev_name = MEMORY_CLASS_NAME, |
4960e05e RW |
74 | .online = memory_subsys_online, |
75 | .offline = memory_subsys_offline, | |
3947be19 DH |
76 | }; |
77 | ||
4fb6eabf SC |
78 | /* |
79 | * Memory blocks are cached in a local radix tree to avoid | |
80 | * a costly linear search for the corresponding device on | |
81 | * the subsystem bus. | |
82 | */ | |
83 | static DEFINE_XARRAY(memory_blocks); | |
84 | ||
e041c683 | 85 | static BLOCKING_NOTIFIER_HEAD(memory_chain); |
3947be19 | 86 | |
98a38ebd | 87 | int register_memory_notifier(struct notifier_block *nb) |
3947be19 | 88 | { |
2aeebca2 | 89 | return blocking_notifier_chain_register(&memory_chain, nb); |
3947be19 | 90 | } |
3c82c30c | 91 | EXPORT_SYMBOL(register_memory_notifier); |
3947be19 | 92 | |
98a38ebd | 93 | void unregister_memory_notifier(struct notifier_block *nb) |
3947be19 | 94 | { |
2aeebca2 | 95 | blocking_notifier_chain_unregister(&memory_chain, nb); |
3947be19 | 96 | } |
3c82c30c | 97 | EXPORT_SYMBOL(unregister_memory_notifier); |
3947be19 | 98 | |
fa7194eb YI |
99 | static void memory_block_release(struct device *dev) |
100 | { | |
7315f0cc | 101 | struct memory_block *mem = to_memory_block(dev); |
fa7194eb YI |
102 | |
103 | kfree(mem); | |
104 | } | |
105 | ||
0c2c99b1 NF |
106 | unsigned long __weak memory_block_size_bytes(void) |
107 | { | |
108 | return MIN_MEMORY_BLOCK_SIZE; | |
109 | } | |
c221c0b0 | 110 | EXPORT_SYMBOL_GPL(memory_block_size_bytes); |
0c2c99b1 | 111 | |
3947be19 | 112 | /* |
f915fb7f | 113 | * Show the first physical section index (number) of this memory block. |
3947be19 | 114 | */ |
3f8e9178 DH |
115 | static ssize_t phys_index_show(struct device *dev, |
116 | struct device_attribute *attr, char *buf) | |
3947be19 | 117 | { |
7315f0cc | 118 | struct memory_block *mem = to_memory_block(dev); |
d3360164 NF |
119 | unsigned long phys_index; |
120 | ||
121 | phys_index = mem->start_section_nr / sections_per_block; | |
948b3edb | 122 | |
aa838896 | 123 | return sysfs_emit(buf, "%08lx\n", phys_index); |
d3360164 NF |
124 | } |
125 | ||
5c755e9f | 126 | /* |
53cdc1cb DH |
127 | * Legacy interface that we cannot remove. Always indicate "removable" |
128 | * with CONFIG_MEMORY_HOTREMOVE - bad heuristic. | |
5c755e9f | 129 | */ |
3f8e9178 DH |
130 | static ssize_t removable_show(struct device *dev, struct device_attribute *attr, |
131 | char *buf) | |
5c755e9f | 132 | { |
aa838896 | 133 | return sysfs_emit(buf, "%d\n", (int)IS_ENABLED(CONFIG_MEMORY_HOTREMOVE)); |
5c755e9f BP |
134 | } |
135 | ||
3947be19 DH |
136 | /* |
137 | * online, offline, going offline, etc. | |
138 | */ | |
3f8e9178 DH |
139 | static ssize_t state_show(struct device *dev, struct device_attribute *attr, |
140 | char *buf) | |
3947be19 | 141 | { |
7315f0cc | 142 | struct memory_block *mem = to_memory_block(dev); |
973c3911 | 143 | const char *output; |
3947be19 DH |
144 | |
145 | /* | |
146 | * We can probably put these states in a nice little array | |
147 | * so that they're not open-coded | |
148 | */ | |
149 | switch (mem->state) { | |
3d3af6af | 150 | case MEM_ONLINE: |
973c3911 | 151 | output = "online"; |
3d3af6af IC |
152 | break; |
153 | case MEM_OFFLINE: | |
973c3911 | 154 | output = "offline"; |
3d3af6af IC |
155 | break; |
156 | case MEM_GOING_OFFLINE: | |
973c3911 | 157 | output = "going-offline"; |
3d3af6af IC |
158 | break; |
159 | default: | |
3d3af6af | 160 | WARN_ON(1); |
973c3911 | 161 | return sysfs_emit(buf, "ERROR-UNKNOWN-%ld\n", mem->state); |
3947be19 DH |
162 | } |
163 | ||
973c3911 | 164 | return sysfs_emit(buf, "%s\n", output); |
3947be19 DH |
165 | } |
166 | ||
7b78d335 | 167 | int memory_notify(unsigned long val, void *v) |
3947be19 | 168 | { |
e041c683 | 169 | return blocking_notifier_call_chain(&memory_chain, val, v); |
3947be19 DH |
170 | } |
171 | ||
172 | /* | |
173 | * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is | |
174 | * OK to have direct references to sparsemem variables in here. | |
175 | */ | |
176 | static int | |
063b8a4c | 177 | memory_block_action(unsigned long start_section_nr, unsigned long action, |
bd5c2344 | 178 | int online_type, int nid) |
3947be19 | 179 | { |
a16cee10 | 180 | unsigned long start_pfn; |
5409d2cd | 181 | unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block; |
3947be19 | 182 | int ret; |
3947be19 | 183 | |
063b8a4c | 184 | start_pfn = section_nr_to_pfn(start_section_nr); |
de0ed36a | 185 | |
3947be19 | 186 | switch (action) { |
3d3af6af | 187 | case MEM_ONLINE: |
bd5c2344 | 188 | ret = online_pages(start_pfn, nr_pages, online_type, nid); |
3d3af6af IC |
189 | break; |
190 | case MEM_OFFLINE: | |
191 | ret = offline_pages(start_pfn, nr_pages); | |
192 | break; | |
193 | default: | |
194 | WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: " | |
063b8a4c | 195 | "%ld\n", __func__, start_section_nr, action, action); |
3d3af6af | 196 | ret = -EINVAL; |
3947be19 | 197 | } |
3947be19 DH |
198 | |
199 | return ret; | |
200 | } | |
201 | ||
dc18d706 | 202 | static int memory_block_change_state(struct memory_block *mem, |
fa2be40f | 203 | unsigned long to_state, unsigned long from_state_req) |
3947be19 | 204 | { |
de0ed36a | 205 | int ret = 0; |
0c2c99b1 | 206 | |
4960e05e RW |
207 | if (mem->state != from_state_req) |
208 | return -EINVAL; | |
3947be19 | 209 | |
0c2c99b1 NF |
210 | if (to_state == MEM_OFFLINE) |
211 | mem->state = MEM_GOING_OFFLINE; | |
212 | ||
fa2be40f | 213 | ret = memory_block_action(mem->start_section_nr, to_state, |
bd5c2344 | 214 | mem->online_type, mem->nid); |
fa2be40f | 215 | |
b2c064b2 | 216 | mem->state = ret ? from_state_req : to_state; |
fa2be40f | 217 | |
4960e05e RW |
218 | return ret; |
219 | } | |
0c2c99b1 | 220 | |
fa2be40f | 221 | /* The device lock serializes operations on memory_subsys_[online|offline] */ |
4960e05e RW |
222 | static int memory_subsys_online(struct device *dev) |
223 | { | |
7315f0cc | 224 | struct memory_block *mem = to_memory_block(dev); |
4960e05e | 225 | int ret; |
3947be19 | 226 | |
fa2be40f SJ |
227 | if (mem->state == MEM_ONLINE) |
228 | return 0; | |
4960e05e | 229 | |
fa2be40f | 230 | /* |
efc978ad DH |
231 | * When called via device_online() without configuring the online_type, |
232 | * we want to default to MMOP_ONLINE. | |
fa2be40f | 233 | */ |
efc978ad | 234 | if (mem->online_type == MMOP_OFFLINE) |
956f8b44 | 235 | mem->online_type = MMOP_ONLINE; |
4960e05e | 236 | |
fa2be40f | 237 | ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE); |
efc978ad | 238 | mem->online_type = MMOP_OFFLINE; |
4960e05e | 239 | |
4960e05e RW |
240 | return ret; |
241 | } | |
242 | ||
4960e05e | 243 | static int memory_subsys_offline(struct device *dev) |
e90bdb7f | 244 | { |
7315f0cc | 245 | struct memory_block *mem = to_memory_block(dev); |
e90bdb7f | 246 | |
fa2be40f SJ |
247 | if (mem->state == MEM_OFFLINE) |
248 | return 0; | |
e90bdb7f | 249 | |
fa2be40f | 250 | return memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE); |
e90bdb7f | 251 | } |
4960e05e | 252 | |
3f8e9178 DH |
253 | static ssize_t state_store(struct device *dev, struct device_attribute *attr, |
254 | const char *buf, size_t count) | |
3947be19 | 255 | { |
4dc8207b | 256 | const int online_type = memhp_online_type_from_str(buf); |
7315f0cc | 257 | struct memory_block *mem = to_memory_block(dev); |
4dc8207b DH |
258 | int ret; |
259 | ||
260 | if (online_type < 0) | |
261 | return -EINVAL; | |
3947be19 | 262 | |
5e33bc41 RW |
263 | ret = lock_device_hotplug_sysfs(); |
264 | if (ret) | |
265 | return ret; | |
4960e05e | 266 | |
fa2be40f | 267 | switch (online_type) { |
4f7c6b49 TC |
268 | case MMOP_ONLINE_KERNEL: |
269 | case MMOP_ONLINE_MOVABLE: | |
956f8b44 | 270 | case MMOP_ONLINE: |
381eab4a | 271 | /* mem->online_type is protected by device_hotplug_lock */ |
fa2be40f SJ |
272 | mem->online_type = online_type; |
273 | ret = device_online(&mem->dev); | |
274 | break; | |
4f7c6b49 | 275 | case MMOP_OFFLINE: |
fa2be40f SJ |
276 | ret = device_offline(&mem->dev); |
277 | break; | |
278 | default: | |
279 | ret = -EINVAL; /* should never happen */ | |
4960e05e | 280 | } |
4960e05e RW |
281 | |
282 | unlock_device_hotplug(); | |
0c2c99b1 | 283 | |
d66ba15b | 284 | if (ret < 0) |
3947be19 | 285 | return ret; |
d66ba15b RA |
286 | if (ret) |
287 | return -EINVAL; | |
288 | ||
3947be19 DH |
289 | return count; |
290 | } | |
291 | ||
292 | /* | |
293 | * phys_device is a bad name for this. What I really want | |
294 | * is a way to differentiate between memory ranges that | |
295 | * are part of physical devices that constitute | |
296 | * a complete removable unit or fru. | |
297 | * i.e. do these ranges belong to the same physical device, | |
298 | * s.t. if I offline all of these sections I can then | |
299 | * remove the physical device? | |
300 | */ | |
3f8e9178 | 301 | static ssize_t phys_device_show(struct device *dev, |
10fbcf4c | 302 | struct device_attribute *attr, char *buf) |
3947be19 | 303 | { |
7315f0cc | 304 | struct memory_block *mem = to_memory_block(dev); |
948b3edb | 305 | |
aa838896 | 306 | return sysfs_emit(buf, "%d\n", mem->phys_device); |
3947be19 DH |
307 | } |
308 | ||
ed2f2400 | 309 | #ifdef CONFIG_MEMORY_HOTREMOVE |
973c3911 JP |
310 | static int print_allowed_zone(char *buf, int len, int nid, |
311 | unsigned long start_pfn, unsigned long nr_pages, | |
312 | int online_type, struct zone *default_zone) | |
e5e68930 MH |
313 | { |
314 | struct zone *zone; | |
315 | ||
e5e68930 | 316 | zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages); |
973c3911 JP |
317 | if (zone == default_zone) |
318 | return 0; | |
948b3edb | 319 | |
973c3911 | 320 | return sysfs_emit_at(buf, len, " %s", zone->name); |
e5e68930 MH |
321 | } |
322 | ||
3f8e9178 | 323 | static ssize_t valid_zones_show(struct device *dev, |
ed2f2400 ZZ |
324 | struct device_attribute *attr, char *buf) |
325 | { | |
326 | struct memory_block *mem = to_memory_block(dev); | |
f1dd2cd1 | 327 | unsigned long start_pfn = section_nr_to_pfn(mem->start_section_nr); |
ed2f2400 | 328 | unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block; |
e5e68930 | 329 | struct zone *default_zone; |
973c3911 | 330 | int len = 0; |
f1dd2cd1 | 331 | int nid; |
ed2f2400 | 332 | |
f1dd2cd1 MH |
333 | /* |
334 | * Check the existing zone. Make sure that we do that only on the | |
335 | * online nodes otherwise the page_zone is not reliable | |
336 | */ | |
337 | if (mem->state == MEM_ONLINE) { | |
4e8346d0 MZ |
338 | /* |
339 | * The block contains more than one zone can not be offlined. | |
340 | * This can happen e.g. for ZONE_DMA and ZONE_DMA32 | |
341 | */ | |
92917998 DH |
342 | default_zone = test_pages_in_a_zone(start_pfn, |
343 | start_pfn + nr_pages); | |
344 | if (!default_zone) | |
973c3911 JP |
345 | return sysfs_emit(buf, "%s\n", "none"); |
346 | len += sysfs_emit_at(buf, len, "%s", default_zone->name); | |
f1dd2cd1 | 347 | goto out; |
ed2f2400 ZZ |
348 | } |
349 | ||
4e8346d0 | 350 | nid = mem->nid; |
956f8b44 DH |
351 | default_zone = zone_for_pfn_range(MMOP_ONLINE, nid, start_pfn, |
352 | nr_pages); | |
ed2f2400 | 353 | |
973c3911 JP |
354 | len += sysfs_emit_at(buf, len, "%s", default_zone->name); |
355 | len += print_allowed_zone(buf, len, nid, start_pfn, nr_pages, | |
356 | MMOP_ONLINE_KERNEL, default_zone); | |
357 | len += print_allowed_zone(buf, len, nid, start_pfn, nr_pages, | |
358 | MMOP_ONLINE_MOVABLE, default_zone); | |
f1dd2cd1 | 359 | out: |
948b3edb | 360 | len += sysfs_emit_at(buf, len, "\n"); |
973c3911 | 361 | return len; |
ed2f2400 | 362 | } |
3f8e9178 | 363 | static DEVICE_ATTR_RO(valid_zones); |
ed2f2400 ZZ |
364 | #endif |
365 | ||
3f8e9178 DH |
366 | static DEVICE_ATTR_RO(phys_index); |
367 | static DEVICE_ATTR_RW(state); | |
368 | static DEVICE_ATTR_RO(phys_device); | |
369 | static DEVICE_ATTR_RO(removable); | |
3947be19 | 370 | |
3947be19 | 371 | /* |
f915fb7f | 372 | * Show the memory block size (shared by all memory blocks). |
3947be19 | 373 | */ |
3f8e9178 DH |
374 | static ssize_t block_size_bytes_show(struct device *dev, |
375 | struct device_attribute *attr, char *buf) | |
3947be19 | 376 | { |
aa838896 | 377 | return sysfs_emit(buf, "%lx\n", memory_block_size_bytes()); |
3947be19 DH |
378 | } |
379 | ||
3f8e9178 | 380 | static DEVICE_ATTR_RO(block_size_bytes); |
3947be19 | 381 | |
31bc3858 VK |
382 | /* |
383 | * Memory auto online policy. | |
384 | */ | |
385 | ||
3f8e9178 DH |
386 | static ssize_t auto_online_blocks_show(struct device *dev, |
387 | struct device_attribute *attr, char *buf) | |
31bc3858 | 388 | { |
aa838896 JP |
389 | return sysfs_emit(buf, "%s\n", |
390 | online_type_to_str[memhp_default_online_type]); | |
31bc3858 VK |
391 | } |
392 | ||
3f8e9178 DH |
393 | static ssize_t auto_online_blocks_store(struct device *dev, |
394 | struct device_attribute *attr, | |
395 | const char *buf, size_t count) | |
31bc3858 | 396 | { |
5f47adf7 DH |
397 | const int online_type = memhp_online_type_from_str(buf); |
398 | ||
399 | if (online_type < 0) | |
31bc3858 VK |
400 | return -EINVAL; |
401 | ||
5f47adf7 | 402 | memhp_default_online_type = online_type; |
31bc3858 VK |
403 | return count; |
404 | } | |
405 | ||
3f8e9178 | 406 | static DEVICE_ATTR_RW(auto_online_blocks); |
31bc3858 | 407 | |
3947be19 DH |
408 | /* |
409 | * Some architectures will have custom drivers to do this, and | |
410 | * will not need to do it from userspace. The fake hot-add code | |
411 | * as well as ppc64 will do all of their discovery in userspace | |
412 | * and will require this interface. | |
413 | */ | |
414 | #ifdef CONFIG_ARCH_MEMORY_PROBE | |
3f8e9178 DH |
415 | static ssize_t probe_store(struct device *dev, struct device_attribute *attr, |
416 | const char *buf, size_t count) | |
3947be19 DH |
417 | { |
418 | u64 phys_addr; | |
cb5490a5 | 419 | int nid, ret; |
61b94fea | 420 | unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block; |
3947be19 | 421 | |
b69deb2b ZZ |
422 | ret = kstrtoull(buf, 0, &phys_addr); |
423 | if (ret) | |
424 | return ret; | |
3947be19 | 425 | |
61b94fea AB |
426 | if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1)) |
427 | return -EINVAL; | |
428 | ||
8df1d0e4 DH |
429 | ret = lock_device_hotplug_sysfs(); |
430 | if (ret) | |
37803841 | 431 | return ret; |
8df1d0e4 | 432 | |
cb5490a5 | 433 | nid = memory_add_physaddr_to_nid(phys_addr); |
8df1d0e4 DH |
434 | ret = __add_memory(nid, phys_addr, |
435 | MIN_MEMORY_BLOCK_SIZE * sections_per_block); | |
6add7cd6 | 436 | |
cb5490a5 JA |
437 | if (ret) |
438 | goto out; | |
3947be19 | 439 | |
9f0af69b NK |
440 | ret = count; |
441 | out: | |
8df1d0e4 | 442 | unlock_device_hotplug(); |
9f0af69b | 443 | return ret; |
3947be19 | 444 | } |
3947be19 | 445 | |
3f8e9178 | 446 | static DEVICE_ATTR_WO(probe); |
3947be19 DH |
447 | #endif |
448 | ||
facb6011 AK |
449 | #ifdef CONFIG_MEMORY_FAILURE |
450 | /* | |
451 | * Support for offlining pages of memory | |
452 | */ | |
453 | ||
454 | /* Soft offline a page */ | |
3f8e9178 DH |
455 | static ssize_t soft_offline_page_store(struct device *dev, |
456 | struct device_attribute *attr, | |
457 | const char *buf, size_t count) | |
facb6011 AK |
458 | { |
459 | int ret; | |
460 | u64 pfn; | |
461 | if (!capable(CAP_SYS_ADMIN)) | |
462 | return -EPERM; | |
34da5e67 | 463 | if (kstrtoull(buf, 0, &pfn) < 0) |
facb6011 AK |
464 | return -EINVAL; |
465 | pfn >>= PAGE_SHIFT; | |
feec24a6 | 466 | ret = soft_offline_page(pfn, 0); |
facb6011 AK |
467 | return ret == 0 ? count : ret; |
468 | } | |
469 | ||
470 | /* Forcibly offline a page, including killing processes. */ | |
3f8e9178 DH |
471 | static ssize_t hard_offline_page_store(struct device *dev, |
472 | struct device_attribute *attr, | |
473 | const char *buf, size_t count) | |
facb6011 AK |
474 | { |
475 | int ret; | |
476 | u64 pfn; | |
477 | if (!capable(CAP_SYS_ADMIN)) | |
478 | return -EPERM; | |
34da5e67 | 479 | if (kstrtoull(buf, 0, &pfn) < 0) |
facb6011 AK |
480 | return -EINVAL; |
481 | pfn >>= PAGE_SHIFT; | |
83b57531 | 482 | ret = memory_failure(pfn, 0); |
facb6011 AK |
483 | return ret ? ret : count; |
484 | } | |
485 | ||
3f8e9178 DH |
486 | static DEVICE_ATTR_WO(soft_offline_page); |
487 | static DEVICE_ATTR_WO(hard_offline_page); | |
facb6011 AK |
488 | #endif |
489 | ||
3947be19 DH |
490 | /* |
491 | * Note that phys_device is optional. It is here to allow for | |
492 | * differentiation between which *physical* devices each | |
493 | * section belongs to... | |
494 | */ | |
bc32df00 HC |
495 | int __weak arch_get_memory_phys_device(unsigned long start_pfn) |
496 | { | |
497 | return 0; | |
498 | } | |
3947be19 | 499 | |
4fb6eabf SC |
500 | /* |
501 | * A reference for the returned memory block device is acquired. | |
502 | * | |
503 | * Called under device_hotplug_lock. | |
504 | */ | |
dd625285 | 505 | static struct memory_block *find_memory_block_by_id(unsigned long block_id) |
3947be19 | 506 | { |
4fb6eabf | 507 | struct memory_block *mem; |
3947be19 | 508 | |
4fb6eabf SC |
509 | mem = xa_load(&memory_blocks, block_id); |
510 | if (mem) | |
511 | get_device(&mem->dev); | |
512 | return mem; | |
db051a0d DH |
513 | } |
514 | ||
98383031 | 515 | /* |
4fb6eabf | 516 | * Called under device_hotplug_lock. |
98383031 RH |
517 | */ |
518 | struct memory_block *find_memory_block(struct mem_section *section) | |
519 | { | |
178bdbed | 520 | unsigned long block_id = memory_block_id(__section_nr(section)); |
dd625285 DH |
521 | |
522 | return find_memory_block_by_id(block_id); | |
98383031 RH |
523 | } |
524 | ||
96b2c0fc NF |
525 | static struct attribute *memory_memblk_attrs[] = { |
526 | &dev_attr_phys_index.attr, | |
96b2c0fc NF |
527 | &dev_attr_state.attr, |
528 | &dev_attr_phys_device.attr, | |
529 | &dev_attr_removable.attr, | |
ed2f2400 ZZ |
530 | #ifdef CONFIG_MEMORY_HOTREMOVE |
531 | &dev_attr_valid_zones.attr, | |
532 | #endif | |
96b2c0fc NF |
533 | NULL |
534 | }; | |
535 | ||
536 | static struct attribute_group memory_memblk_attr_group = { | |
537 | .attrs = memory_memblk_attrs, | |
538 | }; | |
539 | ||
540 | static const struct attribute_group *memory_memblk_attr_groups[] = { | |
541 | &memory_memblk_attr_group, | |
542 | NULL, | |
543 | }; | |
544 | ||
545 | /* | |
546 | * register_memory - Setup a sysfs device for a memory block | |
547 | */ | |
548 | static | |
549 | int register_memory(struct memory_block *memory) | |
550 | { | |
085aa2de AY |
551 | int ret; |
552 | ||
96b2c0fc NF |
553 | memory->dev.bus = &memory_subsys; |
554 | memory->dev.id = memory->start_section_nr / sections_per_block; | |
555 | memory->dev.release = memory_block_release; | |
556 | memory->dev.groups = memory_memblk_attr_groups; | |
f991fae5 | 557 | memory->dev.offline = memory->state == MEM_OFFLINE; |
96b2c0fc | 558 | |
085aa2de | 559 | ret = device_register(&memory->dev); |
4fb6eabf | 560 | if (ret) { |
085aa2de | 561 | put_device(&memory->dev); |
4fb6eabf SC |
562 | return ret; |
563 | } | |
564 | ret = xa_err(xa_store(&memory_blocks, memory->dev.id, memory, | |
565 | GFP_KERNEL)); | |
566 | if (ret) { | |
567 | put_device(&memory->dev); | |
568 | device_unregister(&memory->dev); | |
569 | } | |
085aa2de | 570 | return ret; |
96b2c0fc NF |
571 | } |
572 | ||
40ba2cde | 573 | static int init_memory_block(unsigned long block_id, unsigned long state) |
e4619c85 | 574 | { |
0c2c99b1 | 575 | struct memory_block *mem; |
e4619c85 NF |
576 | unsigned long start_pfn; |
577 | int ret = 0; | |
578 | ||
dd625285 | 579 | mem = find_memory_block_by_id(block_id); |
db051a0d DH |
580 | if (mem) { |
581 | put_device(&mem->dev); | |
582 | return -EEXIST; | |
583 | } | |
0c2c99b1 | 584 | mem = kzalloc(sizeof(*mem), GFP_KERNEL); |
e4619c85 NF |
585 | if (!mem) |
586 | return -ENOMEM; | |
587 | ||
18115825 | 588 | mem->start_section_nr = block_id * sections_per_block; |
e4619c85 | 589 | mem->state = state; |
d3360164 | 590 | start_pfn = section_nr_to_pfn(mem->start_section_nr); |
e4619c85 | 591 | mem->phys_device = arch_get_memory_phys_device(start_pfn); |
d84f2f5a | 592 | mem->nid = NUMA_NO_NODE; |
e4619c85 | 593 | |
0c2c99b1 | 594 | ret = register_memory(mem); |
0c2c99b1 | 595 | |
0c2c99b1 NF |
596 | return ret; |
597 | } | |
598 | ||
2491f0a2 | 599 | static int add_memory_block(unsigned long base_section_nr) |
0c2c99b1 | 600 | { |
68c3a6ac | 601 | int section_count = 0; |
2491f0a2 | 602 | unsigned long nr; |
0c2c99b1 | 603 | |
2491f0a2 DH |
604 | for (nr = base_section_nr; nr < base_section_nr + sections_per_block; |
605 | nr++) | |
606 | if (present_section_nr(nr)) | |
18115825 | 607 | section_count++; |
e4619c85 | 608 | |
cb5e39b8 SJ |
609 | if (section_count == 0) |
610 | return 0; | |
178bdbed | 611 | return init_memory_block(memory_block_id(base_section_nr), |
68c3a6ac | 612 | MEM_ONLINE); |
e4619c85 NF |
613 | } |
614 | ||
db051a0d DH |
615 | static void unregister_memory(struct memory_block *memory) |
616 | { | |
617 | if (WARN_ON_ONCE(memory->dev.bus != &memory_subsys)) | |
618 | return; | |
619 | ||
4fb6eabf SC |
620 | WARN_ON(xa_erase(&memory_blocks, memory->dev.id) == NULL); |
621 | ||
db051a0d DH |
622 | /* drop the ref. we got via find_memory_block() */ |
623 | put_device(&memory->dev); | |
624 | device_unregister(&memory->dev); | |
625 | } | |
626 | ||
4edd7cef | 627 | /* |
db051a0d DH |
628 | * Create memory block devices for the given memory area. Start and size |
629 | * have to be aligned to memory block granularity. Memory block devices | |
630 | * will be initialized as offline. | |
848e19ad DH |
631 | * |
632 | * Called under device_hotplug_lock. | |
4edd7cef | 633 | */ |
db051a0d | 634 | int create_memory_block_devices(unsigned long start, unsigned long size) |
4edd7cef | 635 | { |
90ec010f DH |
636 | const unsigned long start_block_id = pfn_to_block_id(PFN_DOWN(start)); |
637 | unsigned long end_block_id = pfn_to_block_id(PFN_DOWN(start + size)); | |
d7f80530 | 638 | struct memory_block *mem; |
db051a0d DH |
639 | unsigned long block_id; |
640 | int ret = 0; | |
b1eaef3d | 641 | |
db051a0d DH |
642 | if (WARN_ON_ONCE(!IS_ALIGNED(start, memory_block_size_bytes()) || |
643 | !IS_ALIGNED(size, memory_block_size_bytes()))) | |
644 | return -EINVAL; | |
b1eaef3d | 645 | |
db051a0d | 646 | for (block_id = start_block_id; block_id != end_block_id; block_id++) { |
40ba2cde | 647 | ret = init_memory_block(block_id, MEM_OFFLINE); |
d7f80530 | 648 | if (ret) |
db051a0d | 649 | break; |
db051a0d DH |
650 | } |
651 | if (ret) { | |
652 | end_block_id = block_id; | |
653 | for (block_id = start_block_id; block_id != end_block_id; | |
654 | block_id++) { | |
dd625285 | 655 | mem = find_memory_block_by_id(block_id); |
848e19ad DH |
656 | if (WARN_ON_ONCE(!mem)) |
657 | continue; | |
db051a0d DH |
658 | unregister_memory(mem); |
659 | } | |
d7f80530 | 660 | } |
b1eaef3d | 661 | return ret; |
4edd7cef DR |
662 | } |
663 | ||
4c4b7f9b DH |
664 | /* |
665 | * Remove memory block devices for the given memory area. Start and size | |
666 | * have to be aligned to memory block granularity. Memory block devices | |
667 | * have to be offline. | |
848e19ad DH |
668 | * |
669 | * Called under device_hotplug_lock. | |
4c4b7f9b DH |
670 | */ |
671 | void remove_memory_block_devices(unsigned long start, unsigned long size) | |
3947be19 | 672 | { |
90ec010f DH |
673 | const unsigned long start_block_id = pfn_to_block_id(PFN_DOWN(start)); |
674 | const unsigned long end_block_id = pfn_to_block_id(PFN_DOWN(start + size)); | |
3947be19 | 675 | struct memory_block *mem; |
90ec010f | 676 | unsigned long block_id; |
3947be19 | 677 | |
4c4b7f9b DH |
678 | if (WARN_ON_ONCE(!IS_ALIGNED(start, memory_block_size_bytes()) || |
679 | !IS_ALIGNED(size, memory_block_size_bytes()))) | |
cb7b3a36 DH |
680 | return; |
681 | ||
4c4b7f9b | 682 | for (block_id = start_block_id; block_id != end_block_id; block_id++) { |
dd625285 | 683 | mem = find_memory_block_by_id(block_id); |
4c4b7f9b DH |
684 | if (WARN_ON_ONCE(!mem)) |
685 | continue; | |
4c4b7f9b | 686 | unregister_memory_block_under_nodes(mem); |
0c2c99b1 | 687 | unregister_memory(mem); |
4c4b7f9b | 688 | } |
3947be19 DH |
689 | } |
690 | ||
6677e3ea YI |
691 | /* return true if the memory block is offlined, otherwise, return false */ |
692 | bool is_memblock_offlined(struct memory_block *mem) | |
693 | { | |
694 | return mem->state == MEM_OFFLINE; | |
695 | } | |
696 | ||
96b2c0fc NF |
697 | static struct attribute *memory_root_attrs[] = { |
698 | #ifdef CONFIG_ARCH_MEMORY_PROBE | |
699 | &dev_attr_probe.attr, | |
700 | #endif | |
701 | ||
702 | #ifdef CONFIG_MEMORY_FAILURE | |
703 | &dev_attr_soft_offline_page.attr, | |
704 | &dev_attr_hard_offline_page.attr, | |
705 | #endif | |
706 | ||
707 | &dev_attr_block_size_bytes.attr, | |
31bc3858 | 708 | &dev_attr_auto_online_blocks.attr, |
96b2c0fc NF |
709 | NULL |
710 | }; | |
711 | ||
712 | static struct attribute_group memory_root_attr_group = { | |
713 | .attrs = memory_root_attrs, | |
714 | }; | |
715 | ||
716 | static const struct attribute_group *memory_root_attr_groups[] = { | |
717 | &memory_root_attr_group, | |
718 | NULL, | |
719 | }; | |
720 | ||
3947be19 | 721 | /* |
848e19ad DH |
722 | * Initialize the sysfs support for memory devices. At the time this function |
723 | * is called, we cannot have concurrent creation/deletion of memory block | |
724 | * devices, the device_hotplug_lock is not needed. | |
3947be19 | 725 | */ |
902ce63b | 726 | void __init memory_dev_init(void) |
3947be19 | 727 | { |
3947be19 | 728 | int ret; |
2491f0a2 | 729 | unsigned long block_sz, nr; |
3947be19 | 730 | |
902ce63b DH |
731 | /* Validate the configured memory block size */ |
732 | block_sz = memory_block_size_bytes(); | |
733 | if (!is_power_of_2(block_sz) || block_sz < MIN_MEMORY_BLOCK_SIZE) | |
734 | panic("Memory block size not suitable: 0x%lx\n", block_sz); | |
735 | sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE; | |
736 | ||
96b2c0fc | 737 | ret = subsys_system_register(&memory_subsys, memory_root_attr_groups); |
28ec24e2 | 738 | if (ret) |
848e19ad | 739 | panic("%s() failed to register subsystem: %d\n", __func__, ret); |
3947be19 DH |
740 | |
741 | /* | |
742 | * Create entries for memory sections that were found | |
743 | * during boot and have been initialized | |
744 | */ | |
2491f0a2 DH |
745 | for (nr = 0; nr <= __highest_present_section_nr; |
746 | nr += sections_per_block) { | |
848e19ad DH |
747 | ret = add_memory_block(nr); |
748 | if (ret) | |
749 | panic("%s() failed to add memory block: %d\n", __func__, | |
750 | ret); | |
3947be19 | 751 | } |
3947be19 | 752 | } |
ea884641 DH |
753 | |
754 | /** | |
755 | * walk_memory_blocks - walk through all present memory blocks overlapped | |
756 | * by the range [start, start + size) | |
757 | * | |
758 | * @start: start address of the memory range | |
759 | * @size: size of the memory range | |
760 | * @arg: argument passed to func | |
761 | * @func: callback for each memory section walked | |
762 | * | |
763 | * This function walks through all present memory blocks overlapped by the | |
764 | * range [start, start + size), calling func on each memory block. | |
765 | * | |
766 | * In case func() returns an error, walking is aborted and the error is | |
767 | * returned. | |
4fb6eabf SC |
768 | * |
769 | * Called under device_hotplug_lock. | |
ea884641 DH |
770 | */ |
771 | int walk_memory_blocks(unsigned long start, unsigned long size, | |
772 | void *arg, walk_memory_blocks_func_t func) | |
773 | { | |
774 | const unsigned long start_block_id = phys_to_block_id(start); | |
775 | const unsigned long end_block_id = phys_to_block_id(start + size - 1); | |
776 | struct memory_block *mem; | |
777 | unsigned long block_id; | |
778 | int ret = 0; | |
779 | ||
dd625285 DH |
780 | if (!size) |
781 | return 0; | |
782 | ||
ea884641 | 783 | for (block_id = start_block_id; block_id <= end_block_id; block_id++) { |
dd625285 | 784 | mem = find_memory_block_by_id(block_id); |
ea884641 DH |
785 | if (!mem) |
786 | continue; | |
787 | ||
788 | ret = func(mem, arg); | |
789 | put_device(&mem->dev); | |
790 | if (ret) | |
791 | break; | |
792 | } | |
793 | return ret; | |
794 | } | |
2c91f8fc DH |
795 | |
796 | struct for_each_memory_block_cb_data { | |
797 | walk_memory_blocks_func_t func; | |
798 | void *arg; | |
799 | }; | |
800 | ||
801 | static int for_each_memory_block_cb(struct device *dev, void *data) | |
802 | { | |
803 | struct memory_block *mem = to_memory_block(dev); | |
804 | struct for_each_memory_block_cb_data *cb_data = data; | |
805 | ||
806 | return cb_data->func(mem, cb_data->arg); | |
807 | } | |
808 | ||
809 | /** | |
810 | * for_each_memory_block - walk through all present memory blocks | |
811 | * | |
812 | * @arg: argument passed to func | |
813 | * @func: callback for each memory block walked | |
814 | * | |
815 | * This function walks through all present memory blocks, calling func on | |
816 | * each memory block. | |
817 | * | |
818 | * In case func() returns an error, walking is aborted and the error is | |
819 | * returned. | |
820 | */ | |
821 | int for_each_memory_block(void *arg, walk_memory_blocks_func_t func) | |
822 | { | |
823 | struct for_each_memory_block_cb_data cb_data = { | |
824 | .func = func, | |
825 | .arg = arg, | |
826 | }; | |
827 | ||
828 | return bus_for_each_dev(&memory_subsys, NULL, &cb_data, | |
829 | for_each_memory_block_cb); | |
830 | } |