blob: 9f9b2359f718f486ba5336807fcd003ab6bff1f5 [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
Wanlong Gao85723942011-04-23 22:18:26 +080051static int memory_uevent(struct kset *kset, struct kobject *obj,
52 struct kobj_uevent_env *env)
Dave Hansen3947be12005-10-29 18:16:54 -070053{
54 int retval = 0;
55
56 return retval;
57}
58
Emese Revfy9cd43612009-12-31 14:52:51 +010059static const struct kset_uevent_ops memory_uevent_ops = {
Kay Sievers312c0042005-11-16 09:00:00 +010060 .name = memory_uevent_name,
61 .uevent = memory_uevent,
Dave Hansen3947be12005-10-29 18:16:54 -070062};
63
Alan Sterne041c682006-03-27 01:16:30 -080064static BLOCKING_NOTIFIER_HEAD(memory_chain);
Dave Hansen3947be12005-10-29 18:16:54 -070065
Andy Whitcroft98a38eb2006-01-06 00:10:35 -080066int register_memory_notifier(struct notifier_block *nb)
Dave Hansen3947be12005-10-29 18:16:54 -070067{
Alan Sterne041c682006-03-27 01:16:30 -080068 return blocking_notifier_chain_register(&memory_chain, nb);
Dave Hansen3947be12005-10-29 18:16:54 -070069}
Hannes Hering3c82c302008-05-07 14:43:01 +020070EXPORT_SYMBOL(register_memory_notifier);
Dave Hansen3947be12005-10-29 18:16:54 -070071
Andy Whitcroft98a38eb2006-01-06 00:10:35 -080072void unregister_memory_notifier(struct notifier_block *nb)
Dave Hansen3947be12005-10-29 18:16:54 -070073{
Alan Sterne041c682006-03-27 01:16:30 -080074 blocking_notifier_chain_unregister(&memory_chain, nb);
Dave Hansen3947be12005-10-29 18:16:54 -070075}
Hannes Hering3c82c302008-05-07 14:43:01 +020076EXPORT_SYMBOL(unregister_memory_notifier);
Dave Hansen3947be12005-10-29 18:16:54 -070077
Robert Jennings925cc712009-12-17 14:44:38 +000078static ATOMIC_NOTIFIER_HEAD(memory_isolate_chain);
79
80int register_memory_isolate_notifier(struct notifier_block *nb)
81{
82 return atomic_notifier_chain_register(&memory_isolate_chain, nb);
83}
84EXPORT_SYMBOL(register_memory_isolate_notifier);
85
86void unregister_memory_isolate_notifier(struct notifier_block *nb)
87{
88 atomic_notifier_chain_unregister(&memory_isolate_chain, nb);
89}
90EXPORT_SYMBOL(unregister_memory_isolate_notifier);
91
Dave Hansen3947be12005-10-29 18:16:54 -070092/*
93 * register_memory - Setup a sysfs device for a memory block
94 */
Badari Pulavarty00a41db2008-02-11 09:23:18 -080095static
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -060096int register_memory(struct memory_block *memory)
Dave Hansen3947be12005-10-29 18:16:54 -070097{
98 int error;
99
100 memory->sysdev.cls = &memory_sysdev_class;
Nathan Fontenotd3360162011-01-20 10:44:29 -0600101 memory->sysdev.id = memory->start_section_nr / sections_per_block;
Dave Hansen3947be12005-10-29 18:16:54 -0700102
103 error = sysdev_register(&memory->sysdev);
Dave Hansen3947be12005-10-29 18:16:54 -0700104 return error;
105}
106
107static void
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600108unregister_memory(struct memory_block *memory)
Dave Hansen3947be12005-10-29 18:16:54 -0700109{
110 BUG_ON(memory->sysdev.cls != &memory_sysdev_class);
Dave Hansen3947be12005-10-29 18:16:54 -0700111
Badari Pulavarty00a41db2008-02-11 09:23:18 -0800112 /* drop the ref. we got in remove_memory_block() */
113 kobject_put(&memory->sysdev.kobj);
Dave Hansen3947be12005-10-29 18:16:54 -0700114 sysdev_unregister(&memory->sysdev);
Dave Hansen3947be12005-10-29 18:16:54 -0700115}
116
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600117unsigned long __weak memory_block_size_bytes(void)
118{
119 return MIN_MEMORY_BLOCK_SIZE;
120}
121
122static unsigned long get_memory_block_size(void)
123{
124 unsigned long block_sz;
125
126 block_sz = memory_block_size_bytes();
127
128 /* Validate blk_sz is a power of 2 and not less than section size */
129 if ((block_sz & (block_sz - 1)) || (block_sz < MIN_MEMORY_BLOCK_SIZE)) {
130 WARN_ON(1);
131 block_sz = MIN_MEMORY_BLOCK_SIZE;
132 }
133
134 return block_sz;
135}
136
Dave Hansen3947be12005-10-29 18:16:54 -0700137/*
138 * use this as the physical section index that this memsection
139 * uses.
140 */
141
Nathan Fontenotd3360162011-01-20 10:44:29 -0600142static ssize_t show_mem_start_phys_index(struct sys_device *dev,
Andi Kleen4a0b2b42008-07-01 18:48:41 +0200143 struct sysdev_attribute *attr, char *buf)
Dave Hansen3947be12005-10-29 18:16:54 -0700144{
145 struct memory_block *mem =
146 container_of(dev, struct memory_block, sysdev);
Nathan Fontenotd3360162011-01-20 10:44:29 -0600147 unsigned long phys_index;
148
149 phys_index = mem->start_section_nr / sections_per_block;
150 return sprintf(buf, "%08lx\n", phys_index);
151}
152
153static ssize_t show_mem_end_phys_index(struct sys_device *dev,
154 struct sysdev_attribute *attr, char *buf)
155{
156 struct memory_block *mem =
157 container_of(dev, struct memory_block, sysdev);
158 unsigned long phys_index;
159
160 phys_index = mem->end_section_nr / sections_per_block;
161 return sprintf(buf, "%08lx\n", phys_index);
Dave Hansen3947be12005-10-29 18:16:54 -0700162}
163
164/*
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700165 * Show whether the section of memory is likely to be hot-removable
166 */
Stephen Rothwell1f07be12008-07-28 11:05:04 +1000167static ssize_t show_mem_removable(struct sys_device *dev,
168 struct sysdev_attribute *attr, char *buf)
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700169{
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600170 unsigned long i, pfn;
171 int ret = 1;
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700172 struct memory_block *mem =
173 container_of(dev, struct memory_block, sysdev);
174
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600175 for (i = 0; i < sections_per_block; i++) {
Nathan Fontenotd3360162011-01-20 10:44:29 -0600176 pfn = section_nr_to_pfn(mem->start_section_nr + i);
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600177 ret &= is_mem_section_removable(pfn, PAGES_PER_SECTION);
178 }
179
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700180 return sprintf(buf, "%d\n", ret);
181}
182
183/*
Dave Hansen3947be12005-10-29 18:16:54 -0700184 * online, offline, going offline, etc.
185 */
Andi Kleen4a0b2b42008-07-01 18:48:41 +0200186static ssize_t show_mem_state(struct sys_device *dev,
187 struct sysdev_attribute *attr, char *buf)
Dave Hansen3947be12005-10-29 18:16:54 -0700188{
189 struct memory_block *mem =
190 container_of(dev, struct memory_block, sysdev);
191 ssize_t len = 0;
192
193 /*
194 * We can probably put these states in a nice little array
195 * so that they're not open-coded
196 */
197 switch (mem->state) {
198 case MEM_ONLINE:
199 len = sprintf(buf, "online\n");
200 break;
201 case MEM_OFFLINE:
202 len = sprintf(buf, "offline\n");
203 break;
204 case MEM_GOING_OFFLINE:
205 len = sprintf(buf, "going-offline\n");
206 break;
207 default:
208 len = sprintf(buf, "ERROR-UNKNOWN-%ld\n",
209 mem->state);
210 WARN_ON(1);
211 break;
212 }
213
214 return len;
215}
216
Yasunori Goto7b78d332007-10-21 16:41:36 -0700217int memory_notify(unsigned long val, void *v)
Dave Hansen3947be12005-10-29 18:16:54 -0700218{
Alan Sterne041c682006-03-27 01:16:30 -0800219 return blocking_notifier_call_chain(&memory_chain, val, v);
Dave Hansen3947be12005-10-29 18:16:54 -0700220}
221
Robert Jennings925cc712009-12-17 14:44:38 +0000222int memory_isolate_notify(unsigned long val, void *v)
223{
224 return atomic_notifier_call_chain(&memory_isolate_chain, val, v);
225}
226
Dave Hansen3947be12005-10-29 18:16:54 -0700227/*
228 * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is
229 * OK to have direct references to sparsemem variables in here.
230 */
231static int
Anton Blanchard5409d2c2011-05-11 17:25:14 +1000232memory_block_action(unsigned long phys_index, unsigned long action)
Dave Hansen3947be12005-10-29 18:16:54 -0700233{
234 int i;
Dave Hansen3947be12005-10-29 18:16:54 -0700235 unsigned long start_pfn, start_paddr;
Anton Blanchard5409d2c2011-05-11 17:25:14 +1000236 unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
Dave Hansen3947be12005-10-29 18:16:54 -0700237 struct page *first_page;
238 int ret;
Dave Hansen3947be12005-10-29 18:16:54 -0700239
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600240 first_page = pfn_to_page(phys_index << PFN_SECTION_SHIFT);
Dave Hansen3947be12005-10-29 18:16:54 -0700241
242 /*
243 * The probe routines leave the pages reserved, just
244 * as the bootmem code does. Make sure they're still
245 * that way.
246 */
247 if (action == MEM_ONLINE) {
Anton Blanchard5409d2c2011-05-11 17:25:14 +1000248 for (i = 0; i < nr_pages; i++) {
Dave Hansen3947be12005-10-29 18:16:54 -0700249 if (PageReserved(first_page+i))
250 continue;
251
252 printk(KERN_WARNING "section number %ld page number %d "
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600253 "not reserved, was it already online?\n",
254 phys_index, i);
Dave Hansen3947be12005-10-29 18:16:54 -0700255 return -EBUSY;
256 }
257 }
258
259 switch (action) {
260 case MEM_ONLINE:
261 start_pfn = page_to_pfn(first_page);
Anton Blanchard5409d2c2011-05-11 17:25:14 +1000262 ret = online_pages(start_pfn, nr_pages);
Dave Hansen3947be12005-10-29 18:16:54 -0700263 break;
264 case MEM_OFFLINE:
Dave Hansen3947be12005-10-29 18:16:54 -0700265 start_paddr = page_to_pfn(first_page) << PAGE_SHIFT;
266 ret = remove_memory(start_paddr,
Anton Blanchard5409d2c2011-05-11 17:25:14 +1000267 nr_pages << PAGE_SHIFT);
Dave Hansen3947be12005-10-29 18:16:54 -0700268 break;
269 default:
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600270 WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: "
271 "%ld\n", __func__, phys_index, action, action);
Dave Hansen3947be12005-10-29 18:16:54 -0700272 ret = -EINVAL;
273 }
Dave Hansen3947be12005-10-29 18:16:54 -0700274
275 return ret;
276}
277
278static int memory_block_change_state(struct memory_block *mem,
279 unsigned long to_state, unsigned long from_state_req)
280{
Andrew Mortona236c712011-05-12 12:45:31 -0700281 int ret = 0;
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600282
Daniel Walkerda19cbc2008-02-04 23:35:47 -0800283 mutex_lock(&mem->state_mutex);
Dave Hansen3947be12005-10-29 18:16:54 -0700284
285 if (mem->state != from_state_req) {
286 ret = -EINVAL;
287 goto out;
288 }
289
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600290 if (to_state == MEM_OFFLINE)
291 mem->state = MEM_GOING_OFFLINE;
292
Anton Blanchard5409d2c2011-05-11 17:25:14 +1000293 ret = memory_block_action(mem->start_section_nr, to_state);
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600294
Anton Blanchard5409d2c2011-05-11 17:25:14 +1000295 if (ret)
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600296 mem->state = from_state_req;
Anton Blanchard5409d2c2011-05-11 17:25:14 +1000297 else
Dave Hansen3947be12005-10-29 18:16:54 -0700298 mem->state = to_state;
299
300out:
Daniel Walkerda19cbc2008-02-04 23:35:47 -0800301 mutex_unlock(&mem->state_mutex);
Dave Hansen3947be12005-10-29 18:16:54 -0700302 return ret;
303}
304
305static ssize_t
Andi Kleen4a0b2b42008-07-01 18:48:41 +0200306store_mem_state(struct sys_device *dev,
307 struct sysdev_attribute *attr, const char *buf, size_t count)
Dave Hansen3947be12005-10-29 18:16:54 -0700308{
309 struct memory_block *mem;
Dave Hansen3947be12005-10-29 18:16:54 -0700310 int ret = -EINVAL;
311
312 mem = container_of(dev, struct memory_block, sysdev);
Dave Hansen3947be12005-10-29 18:16:54 -0700313
314 if (!strncmp(buf, "online", min((int)count, 6)))
315 ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
316 else if(!strncmp(buf, "offline", min((int)count, 7)))
317 ret = memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE);
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600318
Dave Hansen3947be12005-10-29 18:16:54 -0700319 if (ret)
320 return ret;
321 return count;
322}
323
324/*
325 * phys_device is a bad name for this. What I really want
326 * is a way to differentiate between memory ranges that
327 * are part of physical devices that constitute
328 * a complete removable unit or fru.
329 * i.e. do these ranges belong to the same physical device,
330 * s.t. if I offline all of these sections I can then
331 * remove the physical device?
332 */
Andi Kleen4a0b2b42008-07-01 18:48:41 +0200333static ssize_t show_phys_device(struct sys_device *dev,
334 struct sysdev_attribute *attr, char *buf)
Dave Hansen3947be12005-10-29 18:16:54 -0700335{
336 struct memory_block *mem =
337 container_of(dev, struct memory_block, sysdev);
338 return sprintf(buf, "%d\n", mem->phys_device);
339}
340
Nathan Fontenotd3360162011-01-20 10:44:29 -0600341static SYSDEV_ATTR(phys_index, 0444, show_mem_start_phys_index, NULL);
342static SYSDEV_ATTR(end_phys_index, 0444, show_mem_end_phys_index, NULL);
Dave Hansen3947be12005-10-29 18:16:54 -0700343static SYSDEV_ATTR(state, 0644, show_mem_state, store_mem_state);
344static SYSDEV_ATTR(phys_device, 0444, show_phys_device, NULL);
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700345static SYSDEV_ATTR(removable, 0444, show_mem_removable, NULL);
Dave Hansen3947be12005-10-29 18:16:54 -0700346
347#define mem_create_simple_file(mem, attr_name) \
348 sysdev_create_file(&mem->sysdev, &attr_##attr_name)
349#define mem_remove_simple_file(mem, attr_name) \
350 sysdev_remove_file(&mem->sysdev, &attr_##attr_name)
351
352/*
353 * Block size attribute stuff
354 */
355static ssize_t
Andi Kleen8564a6c2010-01-05 12:48:06 +0100356print_block_size(struct sysdev_class *class, struct sysdev_class_attribute *attr,
357 char *buf)
Dave Hansen3947be12005-10-29 18:16:54 -0700358{
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600359 return sprintf(buf, "%lx\n", get_memory_block_size());
Dave Hansen3947be12005-10-29 18:16:54 -0700360}
361
Andi Kleen8564a6c2010-01-05 12:48:06 +0100362static SYSDEV_CLASS_ATTR(block_size_bytes, 0444, print_block_size, NULL);
Dave Hansen3947be12005-10-29 18:16:54 -0700363
364static int block_size_init(void)
365{
Andrew Morton28ec24e2006-12-06 20:37:29 -0800366 return sysfs_create_file(&memory_sysdev_class.kset.kobj,
Andi Kleen8564a6c2010-01-05 12:48:06 +0100367 &attr_block_size_bytes.attr);
Dave Hansen3947be12005-10-29 18:16:54 -0700368}
369
370/*
371 * Some architectures will have custom drivers to do this, and
372 * will not need to do it from userspace. The fake hot-add code
373 * as well as ppc64 will do all of their discovery in userspace
374 * and will require this interface.
375 */
376#ifdef CONFIG_ARCH_MEMORY_PROBE
377static ssize_t
Andi Kleen28812fe2010-01-05 12:48:07 +0100378memory_probe_store(struct class *class, struct class_attribute *attr,
379 const char *buf, size_t count)
Dave Hansen3947be12005-10-29 18:16:54 -0700380{
381 u64 phys_addr;
Yasunori Gotobc02af92006-06-27 02:53:30 -0700382 int nid;
Nathan Fontenot6add7cd2011-01-31 10:55:23 -0600383 int i, ret;
Dave Hansen3947be12005-10-29 18:16:54 -0700384
385 phys_addr = simple_strtoull(buf, NULL, 0);
386
Nathan Fontenot6add7cd2011-01-31 10:55:23 -0600387 for (i = 0; i < sections_per_block; i++) {
388 nid = memory_add_physaddr_to_nid(phys_addr);
389 ret = add_memory(nid, phys_addr,
390 PAGES_PER_SECTION << PAGE_SHIFT);
391 if (ret)
Nikanth Karthikesan9f0af692011-03-24 11:46:18 +0530392 goto out;
Nathan Fontenot6add7cd2011-01-31 10:55:23 -0600393
394 phys_addr += MIN_MEMORY_BLOCK_SIZE;
395 }
Dave Hansen3947be12005-10-29 18:16:54 -0700396
Nikanth Karthikesan9f0af692011-03-24 11:46:18 +0530397 ret = count;
398out:
399 return ret;
Dave Hansen3947be12005-10-29 18:16:54 -0700400}
Greg Kroah-Hartmanbd796672010-01-19 13:08:16 -0800401static CLASS_ATTR(probe, S_IWUSR, NULL, memory_probe_store);
Dave Hansen3947be12005-10-29 18:16:54 -0700402
403static int memory_probe_init(void)
404{
Andrew Morton28ec24e2006-12-06 20:37:29 -0800405 return sysfs_create_file(&memory_sysdev_class.kset.kobj,
Greg Kroah-Hartmanbd796672010-01-19 13:08:16 -0800406 &class_attr_probe.attr);
Dave Hansen3947be12005-10-29 18:16:54 -0700407}
408#else
Andrew Morton28ec24e2006-12-06 20:37:29 -0800409static inline int memory_probe_init(void)
410{
411 return 0;
412}
Dave Hansen3947be12005-10-29 18:16:54 -0700413#endif
414
Andi Kleenfacb6012009-12-16 12:20:00 +0100415#ifdef CONFIG_MEMORY_FAILURE
416/*
417 * Support for offlining pages of memory
418 */
419
420/* Soft offline a page */
421static ssize_t
Andi Kleen28812fe2010-01-05 12:48:07 +0100422store_soft_offline_page(struct class *class,
423 struct class_attribute *attr,
424 const char *buf, size_t count)
Andi Kleenfacb6012009-12-16 12:20:00 +0100425{
426 int ret;
427 u64 pfn;
428 if (!capable(CAP_SYS_ADMIN))
429 return -EPERM;
430 if (strict_strtoull(buf, 0, &pfn) < 0)
431 return -EINVAL;
432 pfn >>= PAGE_SHIFT;
433 if (!pfn_valid(pfn))
434 return -ENXIO;
435 ret = soft_offline_page(pfn_to_page(pfn), 0);
436 return ret == 0 ? count : ret;
437}
438
439/* Forcibly offline a page, including killing processes. */
440static ssize_t
Andi Kleen28812fe2010-01-05 12:48:07 +0100441store_hard_offline_page(struct class *class,
442 struct class_attribute *attr,
443 const char *buf, size_t count)
Andi Kleenfacb6012009-12-16 12:20:00 +0100444{
445 int ret;
446 u64 pfn;
447 if (!capable(CAP_SYS_ADMIN))
448 return -EPERM;
449 if (strict_strtoull(buf, 0, &pfn) < 0)
450 return -EINVAL;
451 pfn >>= PAGE_SHIFT;
452 ret = __memory_failure(pfn, 0, 0);
453 return ret ? ret : count;
454}
455
Greg Kroah-Hartmanbd796672010-01-19 13:08:16 -0800456static CLASS_ATTR(soft_offline_page, 0644, NULL, store_soft_offline_page);
457static CLASS_ATTR(hard_offline_page, 0644, NULL, store_hard_offline_page);
Andi Kleenfacb6012009-12-16 12:20:00 +0100458
459static __init int memory_fail_init(void)
460{
461 int err;
462
463 err = sysfs_create_file(&memory_sysdev_class.kset.kobj,
Greg Kroah-Hartmanbd796672010-01-19 13:08:16 -0800464 &class_attr_soft_offline_page.attr);
Andi Kleenfacb6012009-12-16 12:20:00 +0100465 if (!err)
466 err = sysfs_create_file(&memory_sysdev_class.kset.kobj,
Greg Kroah-Hartmanbd796672010-01-19 13:08:16 -0800467 &class_attr_hard_offline_page.attr);
Andi Kleenfacb6012009-12-16 12:20:00 +0100468 return err;
469}
470#else
471static inline int memory_fail_init(void)
472{
473 return 0;
474}
475#endif
476
Dave Hansen3947be12005-10-29 18:16:54 -0700477/*
478 * Note that phys_device is optional. It is here to allow for
479 * differentiation between which *physical* devices each
480 * section belongs to...
481 */
Heiko Carstensbc32df02010-03-15 00:35:03 -0400482int __weak arch_get_memory_phys_device(unsigned long start_pfn)
483{
484 return 0;
485}
Dave Hansen3947be12005-10-29 18:16:54 -0700486
Robin Holt98383032010-09-29 14:00:55 -0500487struct memory_block *find_memory_block_hinted(struct mem_section *section,
488 struct memory_block *hint)
489{
490 struct kobject *kobj;
491 struct sys_device *sysdev;
492 struct memory_block *mem;
493 char name[sizeof(MEMORY_CLASS_NAME) + 9 + 1];
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600494 int block_id = base_memory_block_id(__section_nr(section));
Robin Holt98383032010-09-29 14:00:55 -0500495
496 kobj = hint ? &hint->sysdev.kobj : NULL;
497
498 /*
499 * This only works because we know that section == sysdev->id
500 * slightly redundant with sysdev_register()
501 */
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600502 sprintf(&name[0], "%s%d", MEMORY_CLASS_NAME, block_id);
Robin Holt98383032010-09-29 14:00:55 -0500503
504 kobj = kset_find_obj_hinted(&memory_sysdev_class.kset, name, kobj);
505 if (!kobj)
506 return NULL;
507
508 sysdev = container_of(kobj, struct sys_device, kobj);
509 mem = container_of(sysdev, struct memory_block, sysdev);
510
511 return mem;
512}
513
Dave Hansen3947be12005-10-29 18:16:54 -0700514/*
515 * For now, we have a linear search to go find the appropriate
516 * memory_block corresponding to a particular phys_index. If
517 * this gets to be a real problem, we can always use a radix
518 * tree or something here.
519 *
520 * This could be made generic for all sysdev classes.
521 */
Gary Hadec04fc582009-01-06 14:39:14 -0800522struct memory_block *find_memory_block(struct mem_section *section)
Dave Hansen3947be12005-10-29 18:16:54 -0700523{
Robin Holt98383032010-09-29 14:00:55 -0500524 return find_memory_block_hinted(section, NULL);
Dave Hansen3947be12005-10-29 18:16:54 -0700525}
526
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600527static int init_memory_block(struct memory_block **memory,
528 struct mem_section *section, unsigned long state)
Nathan Fontenote4619c82010-10-19 12:44:20 -0500529{
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600530 struct memory_block *mem;
Nathan Fontenote4619c82010-10-19 12:44:20 -0500531 unsigned long start_pfn;
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600532 int scn_nr;
Nathan Fontenote4619c82010-10-19 12:44:20 -0500533 int ret = 0;
534
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600535 mem = kzalloc(sizeof(*mem), GFP_KERNEL);
Nathan Fontenote4619c82010-10-19 12:44:20 -0500536 if (!mem)
537 return -ENOMEM;
538
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600539 scn_nr = __section_nr(section);
Nathan Fontenotd3360162011-01-20 10:44:29 -0600540 mem->start_section_nr =
541 base_memory_block_id(scn_nr) * sections_per_block;
542 mem->end_section_nr = mem->start_section_nr + sections_per_block - 1;
Nathan Fontenote4619c82010-10-19 12:44:20 -0500543 mem->state = state;
Nathan Fontenot07681212010-10-19 12:46:19 -0500544 mem->section_count++;
Nathan Fontenote4619c82010-10-19 12:44:20 -0500545 mutex_init(&mem->state_mutex);
Nathan Fontenotd3360162011-01-20 10:44:29 -0600546 start_pfn = section_nr_to_pfn(mem->start_section_nr);
Nathan Fontenote4619c82010-10-19 12:44:20 -0500547 mem->phys_device = arch_get_memory_phys_device(start_pfn);
548
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600549 ret = register_memory(mem);
Nathan Fontenote4619c82010-10-19 12:44:20 -0500550 if (!ret)
551 ret = mem_create_simple_file(mem, phys_index);
552 if (!ret)
Nathan Fontenotd3360162011-01-20 10:44:29 -0600553 ret = mem_create_simple_file(mem, end_phys_index);
554 if (!ret)
Nathan Fontenote4619c82010-10-19 12:44:20 -0500555 ret = mem_create_simple_file(mem, state);
556 if (!ret)
557 ret = mem_create_simple_file(mem, phys_device);
558 if (!ret)
559 ret = mem_create_simple_file(mem, removable);
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600560
561 *memory = mem;
562 return ret;
563}
564
565static int add_memory_section(int nid, struct mem_section *section,
566 unsigned long state, enum mem_add_context context)
567{
568 struct memory_block *mem;
569 int ret = 0;
570
571 mutex_lock(&mem_sysfs_mutex);
572
573 mem = find_memory_block(section);
574 if (mem) {
575 mem->section_count++;
576 kobject_put(&mem->sysdev.kobj);
577 } else
578 ret = init_memory_block(&mem, section, state);
579
Nathan Fontenote4619c82010-10-19 12:44:20 -0500580 if (!ret) {
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600581 if (context == HOTPLUG &&
582 mem->section_count == sections_per_block)
Nathan Fontenote4619c82010-10-19 12:44:20 -0500583 ret = register_mem_sect_under_node(mem, nid);
584 }
585
Nathan Fontenot2938ffb2010-10-19 12:45:24 -0500586 mutex_unlock(&mem_sysfs_mutex);
Nathan Fontenote4619c82010-10-19 12:44:20 -0500587 return ret;
588}
589
Dave Hansen3947be12005-10-29 18:16:54 -0700590int remove_memory_block(unsigned long node_id, struct mem_section *section,
591 int phys_device)
592{
593 struct memory_block *mem;
594
Nathan Fontenot2938ffb2010-10-19 12:45:24 -0500595 mutex_lock(&mem_sysfs_mutex);
Dave Hansen3947be12005-10-29 18:16:54 -0700596 mem = find_memory_block(section);
Nathan Fontenotd3360162011-01-20 10:44:29 -0600597 unregister_mem_sect_under_nodes(mem, __section_nr(section));
Nathan Fontenot07681212010-10-19 12:46:19 -0500598
599 mem->section_count--;
600 if (mem->section_count == 0) {
Nathan Fontenot07681212010-10-19 12:46:19 -0500601 mem_remove_simple_file(mem, phys_index);
Nathan Fontenotd3360162011-01-20 10:44:29 -0600602 mem_remove_simple_file(mem, end_phys_index);
Nathan Fontenot07681212010-10-19 12:46:19 -0500603 mem_remove_simple_file(mem, state);
604 mem_remove_simple_file(mem, phys_device);
605 mem_remove_simple_file(mem, removable);
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600606 unregister_memory(mem);
607 kfree(mem);
608 } else
609 kobject_put(&mem->sysdev.kobj);
Dave Hansen3947be12005-10-29 18:16:54 -0700610
Nathan Fontenot2938ffb2010-10-19 12:45:24 -0500611 mutex_unlock(&mem_sysfs_mutex);
Dave Hansen3947be12005-10-29 18:16:54 -0700612 return 0;
613}
614
615/*
616 * need an interface for the VM to add new memory regions,
617 * but without onlining it.
618 */
Gary Hadec04fc582009-01-06 14:39:14 -0800619int register_new_memory(int nid, struct mem_section *section)
Dave Hansen3947be12005-10-29 18:16:54 -0700620{
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600621 return add_memory_section(nid, section, MEM_OFFLINE, HOTPLUG);
Dave Hansen3947be12005-10-29 18:16:54 -0700622}
623
624int unregister_memory_section(struct mem_section *section)
625{
Andy Whitcroft540557b2007-10-16 01:24:11 -0700626 if (!present_section(section))
Dave Hansen3947be12005-10-29 18:16:54 -0700627 return -EINVAL;
628
629 return remove_memory_block(0, section, 0);
630}
631
632/*
633 * Initialize the sysfs support for memory devices...
634 */
635int __init memory_dev_init(void)
636{
637 unsigned int i;
638 int ret;
Andrew Morton28ec24e2006-12-06 20:37:29 -0800639 int err;
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600640 unsigned long block_sz;
Dave Hansen3947be12005-10-29 18:16:54 -0700641
Kay Sievers312c0042005-11-16 09:00:00 +0100642 memory_sysdev_class.kset.uevent_ops = &memory_uevent_ops;
Dave Hansen3947be12005-10-29 18:16:54 -0700643 ret = sysdev_class_register(&memory_sysdev_class);
Andrew Morton28ec24e2006-12-06 20:37:29 -0800644 if (ret)
645 goto out;
Dave Hansen3947be12005-10-29 18:16:54 -0700646
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600647 block_sz = get_memory_block_size();
648 sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE;
649
Dave Hansen3947be12005-10-29 18:16:54 -0700650 /*
651 * Create entries for memory sections that were found
652 * during boot and have been initialized
653 */
654 for (i = 0; i < NR_MEM_SECTIONS; i++) {
Andy Whitcroft540557b2007-10-16 01:24:11 -0700655 if (!present_section_nr(i))
Dave Hansen3947be12005-10-29 18:16:54 -0700656 continue;
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600657 err = add_memory_section(0, __nr_to_section(i), MEM_ONLINE,
658 BOOT);
Andrew Morton28ec24e2006-12-06 20:37:29 -0800659 if (!ret)
660 ret = err;
Dave Hansen3947be12005-10-29 18:16:54 -0700661 }
662
Andrew Morton28ec24e2006-12-06 20:37:29 -0800663 err = memory_probe_init();
664 if (!ret)
665 ret = err;
Andi Kleenfacb6012009-12-16 12:20:00 +0100666 err = memory_fail_init();
667 if (!ret)
668 ret = err;
Andrew Morton28ec24e2006-12-06 20:37:29 -0800669 err = block_size_init();
670 if (!ret)
671 ret = err;
672out:
673 if (ret)
Harvey Harrison2b3a3022008-03-04 16:41:05 -0800674 printk(KERN_ERR "%s() failed: %d\n", __func__, ret);
Dave Hansen3947be12005-10-29 18:16:54 -0700675 return ret;
676}