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> | |
19 | #include <linux/kobject.h> | |
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> |
3947be19 DH |
27 | #include <asm/uaccess.h> |
28 | ||
2938ffbd NF |
29 | static DEFINE_MUTEX(mem_sysfs_mutex); |
30 | ||
3947be19 | 31 | #define MEMORY_CLASS_NAME "memory" |
0c2c99b1 NF |
32 | |
33 | static int sections_per_block; | |
34 | ||
35 | static inline int base_memory_block_id(int section_nr) | |
36 | { | |
37 | return section_nr / sections_per_block; | |
38 | } | |
3947be19 | 39 | |
10fbcf4c | 40 | static struct bus_type memory_subsys = { |
af5ca3f4 | 41 | .name = MEMORY_CLASS_NAME, |
10fbcf4c | 42 | .dev_name = MEMORY_CLASS_NAME, |
3947be19 DH |
43 | }; |
44 | ||
e041c683 | 45 | static BLOCKING_NOTIFIER_HEAD(memory_chain); |
3947be19 | 46 | |
98a38ebd | 47 | int register_memory_notifier(struct notifier_block *nb) |
3947be19 | 48 | { |
e041c683 | 49 | return blocking_notifier_chain_register(&memory_chain, nb); |
3947be19 | 50 | } |
3c82c30c | 51 | EXPORT_SYMBOL(register_memory_notifier); |
3947be19 | 52 | |
98a38ebd | 53 | void unregister_memory_notifier(struct notifier_block *nb) |
3947be19 | 54 | { |
e041c683 | 55 | blocking_notifier_chain_unregister(&memory_chain, nb); |
3947be19 | 56 | } |
3c82c30c | 57 | EXPORT_SYMBOL(unregister_memory_notifier); |
3947be19 | 58 | |
925cc71e RJ |
59 | static ATOMIC_NOTIFIER_HEAD(memory_isolate_chain); |
60 | ||
61 | int register_memory_isolate_notifier(struct notifier_block *nb) | |
62 | { | |
63 | return atomic_notifier_chain_register(&memory_isolate_chain, nb); | |
64 | } | |
65 | EXPORT_SYMBOL(register_memory_isolate_notifier); | |
66 | ||
67 | void unregister_memory_isolate_notifier(struct notifier_block *nb) | |
68 | { | |
69 | atomic_notifier_chain_unregister(&memory_isolate_chain, nb); | |
70 | } | |
71 | EXPORT_SYMBOL(unregister_memory_isolate_notifier); | |
72 | ||
fa7194eb YI |
73 | static void memory_block_release(struct device *dev) |
74 | { | |
75 | struct memory_block *mem = container_of(dev, struct memory_block, dev); | |
76 | ||
77 | kfree(mem); | |
78 | } | |
79 | ||
3947be19 DH |
80 | /* |
81 | * register_memory - Setup a sysfs device for a memory block | |
82 | */ | |
00a41db5 | 83 | static |
0c2c99b1 | 84 | int register_memory(struct memory_block *memory) |
3947be19 DH |
85 | { |
86 | int error; | |
87 | ||
10fbcf4c KS |
88 | memory->dev.bus = &memory_subsys; |
89 | memory->dev.id = memory->start_section_nr / sections_per_block; | |
fa7194eb | 90 | memory->dev.release = memory_block_release; |
3947be19 | 91 | |
10fbcf4c | 92 | error = device_register(&memory->dev); |
3947be19 DH |
93 | return error; |
94 | } | |
95 | ||
96 | static void | |
0c2c99b1 | 97 | unregister_memory(struct memory_block *memory) |
3947be19 | 98 | { |
10fbcf4c | 99 | BUG_ON(memory->dev.bus != &memory_subsys); |
3947be19 | 100 | |
00a41db5 | 101 | /* drop the ref. we got in remove_memory_block() */ |
10fbcf4c KS |
102 | kobject_put(&memory->dev.kobj); |
103 | device_unregister(&memory->dev); | |
3947be19 DH |
104 | } |
105 | ||
0c2c99b1 NF |
106 | unsigned long __weak memory_block_size_bytes(void) |
107 | { | |
108 | return MIN_MEMORY_BLOCK_SIZE; | |
109 | } | |
110 | ||
111 | static unsigned long get_memory_block_size(void) | |
112 | { | |
113 | unsigned long block_sz; | |
114 | ||
115 | block_sz = memory_block_size_bytes(); | |
116 | ||
117 | /* Validate blk_sz is a power of 2 and not less than section size */ | |
118 | if ((block_sz & (block_sz - 1)) || (block_sz < MIN_MEMORY_BLOCK_SIZE)) { | |
119 | WARN_ON(1); | |
120 | block_sz = MIN_MEMORY_BLOCK_SIZE; | |
121 | } | |
122 | ||
123 | return block_sz; | |
124 | } | |
125 | ||
3947be19 DH |
126 | /* |
127 | * use this as the physical section index that this memsection | |
128 | * uses. | |
129 | */ | |
130 | ||
10fbcf4c KS |
131 | static ssize_t show_mem_start_phys_index(struct device *dev, |
132 | struct device_attribute *attr, char *buf) | |
3947be19 DH |
133 | { |
134 | struct memory_block *mem = | |
10fbcf4c | 135 | container_of(dev, struct memory_block, dev); |
d3360164 NF |
136 | unsigned long phys_index; |
137 | ||
138 | phys_index = mem->start_section_nr / sections_per_block; | |
139 | return sprintf(buf, "%08lx\n", phys_index); | |
140 | } | |
141 | ||
10fbcf4c KS |
142 | static ssize_t show_mem_end_phys_index(struct device *dev, |
143 | struct device_attribute *attr, char *buf) | |
d3360164 NF |
144 | { |
145 | struct memory_block *mem = | |
10fbcf4c | 146 | container_of(dev, struct memory_block, dev); |
d3360164 NF |
147 | unsigned long phys_index; |
148 | ||
149 | phys_index = mem->end_section_nr / sections_per_block; | |
150 | return sprintf(buf, "%08lx\n", phys_index); | |
3947be19 DH |
151 | } |
152 | ||
5c755e9f BP |
153 | /* |
154 | * Show whether the section of memory is likely to be hot-removable | |
155 | */ | |
10fbcf4c KS |
156 | static ssize_t show_mem_removable(struct device *dev, |
157 | struct device_attribute *attr, char *buf) | |
5c755e9f | 158 | { |
0c2c99b1 NF |
159 | unsigned long i, pfn; |
160 | int ret = 1; | |
5c755e9f | 161 | struct memory_block *mem = |
10fbcf4c | 162 | container_of(dev, struct memory_block, dev); |
5c755e9f | 163 | |
0c2c99b1 | 164 | for (i = 0; i < sections_per_block; i++) { |
d3360164 | 165 | pfn = section_nr_to_pfn(mem->start_section_nr + i); |
0c2c99b1 NF |
166 | ret &= is_mem_section_removable(pfn, PAGES_PER_SECTION); |
167 | } | |
168 | ||
5c755e9f BP |
169 | return sprintf(buf, "%d\n", ret); |
170 | } | |
171 | ||
3947be19 DH |
172 | /* |
173 | * online, offline, going offline, etc. | |
174 | */ | |
10fbcf4c KS |
175 | static ssize_t show_mem_state(struct device *dev, |
176 | struct device_attribute *attr, char *buf) | |
3947be19 DH |
177 | { |
178 | struct memory_block *mem = | |
10fbcf4c | 179 | container_of(dev, struct memory_block, dev); |
3947be19 DH |
180 | ssize_t len = 0; |
181 | ||
182 | /* | |
183 | * We can probably put these states in a nice little array | |
184 | * so that they're not open-coded | |
185 | */ | |
186 | switch (mem->state) { | |
187 | case MEM_ONLINE: | |
188 | len = sprintf(buf, "online\n"); | |
189 | break; | |
190 | case MEM_OFFLINE: | |
191 | len = sprintf(buf, "offline\n"); | |
192 | break; | |
193 | case MEM_GOING_OFFLINE: | |
194 | len = sprintf(buf, "going-offline\n"); | |
195 | break; | |
196 | default: | |
197 | len = sprintf(buf, "ERROR-UNKNOWN-%ld\n", | |
198 | mem->state); | |
199 | WARN_ON(1); | |
200 | break; | |
201 | } | |
202 | ||
203 | return len; | |
204 | } | |
205 | ||
7b78d335 | 206 | int memory_notify(unsigned long val, void *v) |
3947be19 | 207 | { |
e041c683 | 208 | return blocking_notifier_call_chain(&memory_chain, val, v); |
3947be19 DH |
209 | } |
210 | ||
925cc71e RJ |
211 | int memory_isolate_notify(unsigned long val, void *v) |
212 | { | |
213 | return atomic_notifier_call_chain(&memory_isolate_chain, val, v); | |
214 | } | |
215 | ||
2bbcb878 MG |
216 | /* |
217 | * The probe routines leave the pages reserved, just as the bootmem code does. | |
218 | * Make sure they're still that way. | |
219 | */ | |
220 | static bool pages_correctly_reserved(unsigned long start_pfn, | |
221 | unsigned long nr_pages) | |
222 | { | |
223 | int i, j; | |
224 | struct page *page; | |
225 | unsigned long pfn = start_pfn; | |
226 | ||
227 | /* | |
228 | * memmap between sections is not contiguous except with | |
229 | * SPARSEMEM_VMEMMAP. We lookup the page once per section | |
230 | * and assume memmap is contiguous within each section | |
231 | */ | |
232 | for (i = 0; i < sections_per_block; i++, pfn += PAGES_PER_SECTION) { | |
233 | if (WARN_ON_ONCE(!pfn_valid(pfn))) | |
234 | return false; | |
235 | page = pfn_to_page(pfn); | |
236 | ||
237 | for (j = 0; j < PAGES_PER_SECTION; j++) { | |
238 | if (PageReserved(page + j)) | |
239 | continue; | |
240 | ||
241 | printk(KERN_WARNING "section number %ld page number %d " | |
242 | "not reserved, was it already online?\n", | |
243 | pfn_to_section_nr(pfn), j); | |
244 | ||
245 | return false; | |
246 | } | |
247 | } | |
248 | ||
249 | return true; | |
250 | } | |
251 | ||
3947be19 DH |
252 | /* |
253 | * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is | |
254 | * OK to have direct references to sparsemem variables in here. | |
255 | */ | |
256 | static int | |
511c2aba | 257 | memory_block_action(unsigned long phys_index, unsigned long action, int online_type) |
3947be19 | 258 | { |
a16cee10 | 259 | unsigned long start_pfn; |
5409d2cd | 260 | unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block; |
de0ed36a | 261 | struct page *first_page; |
3947be19 | 262 | int ret; |
3947be19 | 263 | |
de0ed36a | 264 | first_page = pfn_to_page(phys_index << PFN_SECTION_SHIFT); |
a16cee10 | 265 | start_pfn = page_to_pfn(first_page); |
de0ed36a | 266 | |
3947be19 DH |
267 | switch (action) { |
268 | case MEM_ONLINE: | |
2bbcb878 MG |
269 | if (!pages_correctly_reserved(start_pfn, nr_pages)) |
270 | return -EBUSY; | |
271 | ||
511c2aba | 272 | ret = online_pages(start_pfn, nr_pages, online_type); |
3947be19 DH |
273 | break; |
274 | case MEM_OFFLINE: | |
a16cee10 | 275 | ret = offline_pages(start_pfn, nr_pages); |
3947be19 DH |
276 | break; |
277 | default: | |
0c2c99b1 NF |
278 | WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: " |
279 | "%ld\n", __func__, phys_index, action, action); | |
3947be19 DH |
280 | ret = -EINVAL; |
281 | } | |
3947be19 DH |
282 | |
283 | return ret; | |
284 | } | |
285 | ||
e90bdb7f | 286 | static int __memory_block_change_state(struct memory_block *mem, |
511c2aba LJ |
287 | unsigned long to_state, unsigned long from_state_req, |
288 | int online_type) | |
3947be19 | 289 | { |
de0ed36a | 290 | int ret = 0; |
0c2c99b1 | 291 | |
3947be19 DH |
292 | if (mem->state != from_state_req) { |
293 | ret = -EINVAL; | |
294 | goto out; | |
295 | } | |
296 | ||
0c2c99b1 NF |
297 | if (to_state == MEM_OFFLINE) |
298 | mem->state = MEM_GOING_OFFLINE; | |
299 | ||
511c2aba | 300 | ret = memory_block_action(mem->start_section_nr, to_state, online_type); |
0c2c99b1 | 301 | |
f5138e42 | 302 | if (ret) { |
0c2c99b1 | 303 | mem->state = from_state_req; |
f5138e42 MH |
304 | goto out; |
305 | } | |
3947be19 | 306 | |
f5138e42 MH |
307 | mem->state = to_state; |
308 | switch (mem->state) { | |
309 | case MEM_OFFLINE: | |
310 | kobject_uevent(&mem->dev.kobj, KOBJ_OFFLINE); | |
311 | break; | |
312 | case MEM_ONLINE: | |
313 | kobject_uevent(&mem->dev.kobj, KOBJ_ONLINE); | |
314 | break; | |
315 | default: | |
316 | break; | |
317 | } | |
3947be19 | 318 | out: |
3947be19 DH |
319 | return ret; |
320 | } | |
321 | ||
e90bdb7f | 322 | static int memory_block_change_state(struct memory_block *mem, |
511c2aba LJ |
323 | unsigned long to_state, unsigned long from_state_req, |
324 | int online_type) | |
e90bdb7f WC |
325 | { |
326 | int ret; | |
327 | ||
328 | mutex_lock(&mem->state_mutex); | |
511c2aba LJ |
329 | ret = __memory_block_change_state(mem, to_state, from_state_req, |
330 | online_type); | |
e90bdb7f WC |
331 | mutex_unlock(&mem->state_mutex); |
332 | ||
333 | return ret; | |
334 | } | |
3947be19 | 335 | static ssize_t |
10fbcf4c KS |
336 | store_mem_state(struct device *dev, |
337 | struct device_attribute *attr, const char *buf, size_t count) | |
3947be19 DH |
338 | { |
339 | struct memory_block *mem; | |
3947be19 DH |
340 | int ret = -EINVAL; |
341 | ||
10fbcf4c | 342 | mem = container_of(dev, struct memory_block, dev); |
3947be19 | 343 | |
511c2aba LJ |
344 | if (!strncmp(buf, "online_kernel", min_t(int, count, 13))) |
345 | ret = memory_block_change_state(mem, MEM_ONLINE, | |
346 | MEM_OFFLINE, ONLINE_KERNEL); | |
347 | else if (!strncmp(buf, "online_movable", min_t(int, count, 14))) | |
348 | ret = memory_block_change_state(mem, MEM_ONLINE, | |
349 | MEM_OFFLINE, ONLINE_MOVABLE); | |
350 | else if (!strncmp(buf, "online", min_t(int, count, 6))) | |
351 | ret = memory_block_change_state(mem, MEM_ONLINE, | |
352 | MEM_OFFLINE, ONLINE_KEEP); | |
353 | else if(!strncmp(buf, "offline", min_t(int, count, 7))) | |
354 | ret = memory_block_change_state(mem, MEM_OFFLINE, | |
355 | MEM_ONLINE, -1); | |
0c2c99b1 | 356 | |
3947be19 DH |
357 | if (ret) |
358 | return ret; | |
359 | return count; | |
360 | } | |
361 | ||
362 | /* | |
363 | * phys_device is a bad name for this. What I really want | |
364 | * is a way to differentiate between memory ranges that | |
365 | * are part of physical devices that constitute | |
366 | * a complete removable unit or fru. | |
367 | * i.e. do these ranges belong to the same physical device, | |
368 | * s.t. if I offline all of these sections I can then | |
369 | * remove the physical device? | |
370 | */ | |
10fbcf4c KS |
371 | static ssize_t show_phys_device(struct device *dev, |
372 | struct device_attribute *attr, char *buf) | |
3947be19 DH |
373 | { |
374 | struct memory_block *mem = | |
10fbcf4c | 375 | container_of(dev, struct memory_block, dev); |
3947be19 DH |
376 | return sprintf(buf, "%d\n", mem->phys_device); |
377 | } | |
378 | ||
10fbcf4c KS |
379 | static DEVICE_ATTR(phys_index, 0444, show_mem_start_phys_index, NULL); |
380 | static DEVICE_ATTR(end_phys_index, 0444, show_mem_end_phys_index, NULL); | |
381 | static DEVICE_ATTR(state, 0644, show_mem_state, store_mem_state); | |
382 | static DEVICE_ATTR(phys_device, 0444, show_phys_device, NULL); | |
383 | static DEVICE_ATTR(removable, 0444, show_mem_removable, NULL); | |
3947be19 DH |
384 | |
385 | #define mem_create_simple_file(mem, attr_name) \ | |
10fbcf4c | 386 | device_create_file(&mem->dev, &dev_attr_##attr_name) |
3947be19 | 387 | #define mem_remove_simple_file(mem, attr_name) \ |
10fbcf4c | 388 | device_remove_file(&mem->dev, &dev_attr_##attr_name) |
3947be19 DH |
389 | |
390 | /* | |
391 | * Block size attribute stuff | |
392 | */ | |
393 | static ssize_t | |
10fbcf4c | 394 | print_block_size(struct device *dev, struct device_attribute *attr, |
8564a6c1 | 395 | char *buf) |
3947be19 | 396 | { |
0c2c99b1 | 397 | return sprintf(buf, "%lx\n", get_memory_block_size()); |
3947be19 DH |
398 | } |
399 | ||
10fbcf4c | 400 | static DEVICE_ATTR(block_size_bytes, 0444, print_block_size, NULL); |
3947be19 DH |
401 | |
402 | static int block_size_init(void) | |
403 | { | |
10fbcf4c KS |
404 | return device_create_file(memory_subsys.dev_root, |
405 | &dev_attr_block_size_bytes); | |
3947be19 DH |
406 | } |
407 | ||
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 | |
415 | static ssize_t | |
10fbcf4c | 416 | memory_probe_store(struct device *dev, struct device_attribute *attr, |
28812fe1 | 417 | const char *buf, size_t count) |
3947be19 DH |
418 | { |
419 | u64 phys_addr; | |
bc02af93 | 420 | int nid; |
6add7cd6 | 421 | int i, ret; |
61b94fea | 422 | unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block; |
3947be19 DH |
423 | |
424 | phys_addr = simple_strtoull(buf, NULL, 0); | |
425 | ||
61b94fea AB |
426 | if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1)) |
427 | return -EINVAL; | |
428 | ||
6add7cd6 NF |
429 | for (i = 0; i < sections_per_block; i++) { |
430 | nid = memory_add_physaddr_to_nid(phys_addr); | |
431 | ret = add_memory(nid, phys_addr, | |
432 | PAGES_PER_SECTION << PAGE_SHIFT); | |
433 | if (ret) | |
9f0af69b | 434 | goto out; |
6add7cd6 NF |
435 | |
436 | phys_addr += MIN_MEMORY_BLOCK_SIZE; | |
437 | } | |
3947be19 | 438 | |
9f0af69b NK |
439 | ret = count; |
440 | out: | |
441 | return ret; | |
3947be19 | 442 | } |
10fbcf4c | 443 | static DEVICE_ATTR(probe, S_IWUSR, NULL, memory_probe_store); |
3947be19 DH |
444 | |
445 | static int memory_probe_init(void) | |
446 | { | |
10fbcf4c | 447 | return device_create_file(memory_subsys.dev_root, &dev_attr_probe); |
3947be19 DH |
448 | } |
449 | #else | |
28ec24e2 AM |
450 | static inline int memory_probe_init(void) |
451 | { | |
452 | return 0; | |
453 | } | |
3947be19 DH |
454 | #endif |
455 | ||
facb6011 AK |
456 | #ifdef CONFIG_MEMORY_FAILURE |
457 | /* | |
458 | * Support for offlining pages of memory | |
459 | */ | |
460 | ||
461 | /* Soft offline a page */ | |
462 | static ssize_t | |
10fbcf4c KS |
463 | store_soft_offline_page(struct device *dev, |
464 | struct device_attribute *attr, | |
28812fe1 | 465 | const char *buf, size_t count) |
facb6011 AK |
466 | { |
467 | int ret; | |
468 | u64 pfn; | |
469 | if (!capable(CAP_SYS_ADMIN)) | |
470 | return -EPERM; | |
471 | if (strict_strtoull(buf, 0, &pfn) < 0) | |
472 | return -EINVAL; | |
473 | pfn >>= PAGE_SHIFT; | |
474 | if (!pfn_valid(pfn)) | |
475 | return -ENXIO; | |
476 | ret = soft_offline_page(pfn_to_page(pfn), 0); | |
477 | return ret == 0 ? count : ret; | |
478 | } | |
479 | ||
480 | /* Forcibly offline a page, including killing processes. */ | |
481 | static ssize_t | |
10fbcf4c KS |
482 | store_hard_offline_page(struct device *dev, |
483 | struct device_attribute *attr, | |
28812fe1 | 484 | const char *buf, size_t count) |
facb6011 AK |
485 | { |
486 | int ret; | |
487 | u64 pfn; | |
488 | if (!capable(CAP_SYS_ADMIN)) | |
489 | return -EPERM; | |
490 | if (strict_strtoull(buf, 0, &pfn) < 0) | |
491 | return -EINVAL; | |
492 | pfn >>= PAGE_SHIFT; | |
cd42f4a3 | 493 | ret = memory_failure(pfn, 0, 0); |
facb6011 AK |
494 | return ret ? ret : count; |
495 | } | |
496 | ||
74fef7a8 FB |
497 | static DEVICE_ATTR(soft_offline_page, S_IWUSR, NULL, store_soft_offline_page); |
498 | static DEVICE_ATTR(hard_offline_page, S_IWUSR, NULL, store_hard_offline_page); | |
facb6011 AK |
499 | |
500 | static __init int memory_fail_init(void) | |
501 | { | |
502 | int err; | |
503 | ||
10fbcf4c KS |
504 | err = device_create_file(memory_subsys.dev_root, |
505 | &dev_attr_soft_offline_page); | |
facb6011 | 506 | if (!err) |
10fbcf4c KS |
507 | err = device_create_file(memory_subsys.dev_root, |
508 | &dev_attr_hard_offline_page); | |
facb6011 AK |
509 | return err; |
510 | } | |
511 | #else | |
512 | static inline int memory_fail_init(void) | |
513 | { | |
514 | return 0; | |
515 | } | |
516 | #endif | |
517 | ||
3947be19 DH |
518 | /* |
519 | * Note that phys_device is optional. It is here to allow for | |
520 | * differentiation between which *physical* devices each | |
521 | * section belongs to... | |
522 | */ | |
bc32df00 HC |
523 | int __weak arch_get_memory_phys_device(unsigned long start_pfn) |
524 | { | |
525 | return 0; | |
526 | } | |
3947be19 | 527 | |
10fbcf4c KS |
528 | /* |
529 | * A reference for the returned object is held and the reference for the | |
530 | * hinted object is released. | |
531 | */ | |
98383031 RH |
532 | struct memory_block *find_memory_block_hinted(struct mem_section *section, |
533 | struct memory_block *hint) | |
3947be19 | 534 | { |
0c2c99b1 | 535 | int block_id = base_memory_block_id(__section_nr(section)); |
10fbcf4c KS |
536 | struct device *hintdev = hint ? &hint->dev : NULL; |
537 | struct device *dev; | |
3947be19 | 538 | |
10fbcf4c KS |
539 | dev = subsys_find_device_by_id(&memory_subsys, block_id, hintdev); |
540 | if (hint) | |
541 | put_device(&hint->dev); | |
542 | if (!dev) | |
3947be19 | 543 | return NULL; |
10fbcf4c | 544 | return container_of(dev, struct memory_block, dev); |
3947be19 DH |
545 | } |
546 | ||
98383031 RH |
547 | /* |
548 | * For now, we have a linear search to go find the appropriate | |
549 | * memory_block corresponding to a particular phys_index. If | |
550 | * this gets to be a real problem, we can always use a radix | |
551 | * tree or something here. | |
552 | * | |
10fbcf4c | 553 | * This could be made generic for all device subsystems. |
98383031 RH |
554 | */ |
555 | struct memory_block *find_memory_block(struct mem_section *section) | |
556 | { | |
557 | return find_memory_block_hinted(section, NULL); | |
558 | } | |
559 | ||
0c2c99b1 NF |
560 | static int init_memory_block(struct memory_block **memory, |
561 | struct mem_section *section, unsigned long state) | |
e4619c85 | 562 | { |
0c2c99b1 | 563 | struct memory_block *mem; |
e4619c85 | 564 | unsigned long start_pfn; |
0c2c99b1 | 565 | int scn_nr; |
e4619c85 NF |
566 | int ret = 0; |
567 | ||
0c2c99b1 | 568 | mem = kzalloc(sizeof(*mem), GFP_KERNEL); |
e4619c85 NF |
569 | if (!mem) |
570 | return -ENOMEM; | |
571 | ||
0c2c99b1 | 572 | scn_nr = __section_nr(section); |
d3360164 NF |
573 | mem->start_section_nr = |
574 | base_memory_block_id(scn_nr) * sections_per_block; | |
575 | mem->end_section_nr = mem->start_section_nr + sections_per_block - 1; | |
e4619c85 | 576 | mem->state = state; |
07681215 | 577 | mem->section_count++; |
e4619c85 | 578 | mutex_init(&mem->state_mutex); |
d3360164 | 579 | start_pfn = section_nr_to_pfn(mem->start_section_nr); |
e4619c85 NF |
580 | mem->phys_device = arch_get_memory_phys_device(start_pfn); |
581 | ||
0c2c99b1 | 582 | ret = register_memory(mem); |
e4619c85 NF |
583 | if (!ret) |
584 | ret = mem_create_simple_file(mem, phys_index); | |
d3360164 NF |
585 | if (!ret) |
586 | ret = mem_create_simple_file(mem, end_phys_index); | |
e4619c85 NF |
587 | if (!ret) |
588 | ret = mem_create_simple_file(mem, state); | |
589 | if (!ret) | |
590 | ret = mem_create_simple_file(mem, phys_device); | |
591 | if (!ret) | |
592 | ret = mem_create_simple_file(mem, removable); | |
0c2c99b1 NF |
593 | |
594 | *memory = mem; | |
595 | return ret; | |
596 | } | |
597 | ||
598 | static int add_memory_section(int nid, struct mem_section *section, | |
321bf4ed | 599 | struct memory_block **mem_p, |
0c2c99b1 NF |
600 | unsigned long state, enum mem_add_context context) |
601 | { | |
321bf4ed YL |
602 | struct memory_block *mem = NULL; |
603 | int scn_nr = __section_nr(section); | |
0c2c99b1 NF |
604 | int ret = 0; |
605 | ||
606 | mutex_lock(&mem_sysfs_mutex); | |
607 | ||
321bf4ed YL |
608 | if (context == BOOT) { |
609 | /* same memory block ? */ | |
610 | if (mem_p && *mem_p) | |
611 | if (scn_nr >= (*mem_p)->start_section_nr && | |
612 | scn_nr <= (*mem_p)->end_section_nr) { | |
613 | mem = *mem_p; | |
614 | kobject_get(&mem->dev.kobj); | |
615 | } | |
616 | } else | |
617 | mem = find_memory_block(section); | |
618 | ||
0c2c99b1 NF |
619 | if (mem) { |
620 | mem->section_count++; | |
10fbcf4c | 621 | kobject_put(&mem->dev.kobj); |
321bf4ed | 622 | } else { |
0c2c99b1 | 623 | ret = init_memory_block(&mem, section, state); |
321bf4ed YL |
624 | /* store memory_block pointer for next loop */ |
625 | if (!ret && context == BOOT) | |
626 | if (mem_p) | |
627 | *mem_p = mem; | |
628 | } | |
0c2c99b1 | 629 | |
e4619c85 | 630 | if (!ret) { |
0c2c99b1 NF |
631 | if (context == HOTPLUG && |
632 | mem->section_count == sections_per_block) | |
e4619c85 NF |
633 | ret = register_mem_sect_under_node(mem, nid); |
634 | } | |
635 | ||
2938ffbd | 636 | mutex_unlock(&mem_sysfs_mutex); |
e4619c85 NF |
637 | return ret; |
638 | } | |
639 | ||
3947be19 DH |
640 | int remove_memory_block(unsigned long node_id, struct mem_section *section, |
641 | int phys_device) | |
642 | { | |
643 | struct memory_block *mem; | |
644 | ||
2938ffbd | 645 | mutex_lock(&mem_sysfs_mutex); |
3947be19 | 646 | mem = find_memory_block(section); |
d3360164 | 647 | unregister_mem_sect_under_nodes(mem, __section_nr(section)); |
07681215 NF |
648 | |
649 | mem->section_count--; | |
650 | if (mem->section_count == 0) { | |
07681215 | 651 | mem_remove_simple_file(mem, phys_index); |
d3360164 | 652 | mem_remove_simple_file(mem, end_phys_index); |
07681215 NF |
653 | mem_remove_simple_file(mem, state); |
654 | mem_remove_simple_file(mem, phys_device); | |
655 | mem_remove_simple_file(mem, removable); | |
0c2c99b1 | 656 | unregister_memory(mem); |
0c2c99b1 | 657 | } else |
10fbcf4c | 658 | kobject_put(&mem->dev.kobj); |
3947be19 | 659 | |
2938ffbd | 660 | mutex_unlock(&mem_sysfs_mutex); |
3947be19 DH |
661 | return 0; |
662 | } | |
663 | ||
664 | /* | |
665 | * need an interface for the VM to add new memory regions, | |
666 | * but without onlining it. | |
667 | */ | |
c04fc586 | 668 | int register_new_memory(int nid, struct mem_section *section) |
3947be19 | 669 | { |
321bf4ed | 670 | return add_memory_section(nid, section, NULL, MEM_OFFLINE, HOTPLUG); |
3947be19 DH |
671 | } |
672 | ||
673 | int unregister_memory_section(struct mem_section *section) | |
674 | { | |
540557b9 | 675 | if (!present_section(section)) |
3947be19 DH |
676 | return -EINVAL; |
677 | ||
678 | return remove_memory_block(0, section, 0); | |
679 | } | |
680 | ||
e90bdb7f WC |
681 | /* |
682 | * offline one memory block. If the memory block has been offlined, do nothing. | |
683 | */ | |
684 | int offline_memory_block(struct memory_block *mem) | |
685 | { | |
686 | int ret = 0; | |
687 | ||
688 | mutex_lock(&mem->state_mutex); | |
689 | if (mem->state != MEM_OFFLINE) | |
511c2aba | 690 | ret = __memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE, -1); |
e90bdb7f WC |
691 | mutex_unlock(&mem->state_mutex); |
692 | ||
693 | return ret; | |
694 | } | |
695 | ||
3947be19 DH |
696 | /* |
697 | * Initialize the sysfs support for memory devices... | |
698 | */ | |
699 | int __init memory_dev_init(void) | |
700 | { | |
701 | unsigned int i; | |
702 | int ret; | |
28ec24e2 | 703 | int err; |
0c2c99b1 | 704 | unsigned long block_sz; |
321bf4ed | 705 | struct memory_block *mem = NULL; |
3947be19 | 706 | |
10fbcf4c | 707 | ret = subsys_system_register(&memory_subsys, NULL); |
28ec24e2 AM |
708 | if (ret) |
709 | goto out; | |
3947be19 | 710 | |
0c2c99b1 NF |
711 | block_sz = get_memory_block_size(); |
712 | sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE; | |
713 | ||
3947be19 DH |
714 | /* |
715 | * Create entries for memory sections that were found | |
716 | * during boot and have been initialized | |
717 | */ | |
718 | for (i = 0; i < NR_MEM_SECTIONS; i++) { | |
540557b9 | 719 | if (!present_section_nr(i)) |
3947be19 | 720 | continue; |
321bf4ed YL |
721 | /* don't need to reuse memory_block if only one per block */ |
722 | err = add_memory_section(0, __nr_to_section(i), | |
723 | (sections_per_block == 1) ? NULL : &mem, | |
724 | MEM_ONLINE, | |
0c2c99b1 | 725 | BOOT); |
28ec24e2 AM |
726 | if (!ret) |
727 | ret = err; | |
3947be19 DH |
728 | } |
729 | ||
28ec24e2 | 730 | err = memory_probe_init(); |
facb6011 AK |
731 | if (!ret) |
732 | ret = err; | |
733 | err = memory_fail_init(); | |
28ec24e2 AM |
734 | if (!ret) |
735 | ret = err; | |
736 | err = block_size_init(); | |
737 | if (!ret) | |
738 | ret = err; | |
739 | out: | |
740 | if (ret) | |
2b3a302a | 741 | printk(KERN_ERR "%s() failed: %d\n", __func__, ret); |
3947be19 DH |
742 | return ret; |
743 | } |