Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
3947be19 | 2 | /* |
10fbcf4c | 3 | * Memory subsystem support |
3947be19 DH |
4 | * |
5 | * Written by Matt Tolentino <matthew.e.tolentino@intel.com> | |
6 | * Dave Hansen <haveblue@us.ibm.com> | |
7 | * | |
8 | * This file provides the necessary infrastructure to represent | |
9 | * a SPARSEMEM-memory-model system's physical memory in /sysfs. | |
10 | * All arch-independent code that assumes MEMORY_HOTPLUG requires | |
11 | * SPARSEMEM should be contained here, or in mm/memory_hotplug.c. | |
12 | */ | |
13 | ||
3947be19 DH |
14 | #include <linux/module.h> |
15 | #include <linux/init.h> | |
3947be19 | 16 | #include <linux/topology.h> |
c59ede7b | 17 | #include <linux/capability.h> |
3947be19 DH |
18 | #include <linux/device.h> |
19 | #include <linux/memory.h> | |
3947be19 DH |
20 | #include <linux/memory_hotplug.h> |
21 | #include <linux/mm.h> | |
da19cbcf | 22 | #include <linux/mutex.h> |
9f1b16a5 | 23 | #include <linux/stat.h> |
5a0e3ad6 | 24 | #include <linux/slab.h> |
9f1b16a5 | 25 | |
60063497 | 26 | #include <linux/atomic.h> |
7c0f6ba6 | 27 | #include <linux/uaccess.h> |
3947be19 | 28 | |
2938ffbd NF |
29 | static DEFINE_MUTEX(mem_sysfs_mutex); |
30 | ||
3947be19 | 31 | #define MEMORY_CLASS_NAME "memory" |
0c2c99b1 | 32 | |
7315f0cc GZ |
33 | #define to_memory_block(dev) container_of(dev, struct memory_block, dev) |
34 | ||
0c2c99b1 NF |
35 | static int sections_per_block; |
36 | ||
37 | static inline int base_memory_block_id(int section_nr) | |
38 | { | |
39 | return section_nr / sections_per_block; | |
40 | } | |
3947be19 | 41 | |
4960e05e RW |
42 | static int memory_subsys_online(struct device *dev); |
43 | static int memory_subsys_offline(struct device *dev); | |
44 | ||
10fbcf4c | 45 | static struct bus_type memory_subsys = { |
af5ca3f4 | 46 | .name = MEMORY_CLASS_NAME, |
10fbcf4c | 47 | .dev_name = MEMORY_CLASS_NAME, |
4960e05e RW |
48 | .online = memory_subsys_online, |
49 | .offline = memory_subsys_offline, | |
3947be19 DH |
50 | }; |
51 | ||
e041c683 | 52 | static BLOCKING_NOTIFIER_HEAD(memory_chain); |
3947be19 | 53 | |
98a38ebd | 54 | int register_memory_notifier(struct notifier_block *nb) |
3947be19 | 55 | { |
2aeebca2 | 56 | return blocking_notifier_chain_register(&memory_chain, nb); |
3947be19 | 57 | } |
3c82c30c | 58 | EXPORT_SYMBOL(register_memory_notifier); |
3947be19 | 59 | |
98a38ebd | 60 | void unregister_memory_notifier(struct notifier_block *nb) |
3947be19 | 61 | { |
2aeebca2 | 62 | blocking_notifier_chain_unregister(&memory_chain, nb); |
3947be19 | 63 | } |
3c82c30c | 64 | EXPORT_SYMBOL(unregister_memory_notifier); |
3947be19 | 65 | |
925cc71e RJ |
66 | static ATOMIC_NOTIFIER_HEAD(memory_isolate_chain); |
67 | ||
68 | int register_memory_isolate_notifier(struct notifier_block *nb) | |
69 | { | |
70 | return atomic_notifier_chain_register(&memory_isolate_chain, nb); | |
71 | } | |
72 | EXPORT_SYMBOL(register_memory_isolate_notifier); | |
73 | ||
74 | void unregister_memory_isolate_notifier(struct notifier_block *nb) | |
75 | { | |
76 | atomic_notifier_chain_unregister(&memory_isolate_chain, nb); | |
77 | } | |
78 | EXPORT_SYMBOL(unregister_memory_isolate_notifier); | |
79 | ||
fa7194eb YI |
80 | static void memory_block_release(struct device *dev) |
81 | { | |
7315f0cc | 82 | struct memory_block *mem = to_memory_block(dev); |
fa7194eb YI |
83 | |
84 | kfree(mem); | |
85 | } | |
86 | ||
0c2c99b1 NF |
87 | unsigned long __weak memory_block_size_bytes(void) |
88 | { | |
89 | return MIN_MEMORY_BLOCK_SIZE; | |
90 | } | |
91 | ||
92 | static unsigned long get_memory_block_size(void) | |
93 | { | |
94 | unsigned long block_sz; | |
95 | ||
96 | block_sz = memory_block_size_bytes(); | |
97 | ||
98 | /* Validate blk_sz is a power of 2 and not less than section size */ | |
99 | if ((block_sz & (block_sz - 1)) || (block_sz < MIN_MEMORY_BLOCK_SIZE)) { | |
100 | WARN_ON(1); | |
101 | block_sz = MIN_MEMORY_BLOCK_SIZE; | |
102 | } | |
103 | ||
104 | return block_sz; | |
105 | } | |
106 | ||
3947be19 DH |
107 | /* |
108 | * use this as the physical section index that this memsection | |
109 | * uses. | |
110 | */ | |
111 | ||
10fbcf4c KS |
112 | static ssize_t show_mem_start_phys_index(struct device *dev, |
113 | struct device_attribute *attr, char *buf) | |
3947be19 | 114 | { |
7315f0cc | 115 | struct memory_block *mem = to_memory_block(dev); |
d3360164 NF |
116 | unsigned long phys_index; |
117 | ||
118 | phys_index = mem->start_section_nr / sections_per_block; | |
119 | return sprintf(buf, "%08lx\n", phys_index); | |
120 | } | |
121 | ||
5c755e9f BP |
122 | /* |
123 | * Show whether the section of memory is likely to be hot-removable | |
124 | */ | |
10fbcf4c KS |
125 | static ssize_t show_mem_removable(struct device *dev, |
126 | struct device_attribute *attr, char *buf) | |
5c755e9f | 127 | { |
0c2c99b1 NF |
128 | unsigned long i, pfn; |
129 | int ret = 1; | |
7315f0cc | 130 | struct memory_block *mem = to_memory_block(dev); |
5c755e9f | 131 | |
8b0662f2 MH |
132 | if (mem->state != MEM_ONLINE) |
133 | goto out; | |
134 | ||
0c2c99b1 | 135 | for (i = 0; i < sections_per_block; i++) { |
21ea9f5a RA |
136 | if (!present_section_nr(mem->start_section_nr + i)) |
137 | continue; | |
d3360164 | 138 | pfn = section_nr_to_pfn(mem->start_section_nr + i); |
0c2c99b1 NF |
139 | ret &= is_mem_section_removable(pfn, PAGES_PER_SECTION); |
140 | } | |
141 | ||
8b0662f2 | 142 | out: |
5c755e9f BP |
143 | return sprintf(buf, "%d\n", ret); |
144 | } | |
145 | ||
3947be19 DH |
146 | /* |
147 | * online, offline, going offline, etc. | |
148 | */ | |
10fbcf4c KS |
149 | static ssize_t show_mem_state(struct device *dev, |
150 | struct device_attribute *attr, char *buf) | |
3947be19 | 151 | { |
7315f0cc | 152 | struct memory_block *mem = to_memory_block(dev); |
3947be19 DH |
153 | ssize_t len = 0; |
154 | ||
155 | /* | |
156 | * We can probably put these states in a nice little array | |
157 | * so that they're not open-coded | |
158 | */ | |
159 | switch (mem->state) { | |
3d3af6af IC |
160 | case MEM_ONLINE: |
161 | len = sprintf(buf, "online\n"); | |
162 | break; | |
163 | case MEM_OFFLINE: | |
164 | len = sprintf(buf, "offline\n"); | |
165 | break; | |
166 | case MEM_GOING_OFFLINE: | |
167 | len = sprintf(buf, "going-offline\n"); | |
168 | break; | |
169 | default: | |
170 | len = sprintf(buf, "ERROR-UNKNOWN-%ld\n", | |
171 | mem->state); | |
172 | WARN_ON(1); | |
173 | break; | |
3947be19 DH |
174 | } |
175 | ||
176 | return len; | |
177 | } | |
178 | ||
7b78d335 | 179 | int memory_notify(unsigned long val, void *v) |
3947be19 | 180 | { |
e041c683 | 181 | return blocking_notifier_call_chain(&memory_chain, val, v); |
3947be19 DH |
182 | } |
183 | ||
925cc71e RJ |
184 | int memory_isolate_notify(unsigned long val, void *v) |
185 | { | |
186 | return atomic_notifier_call_chain(&memory_isolate_chain, val, v); | |
187 | } | |
188 | ||
2bbcb878 | 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 PT |
209 | if (!present_section_nr(section_nr)) { |
210 | pr_warn("section %ld pfn[%lx, %lx) not present", | |
211 | section_nr, pfn, pfn + PAGES_PER_SECTION); | |
212 | return false; | |
213 | } else if (!valid_section_nr(section_nr)) { | |
214 | pr_warn("section %ld pfn[%lx, %lx) no valid memmap", | |
215 | section_nr, pfn, pfn + PAGES_PER_SECTION); | |
216 | return false; | |
217 | } else if (online_section_nr(section_nr)) { | |
218 | pr_warn("section %ld pfn[%lx, %lx) is already online", | |
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. | |
30467e0b | 231 | * Must already be protected by mem_hotplug_begin(). |
3947be19 DH |
232 | */ |
233 | static int | |
511c2aba | 234 | memory_block_action(unsigned long phys_index, unsigned long action, int online_type) |
3947be19 | 235 | { |
a16cee10 | 236 | unsigned long start_pfn; |
5409d2cd | 237 | unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block; |
3947be19 | 238 | int ret; |
3947be19 | 239 | |
19c07d5e | 240 | start_pfn = section_nr_to_pfn(phys_index); |
de0ed36a | 241 | |
3947be19 | 242 | switch (action) { |
3d3af6af | 243 | case MEM_ONLINE: |
b77eab70 | 244 | if (!pages_correctly_probed(start_pfn)) |
3d3af6af IC |
245 | return -EBUSY; |
246 | ||
247 | ret = online_pages(start_pfn, nr_pages, online_type); | |
248 | break; | |
249 | case MEM_OFFLINE: | |
250 | ret = offline_pages(start_pfn, nr_pages); | |
251 | break; | |
252 | default: | |
253 | WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: " | |
254 | "%ld\n", __func__, phys_index, action, action); | |
255 | ret = -EINVAL; | |
3947be19 | 256 | } |
3947be19 DH |
257 | |
258 | return ret; | |
259 | } | |
260 | ||
dc18d706 | 261 | static int memory_block_change_state(struct memory_block *mem, |
fa2be40f | 262 | unsigned long to_state, unsigned long from_state_req) |
3947be19 | 263 | { |
de0ed36a | 264 | int ret = 0; |
0c2c99b1 | 265 | |
4960e05e RW |
266 | if (mem->state != from_state_req) |
267 | return -EINVAL; | |
3947be19 | 268 | |
0c2c99b1 NF |
269 | if (to_state == MEM_OFFLINE) |
270 | mem->state = MEM_GOING_OFFLINE; | |
271 | ||
fa2be40f SJ |
272 | ret = memory_block_action(mem->start_section_nr, to_state, |
273 | mem->online_type); | |
274 | ||
b2c064b2 | 275 | mem->state = ret ? from_state_req : to_state; |
fa2be40f | 276 | |
4960e05e RW |
277 | return ret; |
278 | } | |
0c2c99b1 | 279 | |
fa2be40f | 280 | /* The device lock serializes operations on memory_subsys_[online|offline] */ |
4960e05e RW |
281 | static int memory_subsys_online(struct device *dev) |
282 | { | |
7315f0cc | 283 | struct memory_block *mem = to_memory_block(dev); |
4960e05e | 284 | int ret; |
3947be19 | 285 | |
fa2be40f SJ |
286 | if (mem->state == MEM_ONLINE) |
287 | return 0; | |
4960e05e | 288 | |
fa2be40f SJ |
289 | /* |
290 | * If we are called from store_mem_state(), online_type will be | |
291 | * set >= 0 Otherwise we were called from the device online | |
292 | * attribute and need to set the online_type. | |
293 | */ | |
294 | if (mem->online_type < 0) | |
4f7c6b49 | 295 | mem->online_type = MMOP_ONLINE_KEEP; |
4960e05e | 296 | |
30467e0b | 297 | /* Already under protection of mem_hotplug_begin() */ |
fa2be40f | 298 | ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE); |
4960e05e | 299 | |
fa2be40f SJ |
300 | /* clear online_type */ |
301 | mem->online_type = -1; | |
4960e05e | 302 | |
4960e05e RW |
303 | return ret; |
304 | } | |
305 | ||
4960e05e | 306 | static int memory_subsys_offline(struct device *dev) |
e90bdb7f | 307 | { |
7315f0cc | 308 | struct memory_block *mem = to_memory_block(dev); |
e90bdb7f | 309 | |
fa2be40f SJ |
310 | if (mem->state == MEM_OFFLINE) |
311 | return 0; | |
e90bdb7f | 312 | |
26bbe7ef SJ |
313 | /* Can't offline block with non-present sections */ |
314 | if (mem->section_count != sections_per_block) | |
315 | return -EINVAL; | |
316 | ||
fa2be40f | 317 | return memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE); |
e90bdb7f | 318 | } |
4960e05e | 319 | |
3947be19 | 320 | static ssize_t |
10fbcf4c KS |
321 | store_mem_state(struct device *dev, |
322 | struct device_attribute *attr, const char *buf, size_t count) | |
3947be19 | 323 | { |
7315f0cc | 324 | struct memory_block *mem = to_memory_block(dev); |
fa2be40f | 325 | int ret, online_type; |
3947be19 | 326 | |
5e33bc41 RW |
327 | ret = lock_device_hotplug_sysfs(); |
328 | if (ret) | |
329 | return ret; | |
4960e05e | 330 | |
1f6a6cc8 | 331 | if (sysfs_streq(buf, "online_kernel")) |
4f7c6b49 | 332 | online_type = MMOP_ONLINE_KERNEL; |
1f6a6cc8 | 333 | else if (sysfs_streq(buf, "online_movable")) |
4f7c6b49 | 334 | online_type = MMOP_ONLINE_MOVABLE; |
1f6a6cc8 | 335 | else if (sysfs_streq(buf, "online")) |
4f7c6b49 | 336 | online_type = MMOP_ONLINE_KEEP; |
1f6a6cc8 | 337 | else if (sysfs_streq(buf, "offline")) |
4f7c6b49 | 338 | online_type = MMOP_OFFLINE; |
a37f8630 YI |
339 | else { |
340 | ret = -EINVAL; | |
341 | goto err; | |
342 | } | |
fa2be40f | 343 | |
30467e0b DR |
344 | /* |
345 | * Memory hotplug needs to hold mem_hotplug_begin() for probe to find | |
346 | * the correct memory block to online before doing device_online(dev), | |
347 | * which will take dev->mutex. Take the lock early to prevent an | |
348 | * inversion, memory_subsys_online() callbacks will be implemented by | |
349 | * assuming it's already protected. | |
350 | */ | |
351 | mem_hotplug_begin(); | |
352 | ||
fa2be40f | 353 | switch (online_type) { |
4f7c6b49 TC |
354 | case MMOP_ONLINE_KERNEL: |
355 | case MMOP_ONLINE_MOVABLE: | |
356 | case MMOP_ONLINE_KEEP: | |
fa2be40f SJ |
357 | mem->online_type = online_type; |
358 | ret = device_online(&mem->dev); | |
359 | break; | |
4f7c6b49 | 360 | case MMOP_OFFLINE: |
fa2be40f SJ |
361 | ret = device_offline(&mem->dev); |
362 | break; | |
363 | default: | |
364 | ret = -EINVAL; /* should never happen */ | |
4960e05e | 365 | } |
4960e05e | 366 | |
30467e0b | 367 | mem_hotplug_done(); |
a37f8630 | 368 | err: |
4960e05e | 369 | unlock_device_hotplug(); |
0c2c99b1 | 370 | |
d66ba15b | 371 | if (ret < 0) |
3947be19 | 372 | return ret; |
d66ba15b RA |
373 | if (ret) |
374 | return -EINVAL; | |
375 | ||
3947be19 DH |
376 | return count; |
377 | } | |
378 | ||
379 | /* | |
380 | * phys_device is a bad name for this. What I really want | |
381 | * is a way to differentiate between memory ranges that | |
382 | * are part of physical devices that constitute | |
383 | * a complete removable unit or fru. | |
384 | * i.e. do these ranges belong to the same physical device, | |
385 | * s.t. if I offline all of these sections I can then | |
386 | * remove the physical device? | |
387 | */ | |
10fbcf4c KS |
388 | static ssize_t show_phys_device(struct device *dev, |
389 | struct device_attribute *attr, char *buf) | |
3947be19 | 390 | { |
7315f0cc | 391 | struct memory_block *mem = to_memory_block(dev); |
3947be19 DH |
392 | return sprintf(buf, "%d\n", mem->phys_device); |
393 | } | |
394 | ||
ed2f2400 | 395 | #ifdef CONFIG_MEMORY_HOTREMOVE |
e5e68930 MH |
396 | static void print_allowed_zone(char *buf, int nid, unsigned long start_pfn, |
397 | unsigned long nr_pages, int online_type, | |
398 | struct zone *default_zone) | |
399 | { | |
400 | struct zone *zone; | |
401 | ||
e5e68930 MH |
402 | zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages); |
403 | if (zone != default_zone) { | |
404 | strcat(buf, " "); | |
405 | strcat(buf, zone->name); | |
406 | } | |
407 | } | |
408 | ||
ed2f2400 ZZ |
409 | static ssize_t show_valid_zones(struct device *dev, |
410 | struct device_attribute *attr, char *buf) | |
411 | { | |
412 | struct memory_block *mem = to_memory_block(dev); | |
f1dd2cd1 | 413 | unsigned long start_pfn = section_nr_to_pfn(mem->start_section_nr); |
ed2f2400 | 414 | unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block; |
f1dd2cd1 | 415 | unsigned long valid_start_pfn, valid_end_pfn; |
e5e68930 | 416 | struct zone *default_zone; |
f1dd2cd1 | 417 | int nid; |
ed2f2400 | 418 | |
f1dd2cd1 MH |
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, &valid_start_pfn, &valid_end_pfn)) | |
ed2f2400 ZZ |
424 | return sprintf(buf, "none\n"); |
425 | ||
f1dd2cd1 MH |
426 | start_pfn = valid_start_pfn; |
427 | nr_pages = valid_end_pfn - start_pfn; | |
ed2f2400 | 428 | |
f1dd2cd1 MH |
429 | /* |
430 | * Check the existing zone. Make sure that we do that only on the | |
431 | * online nodes otherwise the page_zone is not reliable | |
432 | */ | |
433 | if (mem->state == MEM_ONLINE) { | |
434 | strcat(buf, page_zone(pfn_to_page(start_pfn))->name); | |
435 | goto out; | |
ed2f2400 ZZ |
436 | } |
437 | ||
f1dd2cd1 | 438 | nid = pfn_to_nid(start_pfn); |
e5e68930 MH |
439 | default_zone = zone_for_pfn_range(MMOP_ONLINE_KEEP, nid, start_pfn, nr_pages); |
440 | strcat(buf, default_zone->name); | |
ed2f2400 | 441 | |
e5e68930 MH |
442 | print_allowed_zone(buf, nid, start_pfn, nr_pages, MMOP_ONLINE_KERNEL, |
443 | default_zone); | |
444 | print_allowed_zone(buf, nid, start_pfn, nr_pages, MMOP_ONLINE_MOVABLE, | |
445 | default_zone); | |
f1dd2cd1 | 446 | out: |
a371d9f1 RA |
447 | strcat(buf, "\n"); |
448 | ||
449 | return strlen(buf); | |
ed2f2400 ZZ |
450 | } |
451 | static DEVICE_ATTR(valid_zones, 0444, show_valid_zones, NULL); | |
452 | #endif | |
453 | ||
10fbcf4c | 454 | static DEVICE_ATTR(phys_index, 0444, show_mem_start_phys_index, NULL); |
10fbcf4c KS |
455 | static DEVICE_ATTR(state, 0644, show_mem_state, store_mem_state); |
456 | static DEVICE_ATTR(phys_device, 0444, show_phys_device, NULL); | |
457 | static DEVICE_ATTR(removable, 0444, show_mem_removable, NULL); | |
3947be19 | 458 | |
3947be19 DH |
459 | /* |
460 | * Block size attribute stuff | |
461 | */ | |
462 | static ssize_t | |
10fbcf4c | 463 | print_block_size(struct device *dev, struct device_attribute *attr, |
8564a6c1 | 464 | char *buf) |
3947be19 | 465 | { |
0c2c99b1 | 466 | return sprintf(buf, "%lx\n", get_memory_block_size()); |
3947be19 DH |
467 | } |
468 | ||
10fbcf4c | 469 | static DEVICE_ATTR(block_size_bytes, 0444, print_block_size, NULL); |
3947be19 | 470 | |
31bc3858 VK |
471 | /* |
472 | * Memory auto online policy. | |
473 | */ | |
474 | ||
475 | static ssize_t | |
476 | show_auto_online_blocks(struct device *dev, struct device_attribute *attr, | |
477 | char *buf) | |
478 | { | |
479 | if (memhp_auto_online) | |
480 | return sprintf(buf, "online\n"); | |
481 | else | |
482 | return sprintf(buf, "offline\n"); | |
483 | } | |
484 | ||
485 | static ssize_t | |
486 | store_auto_online_blocks(struct device *dev, struct device_attribute *attr, | |
487 | const char *buf, size_t count) | |
488 | { | |
489 | if (sysfs_streq(buf, "online")) | |
490 | memhp_auto_online = true; | |
491 | else if (sysfs_streq(buf, "offline")) | |
492 | memhp_auto_online = false; | |
493 | else | |
494 | return -EINVAL; | |
495 | ||
496 | return count; | |
497 | } | |
498 | ||
499 | static DEVICE_ATTR(auto_online_blocks, 0644, show_auto_online_blocks, | |
500 | store_auto_online_blocks); | |
501 | ||
3947be19 DH |
502 | /* |
503 | * Some architectures will have custom drivers to do this, and | |
504 | * will not need to do it from userspace. The fake hot-add code | |
505 | * as well as ppc64 will do all of their discovery in userspace | |
506 | * and will require this interface. | |
507 | */ | |
508 | #ifdef CONFIG_ARCH_MEMORY_PROBE | |
509 | static ssize_t | |
10fbcf4c | 510 | memory_probe_store(struct device *dev, struct device_attribute *attr, |
28812fe1 | 511 | const char *buf, size_t count) |
3947be19 DH |
512 | { |
513 | u64 phys_addr; | |
cb5490a5 | 514 | int nid, ret; |
61b94fea | 515 | unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block; |
3947be19 | 516 | |
b69deb2b ZZ |
517 | ret = kstrtoull(buf, 0, &phys_addr); |
518 | if (ret) | |
519 | return ret; | |
3947be19 | 520 | |
61b94fea AB |
521 | if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1)) |
522 | return -EINVAL; | |
523 | ||
cb5490a5 JA |
524 | nid = memory_add_physaddr_to_nid(phys_addr); |
525 | ret = add_memory(nid, phys_addr, | |
526 | MIN_MEMORY_BLOCK_SIZE * sections_per_block); | |
6add7cd6 | 527 | |
cb5490a5 JA |
528 | if (ret) |
529 | goto out; | |
3947be19 | 530 | |
9f0af69b NK |
531 | ret = count; |
532 | out: | |
533 | return ret; | |
3947be19 | 534 | } |
3947be19 | 535 | |
96b2c0fc | 536 | static DEVICE_ATTR(probe, S_IWUSR, NULL, memory_probe_store); |
3947be19 DH |
537 | #endif |
538 | ||
facb6011 AK |
539 | #ifdef CONFIG_MEMORY_FAILURE |
540 | /* | |
541 | * Support for offlining pages of memory | |
542 | */ | |
543 | ||
544 | /* Soft offline a page */ | |
545 | static ssize_t | |
10fbcf4c KS |
546 | store_soft_offline_page(struct device *dev, |
547 | struct device_attribute *attr, | |
28812fe1 | 548 | const char *buf, size_t count) |
facb6011 AK |
549 | { |
550 | int ret; | |
551 | u64 pfn; | |
552 | if (!capable(CAP_SYS_ADMIN)) | |
553 | return -EPERM; | |
34da5e67 | 554 | if (kstrtoull(buf, 0, &pfn) < 0) |
facb6011 AK |
555 | return -EINVAL; |
556 | pfn >>= PAGE_SHIFT; | |
557 | if (!pfn_valid(pfn)) | |
558 | return -ENXIO; | |
559 | ret = soft_offline_page(pfn_to_page(pfn), 0); | |
560 | return ret == 0 ? count : ret; | |
561 | } | |
562 | ||
563 | /* Forcibly offline a page, including killing processes. */ | |
564 | static ssize_t | |
10fbcf4c KS |
565 | store_hard_offline_page(struct device *dev, |
566 | struct device_attribute *attr, | |
28812fe1 | 567 | const char *buf, size_t count) |
facb6011 AK |
568 | { |
569 | int ret; | |
570 | u64 pfn; | |
571 | if (!capable(CAP_SYS_ADMIN)) | |
572 | return -EPERM; | |
34da5e67 | 573 | if (kstrtoull(buf, 0, &pfn) < 0) |
facb6011 AK |
574 | return -EINVAL; |
575 | pfn >>= PAGE_SHIFT; | |
83b57531 | 576 | ret = memory_failure(pfn, 0); |
facb6011 AK |
577 | return ret ? ret : count; |
578 | } | |
579 | ||
74fef7a8 FB |
580 | static DEVICE_ATTR(soft_offline_page, S_IWUSR, NULL, store_soft_offline_page); |
581 | static DEVICE_ATTR(hard_offline_page, S_IWUSR, NULL, store_hard_offline_page); | |
facb6011 AK |
582 | #endif |
583 | ||
3947be19 DH |
584 | /* |
585 | * Note that phys_device is optional. It is here to allow for | |
586 | * differentiation between which *physical* devices each | |
587 | * section belongs to... | |
588 | */ | |
bc32df00 HC |
589 | int __weak arch_get_memory_phys_device(unsigned long start_pfn) |
590 | { | |
591 | return 0; | |
592 | } | |
3947be19 | 593 | |
10fbcf4c KS |
594 | /* |
595 | * A reference for the returned object is held and the reference for the | |
596 | * hinted object is released. | |
597 | */ | |
98383031 RH |
598 | struct memory_block *find_memory_block_hinted(struct mem_section *section, |
599 | struct memory_block *hint) | |
3947be19 | 600 | { |
0c2c99b1 | 601 | int block_id = base_memory_block_id(__section_nr(section)); |
10fbcf4c KS |
602 | struct device *hintdev = hint ? &hint->dev : NULL; |
603 | struct device *dev; | |
3947be19 | 604 | |
10fbcf4c KS |
605 | dev = subsys_find_device_by_id(&memory_subsys, block_id, hintdev); |
606 | if (hint) | |
607 | put_device(&hint->dev); | |
608 | if (!dev) | |
3947be19 | 609 | return NULL; |
7315f0cc | 610 | return to_memory_block(dev); |
3947be19 DH |
611 | } |
612 | ||
98383031 RH |
613 | /* |
614 | * For now, we have a linear search to go find the appropriate | |
615 | * memory_block corresponding to a particular phys_index. If | |
616 | * this gets to be a real problem, we can always use a radix | |
617 | * tree or something here. | |
618 | * | |
10fbcf4c | 619 | * This could be made generic for all device subsystems. |
98383031 RH |
620 | */ |
621 | struct memory_block *find_memory_block(struct mem_section *section) | |
622 | { | |
623 | return find_memory_block_hinted(section, NULL); | |
624 | } | |
625 | ||
96b2c0fc NF |
626 | static struct attribute *memory_memblk_attrs[] = { |
627 | &dev_attr_phys_index.attr, | |
96b2c0fc NF |
628 | &dev_attr_state.attr, |
629 | &dev_attr_phys_device.attr, | |
630 | &dev_attr_removable.attr, | |
ed2f2400 ZZ |
631 | #ifdef CONFIG_MEMORY_HOTREMOVE |
632 | &dev_attr_valid_zones.attr, | |
633 | #endif | |
96b2c0fc NF |
634 | NULL |
635 | }; | |
636 | ||
637 | static struct attribute_group memory_memblk_attr_group = { | |
638 | .attrs = memory_memblk_attrs, | |
639 | }; | |
640 | ||
641 | static const struct attribute_group *memory_memblk_attr_groups[] = { | |
642 | &memory_memblk_attr_group, | |
643 | NULL, | |
644 | }; | |
645 | ||
646 | /* | |
647 | * register_memory - Setup a sysfs device for a memory block | |
648 | */ | |
649 | static | |
650 | int register_memory(struct memory_block *memory) | |
651 | { | |
085aa2de AY |
652 | int ret; |
653 | ||
96b2c0fc NF |
654 | memory->dev.bus = &memory_subsys; |
655 | memory->dev.id = memory->start_section_nr / sections_per_block; | |
656 | memory->dev.release = memory_block_release; | |
657 | memory->dev.groups = memory_memblk_attr_groups; | |
f991fae5 | 658 | memory->dev.offline = memory->state == MEM_OFFLINE; |
96b2c0fc | 659 | |
085aa2de AY |
660 | ret = device_register(&memory->dev); |
661 | if (ret) | |
662 | put_device(&memory->dev); | |
663 | ||
664 | return ret; | |
96b2c0fc NF |
665 | } |
666 | ||
0c2c99b1 NF |
667 | static int init_memory_block(struct memory_block **memory, |
668 | struct mem_section *section, unsigned long state) | |
e4619c85 | 669 | { |
0c2c99b1 | 670 | struct memory_block *mem; |
e4619c85 | 671 | unsigned long start_pfn; |
0c2c99b1 | 672 | int scn_nr; |
e4619c85 NF |
673 | int ret = 0; |
674 | ||
0c2c99b1 | 675 | mem = kzalloc(sizeof(*mem), GFP_KERNEL); |
e4619c85 NF |
676 | if (!mem) |
677 | return -ENOMEM; | |
678 | ||
0c2c99b1 | 679 | scn_nr = __section_nr(section); |
d3360164 NF |
680 | mem->start_section_nr = |
681 | base_memory_block_id(scn_nr) * sections_per_block; | |
682 | mem->end_section_nr = mem->start_section_nr + sections_per_block - 1; | |
e4619c85 | 683 | mem->state = state; |
d3360164 | 684 | start_pfn = section_nr_to_pfn(mem->start_section_nr); |
e4619c85 NF |
685 | mem->phys_device = arch_get_memory_phys_device(start_pfn); |
686 | ||
0c2c99b1 | 687 | ret = register_memory(mem); |
0c2c99b1 NF |
688 | |
689 | *memory = mem; | |
690 | return ret; | |
691 | } | |
692 | ||
cb5e39b8 | 693 | static int add_memory_block(int base_section_nr) |
0c2c99b1 | 694 | { |
cb5e39b8 SJ |
695 | struct memory_block *mem; |
696 | int i, ret, section_count = 0, section_nr; | |
0c2c99b1 | 697 | |
cb5e39b8 SJ |
698 | for (i = base_section_nr; |
699 | (i < base_section_nr + sections_per_block) && i < NR_MEM_SECTIONS; | |
700 | i++) { | |
701 | if (!present_section_nr(i)) | |
702 | continue; | |
703 | if (section_count == 0) | |
704 | section_nr = i; | |
705 | section_count++; | |
e4619c85 NF |
706 | } |
707 | ||
cb5e39b8 SJ |
708 | if (section_count == 0) |
709 | return 0; | |
710 | ret = init_memory_block(&mem, __nr_to_section(section_nr), MEM_ONLINE); | |
711 | if (ret) | |
712 | return ret; | |
713 | mem->section_count = section_count; | |
714 | return 0; | |
e4619c85 NF |
715 | } |
716 | ||
4edd7cef DR |
717 | /* |
718 | * need an interface for the VM to add new memory regions, | |
719 | * but without onlining it. | |
720 | */ | |
fc44f7f9 | 721 | int hotplug_memory_register(int nid, struct mem_section *section) |
4edd7cef | 722 | { |
d7f80530 SJ |
723 | int ret = 0; |
724 | struct memory_block *mem; | |
b1eaef3d SJ |
725 | |
726 | mutex_lock(&mem_sysfs_mutex); | |
b1eaef3d | 727 | |
d7f80530 SJ |
728 | mem = find_memory_block(section); |
729 | if (mem) { | |
730 | mem->section_count++; | |
731 | put_device(&mem->dev); | |
732 | } else { | |
733 | ret = init_memory_block(&mem, section, MEM_OFFLINE); | |
734 | if (ret) | |
735 | goto out; | |
56c6b5d3 | 736 | mem->section_count++; |
d7f80530 SJ |
737 | } |
738 | ||
d7f80530 SJ |
739 | out: |
740 | mutex_unlock(&mem_sysfs_mutex); | |
b1eaef3d | 741 | return ret; |
4edd7cef DR |
742 | } |
743 | ||
744 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
745 | static void | |
746 | unregister_memory(struct memory_block *memory) | |
747 | { | |
748 | BUG_ON(memory->dev.bus != &memory_subsys); | |
749 | ||
750 | /* drop the ref. we got in remove_memory_block() */ | |
df2b717c | 751 | put_device(&memory->dev); |
4edd7cef DR |
752 | device_unregister(&memory->dev); |
753 | } | |
754 | ||
cc292b0b | 755 | static int remove_memory_section(unsigned long node_id, |
4edd7cef | 756 | struct mem_section *section, int phys_device) |
3947be19 DH |
757 | { |
758 | struct memory_block *mem; | |
759 | ||
2938ffbd | 760 | mutex_lock(&mem_sysfs_mutex); |
1b862aec MH |
761 | |
762 | /* | |
763 | * Some users of the memory hotplug do not want/need memblock to | |
764 | * track all sections. Skip over those. | |
765 | */ | |
3947be19 | 766 | mem = find_memory_block(section); |
1b862aec MH |
767 | if (!mem) |
768 | goto out_unlock; | |
769 | ||
d3360164 | 770 | unregister_mem_sect_under_nodes(mem, __section_nr(section)); |
07681215 NF |
771 | |
772 | mem->section_count--; | |
96b2c0fc | 773 | if (mem->section_count == 0) |
0c2c99b1 | 774 | unregister_memory(mem); |
96b2c0fc | 775 | else |
df2b717c | 776 | put_device(&mem->dev); |
3947be19 | 777 | |
1b862aec | 778 | out_unlock: |
2938ffbd | 779 | mutex_unlock(&mem_sysfs_mutex); |
3947be19 DH |
780 | return 0; |
781 | } | |
782 | ||
3947be19 DH |
783 | int unregister_memory_section(struct mem_section *section) |
784 | { | |
540557b9 | 785 | if (!present_section(section)) |
3947be19 DH |
786 | return -EINVAL; |
787 | ||
cc292b0b | 788 | return remove_memory_section(0, section, 0); |
3947be19 | 789 | } |
4edd7cef | 790 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
3947be19 | 791 | |
6677e3ea YI |
792 | /* return true if the memory block is offlined, otherwise, return false */ |
793 | bool is_memblock_offlined(struct memory_block *mem) | |
794 | { | |
795 | return mem->state == MEM_OFFLINE; | |
796 | } | |
797 | ||
96b2c0fc NF |
798 | static struct attribute *memory_root_attrs[] = { |
799 | #ifdef CONFIG_ARCH_MEMORY_PROBE | |
800 | &dev_attr_probe.attr, | |
801 | #endif | |
802 | ||
803 | #ifdef CONFIG_MEMORY_FAILURE | |
804 | &dev_attr_soft_offline_page.attr, | |
805 | &dev_attr_hard_offline_page.attr, | |
806 | #endif | |
807 | ||
808 | &dev_attr_block_size_bytes.attr, | |
31bc3858 | 809 | &dev_attr_auto_online_blocks.attr, |
96b2c0fc NF |
810 | NULL |
811 | }; | |
812 | ||
813 | static struct attribute_group memory_root_attr_group = { | |
814 | .attrs = memory_root_attrs, | |
815 | }; | |
816 | ||
817 | static const struct attribute_group *memory_root_attr_groups[] = { | |
818 | &memory_root_attr_group, | |
819 | NULL, | |
820 | }; | |
821 | ||
3947be19 DH |
822 | /* |
823 | * Initialize the sysfs support for memory devices... | |
824 | */ | |
825 | int __init memory_dev_init(void) | |
826 | { | |
827 | unsigned int i; | |
828 | int ret; | |
28ec24e2 | 829 | int err; |
0c2c99b1 | 830 | unsigned long block_sz; |
3947be19 | 831 | |
96b2c0fc | 832 | ret = subsys_system_register(&memory_subsys, memory_root_attr_groups); |
28ec24e2 AM |
833 | if (ret) |
834 | goto out; | |
3947be19 | 835 | |
0c2c99b1 NF |
836 | block_sz = get_memory_block_size(); |
837 | sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE; | |
838 | ||
3947be19 DH |
839 | /* |
840 | * Create entries for memory sections that were found | |
841 | * during boot and have been initialized | |
842 | */ | |
b1eaef3d | 843 | mutex_lock(&mem_sysfs_mutex); |
bc8755ba WY |
844 | for (i = 0; i <= __highest_present_section_nr; |
845 | i += sections_per_block) { | |
cb5e39b8 | 846 | err = add_memory_block(i); |
28ec24e2 AM |
847 | if (!ret) |
848 | ret = err; | |
3947be19 | 849 | } |
b1eaef3d | 850 | mutex_unlock(&mem_sysfs_mutex); |
3947be19 | 851 | |
28ec24e2 AM |
852 | out: |
853 | if (ret) | |
2b3a302a | 854 | printk(KERN_ERR "%s() failed: %d\n", __func__, ret); |
3947be19 DH |
855 | return ret; |
856 | } |