Commit | Line | Data |
---|---|---|
3947be19 | 1 | /* |
10fbcf4c | 2 | * Memory subsystem support |
3947be19 DH |
3 | * |
4 | * Written by Matt Tolentino <matthew.e.tolentino@intel.com> | |
5 | * Dave Hansen <haveblue@us.ibm.com> | |
6 | * | |
7 | * This file provides the necessary infrastructure to represent | |
8 | * a SPARSEMEM-memory-model system's physical memory in /sysfs. | |
9 | * All arch-independent code that assumes MEMORY_HOTPLUG requires | |
10 | * SPARSEMEM should be contained here, or in mm/memory_hotplug.c. | |
11 | */ | |
12 | ||
3947be19 DH |
13 | #include <linux/module.h> |
14 | #include <linux/init.h> | |
3947be19 | 15 | #include <linux/topology.h> |
c59ede7b | 16 | #include <linux/capability.h> |
3947be19 DH |
17 | #include <linux/device.h> |
18 | #include <linux/memory.h> | |
3947be19 DH |
19 | #include <linux/memory_hotplug.h> |
20 | #include <linux/mm.h> | |
da19cbcf | 21 | #include <linux/mutex.h> |
9f1b16a5 | 22 | #include <linux/stat.h> |
5a0e3ad6 | 23 | #include <linux/slab.h> |
9f1b16a5 | 24 | |
60063497 | 25 | #include <linux/atomic.h> |
3947be19 DH |
26 | #include <asm/uaccess.h> |
27 | ||
2938ffbd NF |
28 | static DEFINE_MUTEX(mem_sysfs_mutex); |
29 | ||
3947be19 | 30 | #define MEMORY_CLASS_NAME "memory" |
0c2c99b1 | 31 | |
7315f0cc GZ |
32 | #define to_memory_block(dev) container_of(dev, struct memory_block, dev) |
33 | ||
0c2c99b1 NF |
34 | static int sections_per_block; |
35 | ||
36 | static inline int base_memory_block_id(int section_nr) | |
37 | { | |
38 | return section_nr / sections_per_block; | |
39 | } | |
3947be19 | 40 | |
4960e05e RW |
41 | static int memory_subsys_online(struct device *dev); |
42 | static int memory_subsys_offline(struct device *dev); | |
43 | ||
10fbcf4c | 44 | static struct bus_type memory_subsys = { |
af5ca3f4 | 45 | .name = MEMORY_CLASS_NAME, |
10fbcf4c | 46 | .dev_name = MEMORY_CLASS_NAME, |
4960e05e RW |
47 | .online = memory_subsys_online, |
48 | .offline = memory_subsys_offline, | |
3947be19 DH |
49 | }; |
50 | ||
e041c683 | 51 | static BLOCKING_NOTIFIER_HEAD(memory_chain); |
3947be19 | 52 | |
98a38ebd | 53 | int register_memory_notifier(struct notifier_block *nb) |
3947be19 | 54 | { |
2aeebca2 | 55 | return blocking_notifier_chain_register(&memory_chain, nb); |
3947be19 | 56 | } |
3c82c30c | 57 | EXPORT_SYMBOL(register_memory_notifier); |
3947be19 | 58 | |
98a38ebd | 59 | void unregister_memory_notifier(struct notifier_block *nb) |
3947be19 | 60 | { |
2aeebca2 | 61 | blocking_notifier_chain_unregister(&memory_chain, nb); |
3947be19 | 62 | } |
3c82c30c | 63 | EXPORT_SYMBOL(unregister_memory_notifier); |
3947be19 | 64 | |
925cc71e RJ |
65 | static ATOMIC_NOTIFIER_HEAD(memory_isolate_chain); |
66 | ||
67 | int register_memory_isolate_notifier(struct notifier_block *nb) | |
68 | { | |
69 | return atomic_notifier_chain_register(&memory_isolate_chain, nb); | |
70 | } | |
71 | EXPORT_SYMBOL(register_memory_isolate_notifier); | |
72 | ||
73 | void unregister_memory_isolate_notifier(struct notifier_block *nb) | |
74 | { | |
75 | atomic_notifier_chain_unregister(&memory_isolate_chain, nb); | |
76 | } | |
77 | EXPORT_SYMBOL(unregister_memory_isolate_notifier); | |
78 | ||
fa7194eb YI |
79 | static void memory_block_release(struct device *dev) |
80 | { | |
7315f0cc | 81 | struct memory_block *mem = to_memory_block(dev); |
fa7194eb YI |
82 | |
83 | kfree(mem); | |
84 | } | |
85 | ||
0c2c99b1 NF |
86 | unsigned long __weak memory_block_size_bytes(void) |
87 | { | |
88 | return MIN_MEMORY_BLOCK_SIZE; | |
89 | } | |
90 | ||
91 | static unsigned long get_memory_block_size(void) | |
92 | { | |
93 | unsigned long block_sz; | |
94 | ||
95 | block_sz = memory_block_size_bytes(); | |
96 | ||
97 | /* Validate blk_sz is a power of 2 and not less than section size */ | |
98 | if ((block_sz & (block_sz - 1)) || (block_sz < MIN_MEMORY_BLOCK_SIZE)) { | |
99 | WARN_ON(1); | |
100 | block_sz = MIN_MEMORY_BLOCK_SIZE; | |
101 | } | |
102 | ||
103 | return block_sz; | |
104 | } | |
105 | ||
3947be19 DH |
106 | /* |
107 | * use this as the physical section index that this memsection | |
108 | * uses. | |
109 | */ | |
110 | ||
10fbcf4c KS |
111 | static ssize_t show_mem_start_phys_index(struct device *dev, |
112 | struct device_attribute *attr, char *buf) | |
3947be19 | 113 | { |
7315f0cc | 114 | struct memory_block *mem = to_memory_block(dev); |
d3360164 NF |
115 | unsigned long phys_index; |
116 | ||
117 | phys_index = mem->start_section_nr / sections_per_block; | |
118 | return sprintf(buf, "%08lx\n", phys_index); | |
119 | } | |
120 | ||
5c755e9f BP |
121 | /* |
122 | * Show whether the section of memory is likely to be hot-removable | |
123 | */ | |
10fbcf4c KS |
124 | static ssize_t show_mem_removable(struct device *dev, |
125 | struct device_attribute *attr, char *buf) | |
5c755e9f | 126 | { |
0c2c99b1 NF |
127 | unsigned long i, pfn; |
128 | int ret = 1; | |
7315f0cc | 129 | struct memory_block *mem = to_memory_block(dev); |
5c755e9f | 130 | |
0c2c99b1 | 131 | for (i = 0; i < sections_per_block; i++) { |
21ea9f5a RA |
132 | if (!present_section_nr(mem->start_section_nr + i)) |
133 | continue; | |
d3360164 | 134 | pfn = section_nr_to_pfn(mem->start_section_nr + i); |
0c2c99b1 NF |
135 | ret &= is_mem_section_removable(pfn, PAGES_PER_SECTION); |
136 | } | |
137 | ||
5c755e9f BP |
138 | return sprintf(buf, "%d\n", ret); |
139 | } | |
140 | ||
3947be19 DH |
141 | /* |
142 | * online, offline, going offline, etc. | |
143 | */ | |
10fbcf4c KS |
144 | static ssize_t show_mem_state(struct device *dev, |
145 | struct device_attribute *attr, char *buf) | |
3947be19 | 146 | { |
7315f0cc | 147 | struct memory_block *mem = to_memory_block(dev); |
3947be19 DH |
148 | ssize_t len = 0; |
149 | ||
150 | /* | |
151 | * We can probably put these states in a nice little array | |
152 | * so that they're not open-coded | |
153 | */ | |
154 | switch (mem->state) { | |
3d3af6af IC |
155 | case MEM_ONLINE: |
156 | len = sprintf(buf, "online\n"); | |
157 | break; | |
158 | case MEM_OFFLINE: | |
159 | len = sprintf(buf, "offline\n"); | |
160 | break; | |
161 | case MEM_GOING_OFFLINE: | |
162 | len = sprintf(buf, "going-offline\n"); | |
163 | break; | |
164 | default: | |
165 | len = sprintf(buf, "ERROR-UNKNOWN-%ld\n", | |
166 | mem->state); | |
167 | WARN_ON(1); | |
168 | break; | |
3947be19 DH |
169 | } |
170 | ||
171 | return len; | |
172 | } | |
173 | ||
7b78d335 | 174 | int memory_notify(unsigned long val, void *v) |
3947be19 | 175 | { |
e041c683 | 176 | return blocking_notifier_call_chain(&memory_chain, val, v); |
3947be19 DH |
177 | } |
178 | ||
925cc71e RJ |
179 | int memory_isolate_notify(unsigned long val, void *v) |
180 | { | |
181 | return atomic_notifier_call_chain(&memory_isolate_chain, val, v); | |
182 | } | |
183 | ||
2bbcb878 MG |
184 | /* |
185 | * The probe routines leave the pages reserved, just as the bootmem code does. | |
186 | * Make sure they're still that way. | |
187 | */ | |
6056d619 | 188 | static bool pages_correctly_reserved(unsigned long start_pfn) |
2bbcb878 MG |
189 | { |
190 | int i, j; | |
191 | struct page *page; | |
192 | unsigned long pfn = start_pfn; | |
193 | ||
194 | /* | |
195 | * memmap between sections is not contiguous except with | |
196 | * SPARSEMEM_VMEMMAP. We lookup the page once per section | |
197 | * and assume memmap is contiguous within each section | |
198 | */ | |
199 | for (i = 0; i < sections_per_block; i++, pfn += PAGES_PER_SECTION) { | |
200 | if (WARN_ON_ONCE(!pfn_valid(pfn))) | |
201 | return false; | |
202 | page = pfn_to_page(pfn); | |
203 | ||
204 | for (j = 0; j < PAGES_PER_SECTION; j++) { | |
205 | if (PageReserved(page + j)) | |
206 | continue; | |
207 | ||
208 | printk(KERN_WARNING "section number %ld page number %d " | |
209 | "not reserved, was it already online?\n", | |
210 | pfn_to_section_nr(pfn), j); | |
211 | ||
212 | return false; | |
213 | } | |
214 | } | |
215 | ||
216 | return true; | |
217 | } | |
218 | ||
3947be19 DH |
219 | /* |
220 | * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is | |
221 | * OK to have direct references to sparsemem variables in here. | |
30467e0b | 222 | * Must already be protected by mem_hotplug_begin(). |
3947be19 DH |
223 | */ |
224 | static int | |
511c2aba | 225 | memory_block_action(unsigned long phys_index, unsigned long action, int online_type) |
3947be19 | 226 | { |
a16cee10 | 227 | unsigned long start_pfn; |
5409d2cd | 228 | unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block; |
de0ed36a | 229 | struct page *first_page; |
3947be19 | 230 | int ret; |
3947be19 | 231 | |
19c07d5e | 232 | start_pfn = section_nr_to_pfn(phys_index); |
71fbd556 | 233 | first_page = pfn_to_page(start_pfn); |
de0ed36a | 234 | |
3947be19 | 235 | switch (action) { |
3d3af6af IC |
236 | case MEM_ONLINE: |
237 | if (!pages_correctly_reserved(start_pfn)) | |
238 | return -EBUSY; | |
239 | ||
240 | ret = online_pages(start_pfn, nr_pages, online_type); | |
241 | break; | |
242 | case MEM_OFFLINE: | |
243 | ret = offline_pages(start_pfn, nr_pages); | |
244 | break; | |
245 | default: | |
246 | WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: " | |
247 | "%ld\n", __func__, phys_index, action, action); | |
248 | ret = -EINVAL; | |
3947be19 | 249 | } |
3947be19 DH |
250 | |
251 | return ret; | |
252 | } | |
253 | ||
31bc3858 | 254 | int memory_block_change_state(struct memory_block *mem, |
fa2be40f | 255 | unsigned long to_state, unsigned long from_state_req) |
3947be19 | 256 | { |
de0ed36a | 257 | int ret = 0; |
0c2c99b1 | 258 | |
4960e05e RW |
259 | if (mem->state != from_state_req) |
260 | return -EINVAL; | |
3947be19 | 261 | |
0c2c99b1 NF |
262 | if (to_state == MEM_OFFLINE) |
263 | mem->state = MEM_GOING_OFFLINE; | |
264 | ||
fa2be40f SJ |
265 | ret = memory_block_action(mem->start_section_nr, to_state, |
266 | mem->online_type); | |
267 | ||
b2c064b2 | 268 | mem->state = ret ? from_state_req : to_state; |
fa2be40f | 269 | |
4960e05e RW |
270 | return ret; |
271 | } | |
0c2c99b1 | 272 | |
fa2be40f | 273 | /* The device lock serializes operations on memory_subsys_[online|offline] */ |
4960e05e RW |
274 | static int memory_subsys_online(struct device *dev) |
275 | { | |
7315f0cc | 276 | struct memory_block *mem = to_memory_block(dev); |
4960e05e | 277 | int ret; |
3947be19 | 278 | |
fa2be40f SJ |
279 | if (mem->state == MEM_ONLINE) |
280 | return 0; | |
4960e05e | 281 | |
fa2be40f SJ |
282 | /* |
283 | * If we are called from store_mem_state(), online_type will be | |
284 | * set >= 0 Otherwise we were called from the device online | |
285 | * attribute and need to set the online_type. | |
286 | */ | |
287 | if (mem->online_type < 0) | |
4f7c6b49 | 288 | mem->online_type = MMOP_ONLINE_KEEP; |
4960e05e | 289 | |
30467e0b | 290 | /* Already under protection of mem_hotplug_begin() */ |
fa2be40f | 291 | ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE); |
4960e05e | 292 | |
fa2be40f SJ |
293 | /* clear online_type */ |
294 | mem->online_type = -1; | |
4960e05e | 295 | |
4960e05e RW |
296 | return ret; |
297 | } | |
298 | ||
4960e05e | 299 | static int memory_subsys_offline(struct device *dev) |
e90bdb7f | 300 | { |
7315f0cc | 301 | struct memory_block *mem = to_memory_block(dev); |
e90bdb7f | 302 | |
fa2be40f SJ |
303 | if (mem->state == MEM_OFFLINE) |
304 | return 0; | |
e90bdb7f | 305 | |
26bbe7ef SJ |
306 | /* Can't offline block with non-present sections */ |
307 | if (mem->section_count != sections_per_block) | |
308 | return -EINVAL; | |
309 | ||
fa2be40f | 310 | return memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE); |
e90bdb7f | 311 | } |
4960e05e | 312 | |
3947be19 | 313 | static ssize_t |
10fbcf4c KS |
314 | store_mem_state(struct device *dev, |
315 | struct device_attribute *attr, const char *buf, size_t count) | |
3947be19 | 316 | { |
7315f0cc | 317 | struct memory_block *mem = to_memory_block(dev); |
fa2be40f | 318 | int ret, online_type; |
3947be19 | 319 | |
5e33bc41 RW |
320 | ret = lock_device_hotplug_sysfs(); |
321 | if (ret) | |
322 | return ret; | |
4960e05e | 323 | |
1f6a6cc8 | 324 | if (sysfs_streq(buf, "online_kernel")) |
4f7c6b49 | 325 | online_type = MMOP_ONLINE_KERNEL; |
1f6a6cc8 | 326 | else if (sysfs_streq(buf, "online_movable")) |
4f7c6b49 | 327 | online_type = MMOP_ONLINE_MOVABLE; |
1f6a6cc8 | 328 | else if (sysfs_streq(buf, "online")) |
4f7c6b49 | 329 | online_type = MMOP_ONLINE_KEEP; |
1f6a6cc8 | 330 | else if (sysfs_streq(buf, "offline")) |
4f7c6b49 | 331 | online_type = MMOP_OFFLINE; |
a37f8630 YI |
332 | else { |
333 | ret = -EINVAL; | |
334 | goto err; | |
335 | } | |
fa2be40f | 336 | |
30467e0b DR |
337 | /* |
338 | * Memory hotplug needs to hold mem_hotplug_begin() for probe to find | |
339 | * the correct memory block to online before doing device_online(dev), | |
340 | * which will take dev->mutex. Take the lock early to prevent an | |
341 | * inversion, memory_subsys_online() callbacks will be implemented by | |
342 | * assuming it's already protected. | |
343 | */ | |
344 | mem_hotplug_begin(); | |
345 | ||
fa2be40f | 346 | switch (online_type) { |
4f7c6b49 TC |
347 | case MMOP_ONLINE_KERNEL: |
348 | case MMOP_ONLINE_MOVABLE: | |
349 | case MMOP_ONLINE_KEEP: | |
fa2be40f SJ |
350 | mem->online_type = online_type; |
351 | ret = device_online(&mem->dev); | |
352 | break; | |
4f7c6b49 | 353 | case MMOP_OFFLINE: |
fa2be40f SJ |
354 | ret = device_offline(&mem->dev); |
355 | break; | |
356 | default: | |
357 | ret = -EINVAL; /* should never happen */ | |
4960e05e | 358 | } |
4960e05e | 359 | |
30467e0b | 360 | mem_hotplug_done(); |
a37f8630 | 361 | err: |
4960e05e | 362 | unlock_device_hotplug(); |
0c2c99b1 | 363 | |
3947be19 DH |
364 | if (ret) |
365 | return ret; | |
366 | return count; | |
367 | } | |
368 | ||
369 | /* | |
370 | * phys_device is a bad name for this. What I really want | |
371 | * is a way to differentiate between memory ranges that | |
372 | * are part of physical devices that constitute | |
373 | * a complete removable unit or fru. | |
374 | * i.e. do these ranges belong to the same physical device, | |
375 | * s.t. if I offline all of these sections I can then | |
376 | * remove the physical device? | |
377 | */ | |
10fbcf4c KS |
378 | static ssize_t show_phys_device(struct device *dev, |
379 | struct device_attribute *attr, char *buf) | |
3947be19 | 380 | { |
7315f0cc | 381 | struct memory_block *mem = to_memory_block(dev); |
3947be19 DH |
382 | return sprintf(buf, "%d\n", mem->phys_device); |
383 | } | |
384 | ||
ed2f2400 ZZ |
385 | #ifdef CONFIG_MEMORY_HOTREMOVE |
386 | static ssize_t show_valid_zones(struct device *dev, | |
387 | struct device_attribute *attr, char *buf) | |
388 | { | |
389 | struct memory_block *mem = to_memory_block(dev); | |
390 | unsigned long start_pfn, end_pfn; | |
391 | unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block; | |
392 | struct page *first_page; | |
393 | struct zone *zone; | |
394 | ||
395 | start_pfn = section_nr_to_pfn(mem->start_section_nr); | |
396 | end_pfn = start_pfn + nr_pages; | |
397 | first_page = pfn_to_page(start_pfn); | |
398 | ||
399 | /* The block contains more than one zone can not be offlined. */ | |
400 | if (!test_pages_in_a_zone(start_pfn, end_pfn)) | |
401 | return sprintf(buf, "none\n"); | |
402 | ||
403 | zone = page_zone(first_page); | |
404 | ||
405 | if (zone_idx(zone) == ZONE_MOVABLE - 1) { | |
406 | /*The mem block is the last memoryblock of this zone.*/ | |
407 | if (end_pfn == zone_end_pfn(zone)) | |
408 | return sprintf(buf, "%s %s\n", | |
409 | zone->name, (zone + 1)->name); | |
410 | } | |
411 | ||
412 | if (zone_idx(zone) == ZONE_MOVABLE) { | |
413 | /*The mem block is the first memoryblock of ZONE_MOVABLE.*/ | |
414 | if (start_pfn == zone->zone_start_pfn) | |
415 | return sprintf(buf, "%s %s\n", | |
416 | zone->name, (zone - 1)->name); | |
417 | } | |
418 | ||
419 | return sprintf(buf, "%s\n", zone->name); | |
420 | } | |
421 | static DEVICE_ATTR(valid_zones, 0444, show_valid_zones, NULL); | |
422 | #endif | |
423 | ||
10fbcf4c | 424 | static DEVICE_ATTR(phys_index, 0444, show_mem_start_phys_index, NULL); |
10fbcf4c KS |
425 | static DEVICE_ATTR(state, 0644, show_mem_state, store_mem_state); |
426 | static DEVICE_ATTR(phys_device, 0444, show_phys_device, NULL); | |
427 | static DEVICE_ATTR(removable, 0444, show_mem_removable, NULL); | |
3947be19 | 428 | |
3947be19 DH |
429 | /* |
430 | * Block size attribute stuff | |
431 | */ | |
432 | static ssize_t | |
10fbcf4c | 433 | print_block_size(struct device *dev, struct device_attribute *attr, |
8564a6c1 | 434 | char *buf) |
3947be19 | 435 | { |
0c2c99b1 | 436 | return sprintf(buf, "%lx\n", get_memory_block_size()); |
3947be19 DH |
437 | } |
438 | ||
10fbcf4c | 439 | static DEVICE_ATTR(block_size_bytes, 0444, print_block_size, NULL); |
3947be19 | 440 | |
31bc3858 VK |
441 | /* |
442 | * Memory auto online policy. | |
443 | */ | |
444 | ||
445 | static ssize_t | |
446 | show_auto_online_blocks(struct device *dev, struct device_attribute *attr, | |
447 | char *buf) | |
448 | { | |
449 | if (memhp_auto_online) | |
450 | return sprintf(buf, "online\n"); | |
451 | else | |
452 | return sprintf(buf, "offline\n"); | |
453 | } | |
454 | ||
455 | static ssize_t | |
456 | store_auto_online_blocks(struct device *dev, struct device_attribute *attr, | |
457 | const char *buf, size_t count) | |
458 | { | |
459 | if (sysfs_streq(buf, "online")) | |
460 | memhp_auto_online = true; | |
461 | else if (sysfs_streq(buf, "offline")) | |
462 | memhp_auto_online = false; | |
463 | else | |
464 | return -EINVAL; | |
465 | ||
466 | return count; | |
467 | } | |
468 | ||
469 | static DEVICE_ATTR(auto_online_blocks, 0644, show_auto_online_blocks, | |
470 | store_auto_online_blocks); | |
471 | ||
3947be19 DH |
472 | /* |
473 | * Some architectures will have custom drivers to do this, and | |
474 | * will not need to do it from userspace. The fake hot-add code | |
475 | * as well as ppc64 will do all of their discovery in userspace | |
476 | * and will require this interface. | |
477 | */ | |
478 | #ifdef CONFIG_ARCH_MEMORY_PROBE | |
479 | static ssize_t | |
10fbcf4c | 480 | memory_probe_store(struct device *dev, struct device_attribute *attr, |
28812fe1 | 481 | const char *buf, size_t count) |
3947be19 DH |
482 | { |
483 | u64 phys_addr; | |
cb5490a5 | 484 | int nid, ret; |
61b94fea | 485 | unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block; |
3947be19 | 486 | |
b69deb2b ZZ |
487 | ret = kstrtoull(buf, 0, &phys_addr); |
488 | if (ret) | |
489 | return ret; | |
3947be19 | 490 | |
61b94fea AB |
491 | if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1)) |
492 | return -EINVAL; | |
493 | ||
cb5490a5 JA |
494 | nid = memory_add_physaddr_to_nid(phys_addr); |
495 | ret = add_memory(nid, phys_addr, | |
496 | MIN_MEMORY_BLOCK_SIZE * sections_per_block); | |
6add7cd6 | 497 | |
cb5490a5 JA |
498 | if (ret) |
499 | goto out; | |
3947be19 | 500 | |
9f0af69b NK |
501 | ret = count; |
502 | out: | |
503 | return ret; | |
3947be19 | 504 | } |
3947be19 | 505 | |
96b2c0fc | 506 | static DEVICE_ATTR(probe, S_IWUSR, NULL, memory_probe_store); |
3947be19 DH |
507 | #endif |
508 | ||
facb6011 AK |
509 | #ifdef CONFIG_MEMORY_FAILURE |
510 | /* | |
511 | * Support for offlining pages of memory | |
512 | */ | |
513 | ||
514 | /* Soft offline a page */ | |
515 | static ssize_t | |
10fbcf4c KS |
516 | store_soft_offline_page(struct device *dev, |
517 | struct device_attribute *attr, | |
28812fe1 | 518 | const char *buf, size_t count) |
facb6011 AK |
519 | { |
520 | int ret; | |
521 | u64 pfn; | |
522 | if (!capable(CAP_SYS_ADMIN)) | |
523 | return -EPERM; | |
34da5e67 | 524 | if (kstrtoull(buf, 0, &pfn) < 0) |
facb6011 AK |
525 | return -EINVAL; |
526 | pfn >>= PAGE_SHIFT; | |
527 | if (!pfn_valid(pfn)) | |
528 | return -ENXIO; | |
529 | ret = soft_offline_page(pfn_to_page(pfn), 0); | |
530 | return ret == 0 ? count : ret; | |
531 | } | |
532 | ||
533 | /* Forcibly offline a page, including killing processes. */ | |
534 | static ssize_t | |
10fbcf4c KS |
535 | store_hard_offline_page(struct device *dev, |
536 | struct device_attribute *attr, | |
28812fe1 | 537 | const char *buf, size_t count) |
facb6011 AK |
538 | { |
539 | int ret; | |
540 | u64 pfn; | |
541 | if (!capable(CAP_SYS_ADMIN)) | |
542 | return -EPERM; | |
34da5e67 | 543 | if (kstrtoull(buf, 0, &pfn) < 0) |
facb6011 AK |
544 | return -EINVAL; |
545 | pfn >>= PAGE_SHIFT; | |
cd42f4a3 | 546 | ret = memory_failure(pfn, 0, 0); |
facb6011 AK |
547 | return ret ? ret : count; |
548 | } | |
549 | ||
74fef7a8 FB |
550 | static DEVICE_ATTR(soft_offline_page, S_IWUSR, NULL, store_soft_offline_page); |
551 | static DEVICE_ATTR(hard_offline_page, S_IWUSR, NULL, store_hard_offline_page); | |
facb6011 AK |
552 | #endif |
553 | ||
3947be19 DH |
554 | /* |
555 | * Note that phys_device is optional. It is here to allow for | |
556 | * differentiation between which *physical* devices each | |
557 | * section belongs to... | |
558 | */ | |
bc32df00 HC |
559 | int __weak arch_get_memory_phys_device(unsigned long start_pfn) |
560 | { | |
561 | return 0; | |
562 | } | |
3947be19 | 563 | |
10fbcf4c KS |
564 | /* |
565 | * A reference for the returned object is held and the reference for the | |
566 | * hinted object is released. | |
567 | */ | |
98383031 RH |
568 | struct memory_block *find_memory_block_hinted(struct mem_section *section, |
569 | struct memory_block *hint) | |
3947be19 | 570 | { |
0c2c99b1 | 571 | int block_id = base_memory_block_id(__section_nr(section)); |
10fbcf4c KS |
572 | struct device *hintdev = hint ? &hint->dev : NULL; |
573 | struct device *dev; | |
3947be19 | 574 | |
10fbcf4c KS |
575 | dev = subsys_find_device_by_id(&memory_subsys, block_id, hintdev); |
576 | if (hint) | |
577 | put_device(&hint->dev); | |
578 | if (!dev) | |
3947be19 | 579 | return NULL; |
7315f0cc | 580 | return to_memory_block(dev); |
3947be19 DH |
581 | } |
582 | ||
98383031 RH |
583 | /* |
584 | * For now, we have a linear search to go find the appropriate | |
585 | * memory_block corresponding to a particular phys_index. If | |
586 | * this gets to be a real problem, we can always use a radix | |
587 | * tree or something here. | |
588 | * | |
10fbcf4c | 589 | * This could be made generic for all device subsystems. |
98383031 RH |
590 | */ |
591 | struct memory_block *find_memory_block(struct mem_section *section) | |
592 | { | |
593 | return find_memory_block_hinted(section, NULL); | |
594 | } | |
595 | ||
96b2c0fc NF |
596 | static struct attribute *memory_memblk_attrs[] = { |
597 | &dev_attr_phys_index.attr, | |
96b2c0fc NF |
598 | &dev_attr_state.attr, |
599 | &dev_attr_phys_device.attr, | |
600 | &dev_attr_removable.attr, | |
ed2f2400 ZZ |
601 | #ifdef CONFIG_MEMORY_HOTREMOVE |
602 | &dev_attr_valid_zones.attr, | |
603 | #endif | |
96b2c0fc NF |
604 | NULL |
605 | }; | |
606 | ||
607 | static struct attribute_group memory_memblk_attr_group = { | |
608 | .attrs = memory_memblk_attrs, | |
609 | }; | |
610 | ||
611 | static const struct attribute_group *memory_memblk_attr_groups[] = { | |
612 | &memory_memblk_attr_group, | |
613 | NULL, | |
614 | }; | |
615 | ||
616 | /* | |
617 | * register_memory - Setup a sysfs device for a memory block | |
618 | */ | |
619 | static | |
620 | int register_memory(struct memory_block *memory) | |
621 | { | |
96b2c0fc NF |
622 | memory->dev.bus = &memory_subsys; |
623 | memory->dev.id = memory->start_section_nr / sections_per_block; | |
624 | memory->dev.release = memory_block_release; | |
625 | memory->dev.groups = memory_memblk_attr_groups; | |
f991fae5 | 626 | memory->dev.offline = memory->state == MEM_OFFLINE; |
96b2c0fc | 627 | |
879f1bec | 628 | return device_register(&memory->dev); |
96b2c0fc NF |
629 | } |
630 | ||
0c2c99b1 NF |
631 | static int init_memory_block(struct memory_block **memory, |
632 | struct mem_section *section, unsigned long state) | |
e4619c85 | 633 | { |
0c2c99b1 | 634 | struct memory_block *mem; |
e4619c85 | 635 | unsigned long start_pfn; |
0c2c99b1 | 636 | int scn_nr; |
e4619c85 NF |
637 | int ret = 0; |
638 | ||
0c2c99b1 | 639 | mem = kzalloc(sizeof(*mem), GFP_KERNEL); |
e4619c85 NF |
640 | if (!mem) |
641 | return -ENOMEM; | |
642 | ||
0c2c99b1 | 643 | scn_nr = __section_nr(section); |
d3360164 NF |
644 | mem->start_section_nr = |
645 | base_memory_block_id(scn_nr) * sections_per_block; | |
646 | mem->end_section_nr = mem->start_section_nr + sections_per_block - 1; | |
e4619c85 | 647 | mem->state = state; |
d3360164 | 648 | start_pfn = section_nr_to_pfn(mem->start_section_nr); |
e4619c85 NF |
649 | mem->phys_device = arch_get_memory_phys_device(start_pfn); |
650 | ||
0c2c99b1 | 651 | ret = register_memory(mem); |
0c2c99b1 NF |
652 | |
653 | *memory = mem; | |
654 | return ret; | |
655 | } | |
656 | ||
cb5e39b8 | 657 | static int add_memory_block(int base_section_nr) |
0c2c99b1 | 658 | { |
cb5e39b8 SJ |
659 | struct memory_block *mem; |
660 | int i, ret, section_count = 0, section_nr; | |
0c2c99b1 | 661 | |
cb5e39b8 SJ |
662 | for (i = base_section_nr; |
663 | (i < base_section_nr + sections_per_block) && i < NR_MEM_SECTIONS; | |
664 | i++) { | |
665 | if (!present_section_nr(i)) | |
666 | continue; | |
667 | if (section_count == 0) | |
668 | section_nr = i; | |
669 | section_count++; | |
e4619c85 NF |
670 | } |
671 | ||
cb5e39b8 SJ |
672 | if (section_count == 0) |
673 | return 0; | |
674 | ret = init_memory_block(&mem, __nr_to_section(section_nr), MEM_ONLINE); | |
675 | if (ret) | |
676 | return ret; | |
677 | mem->section_count = section_count; | |
678 | return 0; | |
e4619c85 NF |
679 | } |
680 | ||
260ae3f7 DW |
681 | static bool is_zone_device_section(struct mem_section *ms) |
682 | { | |
683 | struct page *page; | |
684 | ||
685 | page = sparse_decode_mem_map(ms->section_mem_map, __section_nr(ms)); | |
686 | return is_zone_device_page(page); | |
687 | } | |
cb5e39b8 | 688 | |
4edd7cef DR |
689 | /* |
690 | * need an interface for the VM to add new memory regions, | |
691 | * but without onlining it. | |
692 | */ | |
693 | int register_new_memory(int nid, struct mem_section *section) | |
694 | { | |
d7f80530 SJ |
695 | int ret = 0; |
696 | struct memory_block *mem; | |
b1eaef3d | 697 | |
260ae3f7 DW |
698 | if (is_zone_device_section(section)) |
699 | return 0; | |
700 | ||
b1eaef3d | 701 | mutex_lock(&mem_sysfs_mutex); |
b1eaef3d | 702 | |
d7f80530 SJ |
703 | mem = find_memory_block(section); |
704 | if (mem) { | |
705 | mem->section_count++; | |
706 | put_device(&mem->dev); | |
707 | } else { | |
708 | ret = init_memory_block(&mem, section, MEM_OFFLINE); | |
709 | if (ret) | |
710 | goto out; | |
56c6b5d3 | 711 | mem->section_count++; |
d7f80530 SJ |
712 | } |
713 | ||
714 | if (mem->section_count == sections_per_block) | |
715 | ret = register_mem_sect_under_node(mem, nid); | |
716 | out: | |
717 | mutex_unlock(&mem_sysfs_mutex); | |
b1eaef3d | 718 | return ret; |
4edd7cef DR |
719 | } |
720 | ||
721 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
722 | static void | |
723 | unregister_memory(struct memory_block *memory) | |
724 | { | |
725 | BUG_ON(memory->dev.bus != &memory_subsys); | |
726 | ||
727 | /* drop the ref. we got in remove_memory_block() */ | |
df2b717c | 728 | put_device(&memory->dev); |
4edd7cef DR |
729 | device_unregister(&memory->dev); |
730 | } | |
731 | ||
cc292b0b | 732 | static int remove_memory_section(unsigned long node_id, |
4edd7cef | 733 | struct mem_section *section, int phys_device) |
3947be19 DH |
734 | { |
735 | struct memory_block *mem; | |
736 | ||
260ae3f7 DW |
737 | if (is_zone_device_section(section)) |
738 | return 0; | |
739 | ||
2938ffbd | 740 | mutex_lock(&mem_sysfs_mutex); |
3947be19 | 741 | mem = find_memory_block(section); |
d3360164 | 742 | unregister_mem_sect_under_nodes(mem, __section_nr(section)); |
07681215 NF |
743 | |
744 | mem->section_count--; | |
96b2c0fc | 745 | if (mem->section_count == 0) |
0c2c99b1 | 746 | unregister_memory(mem); |
96b2c0fc | 747 | else |
df2b717c | 748 | put_device(&mem->dev); |
3947be19 | 749 | |
2938ffbd | 750 | mutex_unlock(&mem_sysfs_mutex); |
3947be19 DH |
751 | return 0; |
752 | } | |
753 | ||
3947be19 DH |
754 | int unregister_memory_section(struct mem_section *section) |
755 | { | |
540557b9 | 756 | if (!present_section(section)) |
3947be19 DH |
757 | return -EINVAL; |
758 | ||
cc292b0b | 759 | return remove_memory_section(0, section, 0); |
3947be19 | 760 | } |
4edd7cef | 761 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
3947be19 | 762 | |
6677e3ea YI |
763 | /* return true if the memory block is offlined, otherwise, return false */ |
764 | bool is_memblock_offlined(struct memory_block *mem) | |
765 | { | |
766 | return mem->state == MEM_OFFLINE; | |
767 | } | |
768 | ||
96b2c0fc NF |
769 | static struct attribute *memory_root_attrs[] = { |
770 | #ifdef CONFIG_ARCH_MEMORY_PROBE | |
771 | &dev_attr_probe.attr, | |
772 | #endif | |
773 | ||
774 | #ifdef CONFIG_MEMORY_FAILURE | |
775 | &dev_attr_soft_offline_page.attr, | |
776 | &dev_attr_hard_offline_page.attr, | |
777 | #endif | |
778 | ||
779 | &dev_attr_block_size_bytes.attr, | |
31bc3858 | 780 | &dev_attr_auto_online_blocks.attr, |
96b2c0fc NF |
781 | NULL |
782 | }; | |
783 | ||
784 | static struct attribute_group memory_root_attr_group = { | |
785 | .attrs = memory_root_attrs, | |
786 | }; | |
787 | ||
788 | static const struct attribute_group *memory_root_attr_groups[] = { | |
789 | &memory_root_attr_group, | |
790 | NULL, | |
791 | }; | |
792 | ||
3947be19 DH |
793 | /* |
794 | * Initialize the sysfs support for memory devices... | |
795 | */ | |
796 | int __init memory_dev_init(void) | |
797 | { | |
798 | unsigned int i; | |
799 | int ret; | |
28ec24e2 | 800 | int err; |
0c2c99b1 | 801 | unsigned long block_sz; |
3947be19 | 802 | |
96b2c0fc | 803 | ret = subsys_system_register(&memory_subsys, memory_root_attr_groups); |
28ec24e2 AM |
804 | if (ret) |
805 | goto out; | |
3947be19 | 806 | |
0c2c99b1 NF |
807 | block_sz = get_memory_block_size(); |
808 | sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE; | |
809 | ||
3947be19 DH |
810 | /* |
811 | * Create entries for memory sections that were found | |
812 | * during boot and have been initialized | |
813 | */ | |
b1eaef3d | 814 | mutex_lock(&mem_sysfs_mutex); |
cb5e39b8 SJ |
815 | for (i = 0; i < NR_MEM_SECTIONS; i += sections_per_block) { |
816 | err = add_memory_block(i); | |
28ec24e2 AM |
817 | if (!ret) |
818 | ret = err; | |
3947be19 | 819 | } |
b1eaef3d | 820 | mutex_unlock(&mem_sysfs_mutex); |
3947be19 | 821 | |
28ec24e2 AM |
822 | out: |
823 | if (ret) | |
2b3a302a | 824 | printk(KERN_ERR "%s() failed: %d\n", __func__, ret); |
3947be19 DH |
825 | return ret; |
826 | } |