blob: 3da6a43b77567d622f0768e6049c71533e55ed0e [file] [log] [blame]
Dave Hansen3947be12005-10-29 18:16:54 -07001/*
2 * drivers/base/memory.c - basic Memory class support
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
13#include <linux/sysdev.h>
14#include <linux/module.h>
15#include <linux/init.h>
Dave Hansen3947be12005-10-29 18:16:54 -070016#include <linux/topology.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080017#include <linux/capability.h>
Dave Hansen3947be12005-10-29 18:16:54 -070018#include <linux/device.h>
19#include <linux/memory.h>
20#include <linux/kobject.h>
21#include <linux/memory_hotplug.h>
22#include <linux/mm.h>
Daniel Walkerda19cbc2008-02-04 23:35:47 -080023#include <linux/mutex.h>
Shaohua Li9f1b16a2008-10-18 20:27:12 -070024#include <linux/stat.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090025#include <linux/slab.h>
Shaohua Li9f1b16a2008-10-18 20:27:12 -070026
Dave Hansen3947be12005-10-29 18:16:54 -070027#include <asm/atomic.h>
28#include <asm/uaccess.h>
29
Nathan Fontenot2938ffb2010-10-19 12:45:24 -050030static DEFINE_MUTEX(mem_sysfs_mutex);
31
Dave Hansen3947be12005-10-29 18:16:54 -070032#define MEMORY_CLASS_NAME "memory"
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -060033#define MIN_MEMORY_BLOCK_SIZE (1 << SECTION_SIZE_BITS)
34
35static int sections_per_block;
36
37static inline int base_memory_block_id(int section_nr)
38{
39 return section_nr / sections_per_block;
40}
Dave Hansen3947be12005-10-29 18:16:54 -070041
42static struct sysdev_class memory_sysdev_class = {
Kay Sieversaf5ca3f2007-12-20 02:09:39 +010043 .name = MEMORY_CLASS_NAME,
Dave Hansen3947be12005-10-29 18:16:54 -070044};
Dave Hansen3947be12005-10-29 18:16:54 -070045
Kay Sievers312c0042005-11-16 09:00:00 +010046static const char *memory_uevent_name(struct kset *kset, struct kobject *kobj)
Dave Hansen3947be12005-10-29 18:16:54 -070047{
48 return MEMORY_CLASS_NAME;
49}
50
Al Viro9ec0fd42007-10-14 06:53:45 +010051static int memory_uevent(struct kset *kset, struct kobject *obj, struct kobj_uevent_env *env)
Dave Hansen3947be12005-10-29 18:16:54 -070052{
53 int retval = 0;
54
55 return retval;
56}
57
Emese Revfy9cd43612009-12-31 14:52:51 +010058static const struct kset_uevent_ops memory_uevent_ops = {
Kay Sievers312c0042005-11-16 09:00:00 +010059 .name = memory_uevent_name,
60 .uevent = memory_uevent,
Dave Hansen3947be12005-10-29 18:16:54 -070061};
62
Alan Sterne041c682006-03-27 01:16:30 -080063static BLOCKING_NOTIFIER_HEAD(memory_chain);
Dave Hansen3947be12005-10-29 18:16:54 -070064
Andy Whitcroft98a38eb2006-01-06 00:10:35 -080065int register_memory_notifier(struct notifier_block *nb)
Dave Hansen3947be12005-10-29 18:16:54 -070066{
Alan Sterne041c682006-03-27 01:16:30 -080067 return blocking_notifier_chain_register(&memory_chain, nb);
Dave Hansen3947be12005-10-29 18:16:54 -070068}
Hannes Hering3c82c302008-05-07 14:43:01 +020069EXPORT_SYMBOL(register_memory_notifier);
Dave Hansen3947be12005-10-29 18:16:54 -070070
Andy Whitcroft98a38eb2006-01-06 00:10:35 -080071void unregister_memory_notifier(struct notifier_block *nb)
Dave Hansen3947be12005-10-29 18:16:54 -070072{
Alan Sterne041c682006-03-27 01:16:30 -080073 blocking_notifier_chain_unregister(&memory_chain, nb);
Dave Hansen3947be12005-10-29 18:16:54 -070074}
Hannes Hering3c82c302008-05-07 14:43:01 +020075EXPORT_SYMBOL(unregister_memory_notifier);
Dave Hansen3947be12005-10-29 18:16:54 -070076
Robert Jennings925cc712009-12-17 14:44:38 +000077static ATOMIC_NOTIFIER_HEAD(memory_isolate_chain);
78
79int register_memory_isolate_notifier(struct notifier_block *nb)
80{
81 return atomic_notifier_chain_register(&memory_isolate_chain, nb);
82}
83EXPORT_SYMBOL(register_memory_isolate_notifier);
84
85void unregister_memory_isolate_notifier(struct notifier_block *nb)
86{
87 atomic_notifier_chain_unregister(&memory_isolate_chain, nb);
88}
89EXPORT_SYMBOL(unregister_memory_isolate_notifier);
90
Dave Hansen3947be12005-10-29 18:16:54 -070091/*
92 * register_memory - Setup a sysfs device for a memory block
93 */
Badari Pulavarty00a41db2008-02-11 09:23:18 -080094static
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -060095int register_memory(struct memory_block *memory)
Dave Hansen3947be12005-10-29 18:16:54 -070096{
97 int error;
98
99 memory->sysdev.cls = &memory_sysdev_class;
Nathan Fontenotd3360162011-01-20 10:44:29 -0600100 memory->sysdev.id = memory->start_section_nr / sections_per_block;
Dave Hansen3947be12005-10-29 18:16:54 -0700101
102 error = sysdev_register(&memory->sysdev);
Dave Hansen3947be12005-10-29 18:16:54 -0700103 return error;
104}
105
106static void
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600107unregister_memory(struct memory_block *memory)
Dave Hansen3947be12005-10-29 18:16:54 -0700108{
109 BUG_ON(memory->sysdev.cls != &memory_sysdev_class);
Dave Hansen3947be12005-10-29 18:16:54 -0700110
Badari Pulavarty00a41db2008-02-11 09:23:18 -0800111 /* drop the ref. we got in remove_memory_block() */
112 kobject_put(&memory->sysdev.kobj);
Dave Hansen3947be12005-10-29 18:16:54 -0700113 sysdev_unregister(&memory->sysdev);
Dave Hansen3947be12005-10-29 18:16:54 -0700114}
115
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600116unsigned long __weak memory_block_size_bytes(void)
117{
118 return MIN_MEMORY_BLOCK_SIZE;
119}
120
121static unsigned long get_memory_block_size(void)
122{
123 unsigned long block_sz;
124
125 block_sz = memory_block_size_bytes();
126
127 /* Validate blk_sz is a power of 2 and not less than section size */
128 if ((block_sz & (block_sz - 1)) || (block_sz < MIN_MEMORY_BLOCK_SIZE)) {
129 WARN_ON(1);
130 block_sz = MIN_MEMORY_BLOCK_SIZE;
131 }
132
133 return block_sz;
134}
135
Dave Hansen3947be12005-10-29 18:16:54 -0700136/*
137 * use this as the physical section index that this memsection
138 * uses.
139 */
140
Nathan Fontenotd3360162011-01-20 10:44:29 -0600141static ssize_t show_mem_start_phys_index(struct sys_device *dev,
Andi Kleen4a0b2b42008-07-01 18:48:41 +0200142 struct sysdev_attribute *attr, char *buf)
Dave Hansen3947be12005-10-29 18:16:54 -0700143{
144 struct memory_block *mem =
145 container_of(dev, struct memory_block, sysdev);
Nathan Fontenotd3360162011-01-20 10:44:29 -0600146 unsigned long phys_index;
147
148 phys_index = mem->start_section_nr / sections_per_block;
149 return sprintf(buf, "%08lx\n", phys_index);
150}
151
152static ssize_t show_mem_end_phys_index(struct sys_device *dev,
153 struct sysdev_attribute *attr, char *buf)
154{
155 struct memory_block *mem =
156 container_of(dev, struct memory_block, sysdev);
157 unsigned long phys_index;
158
159 phys_index = mem->end_section_nr / sections_per_block;
160 return sprintf(buf, "%08lx\n", phys_index);
Dave Hansen3947be12005-10-29 18:16:54 -0700161}
162
163/*
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700164 * Show whether the section of memory is likely to be hot-removable
165 */
Stephen Rothwell1f07be12008-07-28 11:05:04 +1000166static ssize_t show_mem_removable(struct sys_device *dev,
167 struct sysdev_attribute *attr, char *buf)
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700168{
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600169 unsigned long i, pfn;
170 int ret = 1;
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700171 struct memory_block *mem =
172 container_of(dev, struct memory_block, sysdev);
173
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600174 for (i = 0; i < sections_per_block; i++) {
Nathan Fontenotd3360162011-01-20 10:44:29 -0600175 pfn = section_nr_to_pfn(mem->start_section_nr + i);
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600176 ret &= is_mem_section_removable(pfn, PAGES_PER_SECTION);
177 }
178
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700179 return sprintf(buf, "%d\n", ret);
180}
181
182/*
Dave Hansen3947be12005-10-29 18:16:54 -0700183 * online, offline, going offline, etc.
184 */
Andi Kleen4a0b2b42008-07-01 18:48:41 +0200185static ssize_t show_mem_state(struct sys_device *dev,
186 struct sysdev_attribute *attr, char *buf)
Dave Hansen3947be12005-10-29 18:16:54 -0700187{
188 struct memory_block *mem =
189 container_of(dev, struct memory_block, sysdev);
190 ssize_t len = 0;
191
192 /*
193 * We can probably put these states in a nice little array
194 * so that they're not open-coded
195 */
196 switch (mem->state) {
197 case MEM_ONLINE:
198 len = sprintf(buf, "online\n");
199 break;
200 case MEM_OFFLINE:
201 len = sprintf(buf, "offline\n");
202 break;
203 case MEM_GOING_OFFLINE:
204 len = sprintf(buf, "going-offline\n");
205 break;
206 default:
207 len = sprintf(buf, "ERROR-UNKNOWN-%ld\n",
208 mem->state);
209 WARN_ON(1);
210 break;
211 }
212
213 return len;
214}
215
Yasunori Goto7b78d332007-10-21 16:41:36 -0700216int memory_notify(unsigned long val, void *v)
Dave Hansen3947be12005-10-29 18:16:54 -0700217{
Alan Sterne041c682006-03-27 01:16:30 -0800218 return blocking_notifier_call_chain(&memory_chain, val, v);
Dave Hansen3947be12005-10-29 18:16:54 -0700219}
220
Robert Jennings925cc712009-12-17 14:44:38 +0000221int memory_isolate_notify(unsigned long val, void *v)
222{
223 return atomic_notifier_call_chain(&memory_isolate_chain, val, v);
224}
225
Dave Hansen3947be12005-10-29 18:16:54 -0700226/*
227 * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is
228 * OK to have direct references to sparsemem variables in here.
229 */
230static int
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600231memory_section_action(unsigned long phys_index, unsigned long action)
Dave Hansen3947be12005-10-29 18:16:54 -0700232{
233 int i;
Dave Hansen3947be12005-10-29 18:16:54 -0700234 unsigned long start_pfn, start_paddr;
235 struct page *first_page;
236 int ret;
Dave Hansen3947be12005-10-29 18:16:54 -0700237
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600238 first_page = pfn_to_page(phys_index << PFN_SECTION_SHIFT);
Dave Hansen3947be12005-10-29 18:16:54 -0700239
240 /*
241 * The probe routines leave the pages reserved, just
242 * as the bootmem code does. Make sure they're still
243 * that way.
244 */
245 if (action == MEM_ONLINE) {
246 for (i = 0; i < PAGES_PER_SECTION; i++) {
247 if (PageReserved(first_page+i))
248 continue;
249
250 printk(KERN_WARNING "section number %ld page number %d "
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600251 "not reserved, was it already online?\n",
252 phys_index, i);
Dave Hansen3947be12005-10-29 18:16:54 -0700253 return -EBUSY;
254 }
255 }
256
257 switch (action) {
258 case MEM_ONLINE:
259 start_pfn = page_to_pfn(first_page);
260 ret = online_pages(start_pfn, PAGES_PER_SECTION);
261 break;
262 case MEM_OFFLINE:
Dave Hansen3947be12005-10-29 18:16:54 -0700263 start_paddr = page_to_pfn(first_page) << PAGE_SHIFT;
264 ret = remove_memory(start_paddr,
265 PAGES_PER_SECTION << PAGE_SHIFT);
Dave Hansen3947be12005-10-29 18:16:54 -0700266 break;
267 default:
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600268 WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: "
269 "%ld\n", __func__, phys_index, action, action);
Dave Hansen3947be12005-10-29 18:16:54 -0700270 ret = -EINVAL;
271 }
Dave Hansen3947be12005-10-29 18:16:54 -0700272
273 return ret;
274}
275
276static int memory_block_change_state(struct memory_block *mem,
277 unsigned long to_state, unsigned long from_state_req)
278{
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600279 int i, ret = 0;
280
Daniel Walkerda19cbc2008-02-04 23:35:47 -0800281 mutex_lock(&mem->state_mutex);
Dave Hansen3947be12005-10-29 18:16:54 -0700282
283 if (mem->state != from_state_req) {
284 ret = -EINVAL;
285 goto out;
286 }
287
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600288 if (to_state == MEM_OFFLINE)
289 mem->state = MEM_GOING_OFFLINE;
290
291 for (i = 0; i < sections_per_block; i++) {
Nathan Fontenotd3360162011-01-20 10:44:29 -0600292 ret = memory_section_action(mem->start_section_nr + i,
293 to_state);
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600294 if (ret)
295 break;
296 }
297
298 if (ret) {
299 for (i = 0; i < sections_per_block; i++)
Nathan Fontenotd3360162011-01-20 10:44:29 -0600300 memory_section_action(mem->start_section_nr + i,
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600301 from_state_req);
302
303 mem->state = from_state_req;
304 } else
Dave Hansen3947be12005-10-29 18:16:54 -0700305 mem->state = to_state;
306
307out:
Daniel Walkerda19cbc2008-02-04 23:35:47 -0800308 mutex_unlock(&mem->state_mutex);
Dave Hansen3947be12005-10-29 18:16:54 -0700309 return ret;
310}
311
312static ssize_t
Andi Kleen4a0b2b42008-07-01 18:48:41 +0200313store_mem_state(struct sys_device *dev,
314 struct sysdev_attribute *attr, const char *buf, size_t count)
Dave Hansen3947be12005-10-29 18:16:54 -0700315{
316 struct memory_block *mem;
Dave Hansen3947be12005-10-29 18:16:54 -0700317 int ret = -EINVAL;
318
319 mem = container_of(dev, struct memory_block, sysdev);
Dave Hansen3947be12005-10-29 18:16:54 -0700320
321 if (!strncmp(buf, "online", min((int)count, 6)))
322 ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
323 else if(!strncmp(buf, "offline", min((int)count, 7)))
324 ret = memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE);
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600325
Dave Hansen3947be12005-10-29 18:16:54 -0700326 if (ret)
327 return ret;
328 return count;
329}
330
331/*
332 * phys_device is a bad name for this. What I really want
333 * is a way to differentiate between memory ranges that
334 * are part of physical devices that constitute
335 * a complete removable unit or fru.
336 * i.e. do these ranges belong to the same physical device,
337 * s.t. if I offline all of these sections I can then
338 * remove the physical device?
339 */
Andi Kleen4a0b2b42008-07-01 18:48:41 +0200340static ssize_t show_phys_device(struct sys_device *dev,
341 struct sysdev_attribute *attr, char *buf)
Dave Hansen3947be12005-10-29 18:16:54 -0700342{
343 struct memory_block *mem =
344 container_of(dev, struct memory_block, sysdev);
345 return sprintf(buf, "%d\n", mem->phys_device);
346}
347
Nathan Fontenotd3360162011-01-20 10:44:29 -0600348static SYSDEV_ATTR(phys_index, 0444, show_mem_start_phys_index, NULL);
349static SYSDEV_ATTR(end_phys_index, 0444, show_mem_end_phys_index, NULL);
Dave Hansen3947be12005-10-29 18:16:54 -0700350static SYSDEV_ATTR(state, 0644, show_mem_state, store_mem_state);
351static SYSDEV_ATTR(phys_device, 0444, show_phys_device, NULL);
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700352static SYSDEV_ATTR(removable, 0444, show_mem_removable, NULL);
Dave Hansen3947be12005-10-29 18:16:54 -0700353
354#define mem_create_simple_file(mem, attr_name) \
355 sysdev_create_file(&mem->sysdev, &attr_##attr_name)
356#define mem_remove_simple_file(mem, attr_name) \
357 sysdev_remove_file(&mem->sysdev, &attr_##attr_name)
358
359/*
360 * Block size attribute stuff
361 */
362static ssize_t
Andi Kleen8564a6c2010-01-05 12:48:06 +0100363print_block_size(struct sysdev_class *class, struct sysdev_class_attribute *attr,
364 char *buf)
Dave Hansen3947be12005-10-29 18:16:54 -0700365{
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600366 return sprintf(buf, "%lx\n", get_memory_block_size());
Dave Hansen3947be12005-10-29 18:16:54 -0700367}
368
Andi Kleen8564a6c2010-01-05 12:48:06 +0100369static SYSDEV_CLASS_ATTR(block_size_bytes, 0444, print_block_size, NULL);
Dave Hansen3947be12005-10-29 18:16:54 -0700370
371static int block_size_init(void)
372{
Andrew Morton28ec24e2006-12-06 20:37:29 -0800373 return sysfs_create_file(&memory_sysdev_class.kset.kobj,
Andi Kleen8564a6c2010-01-05 12:48:06 +0100374 &attr_block_size_bytes.attr);
Dave Hansen3947be12005-10-29 18:16:54 -0700375}
376
377/*
378 * Some architectures will have custom drivers to do this, and
379 * will not need to do it from userspace. The fake hot-add code
380 * as well as ppc64 will do all of their discovery in userspace
381 * and will require this interface.
382 */
383#ifdef CONFIG_ARCH_MEMORY_PROBE
384static ssize_t
Andi Kleen28812fe2010-01-05 12:48:07 +0100385memory_probe_store(struct class *class, struct class_attribute *attr,
386 const char *buf, size_t count)
Dave Hansen3947be12005-10-29 18:16:54 -0700387{
388 u64 phys_addr;
Yasunori Gotobc02af92006-06-27 02:53:30 -0700389 int nid;
Nathan Fontenot6add7cd2011-01-31 10:55:23 -0600390 int i, ret;
Dave Hansen3947be12005-10-29 18:16:54 -0700391
392 phys_addr = simple_strtoull(buf, NULL, 0);
393
Nathan Fontenot6add7cd2011-01-31 10:55:23 -0600394 for (i = 0; i < sections_per_block; i++) {
395 nid = memory_add_physaddr_to_nid(phys_addr);
396 ret = add_memory(nid, phys_addr,
397 PAGES_PER_SECTION << PAGE_SHIFT);
398 if (ret)
399 break;
400
401 phys_addr += MIN_MEMORY_BLOCK_SIZE;
402 }
Dave Hansen3947be12005-10-29 18:16:54 -0700403
404 if (ret)
405 count = ret;
406
407 return count;
408}
Greg Kroah-Hartmanbd796672010-01-19 13:08:16 -0800409static CLASS_ATTR(probe, S_IWUSR, NULL, memory_probe_store);
Dave Hansen3947be12005-10-29 18:16:54 -0700410
411static int memory_probe_init(void)
412{
Andrew Morton28ec24e2006-12-06 20:37:29 -0800413 return sysfs_create_file(&memory_sysdev_class.kset.kobj,
Greg Kroah-Hartmanbd796672010-01-19 13:08:16 -0800414 &class_attr_probe.attr);
Dave Hansen3947be12005-10-29 18:16:54 -0700415}
416#else
Andrew Morton28ec24e2006-12-06 20:37:29 -0800417static inline int memory_probe_init(void)
418{
419 return 0;
420}
Dave Hansen3947be12005-10-29 18:16:54 -0700421#endif
422
Andi Kleenfacb6012009-12-16 12:20:00 +0100423#ifdef CONFIG_MEMORY_FAILURE
424/*
425 * Support for offlining pages of memory
426 */
427
428/* Soft offline a page */
429static ssize_t
Andi Kleen28812fe2010-01-05 12:48:07 +0100430store_soft_offline_page(struct class *class,
431 struct class_attribute *attr,
432 const char *buf, size_t count)
Andi Kleenfacb6012009-12-16 12:20:00 +0100433{
434 int ret;
435 u64 pfn;
436 if (!capable(CAP_SYS_ADMIN))
437 return -EPERM;
438 if (strict_strtoull(buf, 0, &pfn) < 0)
439 return -EINVAL;
440 pfn >>= PAGE_SHIFT;
441 if (!pfn_valid(pfn))
442 return -ENXIO;
443 ret = soft_offline_page(pfn_to_page(pfn), 0);
444 return ret == 0 ? count : ret;
445}
446
447/* Forcibly offline a page, including killing processes. */
448static ssize_t
Andi Kleen28812fe2010-01-05 12:48:07 +0100449store_hard_offline_page(struct class *class,
450 struct class_attribute *attr,
451 const char *buf, size_t count)
Andi Kleenfacb6012009-12-16 12:20:00 +0100452{
453 int ret;
454 u64 pfn;
455 if (!capable(CAP_SYS_ADMIN))
456 return -EPERM;
457 if (strict_strtoull(buf, 0, &pfn) < 0)
458 return -EINVAL;
459 pfn >>= PAGE_SHIFT;
460 ret = __memory_failure(pfn, 0, 0);
461 return ret ? ret : count;
462}
463
Greg Kroah-Hartmanbd796672010-01-19 13:08:16 -0800464static CLASS_ATTR(soft_offline_page, 0644, NULL, store_soft_offline_page);
465static CLASS_ATTR(hard_offline_page, 0644, NULL, store_hard_offline_page);
Andi Kleenfacb6012009-12-16 12:20:00 +0100466
467static __init int memory_fail_init(void)
468{
469 int err;
470
471 err = sysfs_create_file(&memory_sysdev_class.kset.kobj,
Greg Kroah-Hartmanbd796672010-01-19 13:08:16 -0800472 &class_attr_soft_offline_page.attr);
Andi Kleenfacb6012009-12-16 12:20:00 +0100473 if (!err)
474 err = sysfs_create_file(&memory_sysdev_class.kset.kobj,
Greg Kroah-Hartmanbd796672010-01-19 13:08:16 -0800475 &class_attr_hard_offline_page.attr);
Andi Kleenfacb6012009-12-16 12:20:00 +0100476 return err;
477}
478#else
479static inline int memory_fail_init(void)
480{
481 return 0;
482}
483#endif
484
Dave Hansen3947be12005-10-29 18:16:54 -0700485/*
486 * Note that phys_device is optional. It is here to allow for
487 * differentiation between which *physical* devices each
488 * section belongs to...
489 */
Heiko Carstensbc32df02010-03-15 00:35:03 -0400490int __weak arch_get_memory_phys_device(unsigned long start_pfn)
491{
492 return 0;
493}
Dave Hansen3947be12005-10-29 18:16:54 -0700494
Robin Holt98383032010-09-29 14:00:55 -0500495struct memory_block *find_memory_block_hinted(struct mem_section *section,
496 struct memory_block *hint)
497{
498 struct kobject *kobj;
499 struct sys_device *sysdev;
500 struct memory_block *mem;
501 char name[sizeof(MEMORY_CLASS_NAME) + 9 + 1];
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600502 int block_id = base_memory_block_id(__section_nr(section));
Robin Holt98383032010-09-29 14:00:55 -0500503
504 kobj = hint ? &hint->sysdev.kobj : NULL;
505
506 /*
507 * This only works because we know that section == sysdev->id
508 * slightly redundant with sysdev_register()
509 */
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600510 sprintf(&name[0], "%s%d", MEMORY_CLASS_NAME, block_id);
Robin Holt98383032010-09-29 14:00:55 -0500511
512 kobj = kset_find_obj_hinted(&memory_sysdev_class.kset, name, kobj);
513 if (!kobj)
514 return NULL;
515
516 sysdev = container_of(kobj, struct sys_device, kobj);
517 mem = container_of(sysdev, struct memory_block, sysdev);
518
519 return mem;
520}
521
Dave Hansen3947be12005-10-29 18:16:54 -0700522/*
523 * For now, we have a linear search to go find the appropriate
524 * memory_block corresponding to a particular phys_index. If
525 * this gets to be a real problem, we can always use a radix
526 * tree or something here.
527 *
528 * This could be made generic for all sysdev classes.
529 */
Gary Hadec04fc582009-01-06 14:39:14 -0800530struct memory_block *find_memory_block(struct mem_section *section)
Dave Hansen3947be12005-10-29 18:16:54 -0700531{
Robin Holt98383032010-09-29 14:00:55 -0500532 return find_memory_block_hinted(section, NULL);
Dave Hansen3947be12005-10-29 18:16:54 -0700533}
534
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600535static int init_memory_block(struct memory_block **memory,
536 struct mem_section *section, unsigned long state)
Nathan Fontenote4619c82010-10-19 12:44:20 -0500537{
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600538 struct memory_block *mem;
Nathan Fontenote4619c82010-10-19 12:44:20 -0500539 unsigned long start_pfn;
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600540 int scn_nr;
Nathan Fontenote4619c82010-10-19 12:44:20 -0500541 int ret = 0;
542
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600543 mem = kzalloc(sizeof(*mem), GFP_KERNEL);
Nathan Fontenote4619c82010-10-19 12:44:20 -0500544 if (!mem)
545 return -ENOMEM;
546
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600547 scn_nr = __section_nr(section);
Nathan Fontenotd3360162011-01-20 10:44:29 -0600548 mem->start_section_nr =
549 base_memory_block_id(scn_nr) * sections_per_block;
550 mem->end_section_nr = mem->start_section_nr + sections_per_block - 1;
Nathan Fontenote4619c82010-10-19 12:44:20 -0500551 mem->state = state;
Nathan Fontenot07681212010-10-19 12:46:19 -0500552 mem->section_count++;
Nathan Fontenote4619c82010-10-19 12:44:20 -0500553 mutex_init(&mem->state_mutex);
Nathan Fontenotd3360162011-01-20 10:44:29 -0600554 start_pfn = section_nr_to_pfn(mem->start_section_nr);
Nathan Fontenote4619c82010-10-19 12:44:20 -0500555 mem->phys_device = arch_get_memory_phys_device(start_pfn);
556
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600557 ret = register_memory(mem);
Nathan Fontenote4619c82010-10-19 12:44:20 -0500558 if (!ret)
559 ret = mem_create_simple_file(mem, phys_index);
560 if (!ret)
Nathan Fontenotd3360162011-01-20 10:44:29 -0600561 ret = mem_create_simple_file(mem, end_phys_index);
562 if (!ret)
Nathan Fontenote4619c82010-10-19 12:44:20 -0500563 ret = mem_create_simple_file(mem, state);
564 if (!ret)
565 ret = mem_create_simple_file(mem, phys_device);
566 if (!ret)
567 ret = mem_create_simple_file(mem, removable);
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600568
569 *memory = mem;
570 return ret;
571}
572
573static int add_memory_section(int nid, struct mem_section *section,
574 unsigned long state, enum mem_add_context context)
575{
576 struct memory_block *mem;
577 int ret = 0;
578
579 mutex_lock(&mem_sysfs_mutex);
580
581 mem = find_memory_block(section);
582 if (mem) {
583 mem->section_count++;
584 kobject_put(&mem->sysdev.kobj);
585 } else
586 ret = init_memory_block(&mem, section, state);
587
Nathan Fontenote4619c82010-10-19 12:44:20 -0500588 if (!ret) {
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600589 if (context == HOTPLUG &&
590 mem->section_count == sections_per_block)
Nathan Fontenote4619c82010-10-19 12:44:20 -0500591 ret = register_mem_sect_under_node(mem, nid);
592 }
593
Nathan Fontenot2938ffb2010-10-19 12:45:24 -0500594 mutex_unlock(&mem_sysfs_mutex);
Nathan Fontenote4619c82010-10-19 12:44:20 -0500595 return ret;
596}
597
Dave Hansen3947be12005-10-29 18:16:54 -0700598int remove_memory_block(unsigned long node_id, struct mem_section *section,
599 int phys_device)
600{
601 struct memory_block *mem;
602
Nathan Fontenot2938ffb2010-10-19 12:45:24 -0500603 mutex_lock(&mem_sysfs_mutex);
Dave Hansen3947be12005-10-29 18:16:54 -0700604 mem = find_memory_block(section);
Nathan Fontenotd3360162011-01-20 10:44:29 -0600605 unregister_mem_sect_under_nodes(mem, __section_nr(section));
Nathan Fontenot07681212010-10-19 12:46:19 -0500606
607 mem->section_count--;
608 if (mem->section_count == 0) {
Nathan Fontenot07681212010-10-19 12:46:19 -0500609 mem_remove_simple_file(mem, phys_index);
Nathan Fontenotd3360162011-01-20 10:44:29 -0600610 mem_remove_simple_file(mem, end_phys_index);
Nathan Fontenot07681212010-10-19 12:46:19 -0500611 mem_remove_simple_file(mem, state);
612 mem_remove_simple_file(mem, phys_device);
613 mem_remove_simple_file(mem, removable);
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600614 unregister_memory(mem);
615 kfree(mem);
616 } else
617 kobject_put(&mem->sysdev.kobj);
Dave Hansen3947be12005-10-29 18:16:54 -0700618
Nathan Fontenot2938ffb2010-10-19 12:45:24 -0500619 mutex_unlock(&mem_sysfs_mutex);
Dave Hansen3947be12005-10-29 18:16:54 -0700620 return 0;
621}
622
623/*
624 * need an interface for the VM to add new memory regions,
625 * but without onlining it.
626 */
Gary Hadec04fc582009-01-06 14:39:14 -0800627int register_new_memory(int nid, struct mem_section *section)
Dave Hansen3947be12005-10-29 18:16:54 -0700628{
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600629 return add_memory_section(nid, section, MEM_OFFLINE, HOTPLUG);
Dave Hansen3947be12005-10-29 18:16:54 -0700630}
631
632int unregister_memory_section(struct mem_section *section)
633{
Andy Whitcroft540557b2007-10-16 01:24:11 -0700634 if (!present_section(section))
Dave Hansen3947be12005-10-29 18:16:54 -0700635 return -EINVAL;
636
637 return remove_memory_block(0, section, 0);
638}
639
640/*
641 * Initialize the sysfs support for memory devices...
642 */
643int __init memory_dev_init(void)
644{
645 unsigned int i;
646 int ret;
Andrew Morton28ec24e2006-12-06 20:37:29 -0800647 int err;
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600648 unsigned long block_sz;
Dave Hansen3947be12005-10-29 18:16:54 -0700649
Kay Sievers312c0042005-11-16 09:00:00 +0100650 memory_sysdev_class.kset.uevent_ops = &memory_uevent_ops;
Dave Hansen3947be12005-10-29 18:16:54 -0700651 ret = sysdev_class_register(&memory_sysdev_class);
Andrew Morton28ec24e2006-12-06 20:37:29 -0800652 if (ret)
653 goto out;
Dave Hansen3947be12005-10-29 18:16:54 -0700654
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600655 block_sz = get_memory_block_size();
656 sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE;
657
Dave Hansen3947be12005-10-29 18:16:54 -0700658 /*
659 * Create entries for memory sections that were found
660 * during boot and have been initialized
661 */
662 for (i = 0; i < NR_MEM_SECTIONS; i++) {
Andy Whitcroft540557b2007-10-16 01:24:11 -0700663 if (!present_section_nr(i))
Dave Hansen3947be12005-10-29 18:16:54 -0700664 continue;
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600665 err = add_memory_section(0, __nr_to_section(i), MEM_ONLINE,
666 BOOT);
Andrew Morton28ec24e2006-12-06 20:37:29 -0800667 if (!ret)
668 ret = err;
Dave Hansen3947be12005-10-29 18:16:54 -0700669 }
670
Andrew Morton28ec24e2006-12-06 20:37:29 -0800671 err = memory_probe_init();
672 if (!ret)
673 ret = err;
Andi Kleenfacb6012009-12-16 12:20:00 +0100674 err = memory_fail_init();
675 if (!ret)
676 ret = err;
Andrew Morton28ec24e2006-12-06 20:37:29 -0800677 err = block_size_init();
678 if (!ret)
679 ret = err;
680out:
681 if (ret)
Harvey Harrison2b3a3022008-03-04 16:41:05 -0800682 printk(KERN_ERR "%s() failed: %d\n", __func__, ret);
Dave Hansen3947be12005-10-29 18:16:54 -0700683 return ret;
684}