blob: 9e60dbe9fd94884cb5850f10af74567a9bc6300f [file] [log] [blame]
Dave Hansen3947be12005-10-29 18:16:54 -07001/*
Kay Sievers10fbcf42011-12-21 14:48:43 -08002 * Memory subsystem support
Dave Hansen3947be12005-10-29 18:16:54 -07003 *
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
Dave Hansen3947be12005-10-29 18:16:54 -070013#include <linux/module.h>
14#include <linux/init.h>
Dave Hansen3947be12005-10-29 18:16:54 -070015#include <linux/topology.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080016#include <linux/capability.h>
Dave Hansen3947be12005-10-29 18:16:54 -070017#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>
Daniel Walkerda19cbc2008-02-04 23:35:47 -080022#include <linux/mutex.h>
Shaohua Li9f1b16a2008-10-18 20:27:12 -070023#include <linux/stat.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090024#include <linux/slab.h>
Shaohua Li9f1b16a2008-10-18 20:27:12 -070025
Arun Sharma60063492011-07-26 16:09:06 -070026#include <linux/atomic.h>
Dave Hansen3947be12005-10-29 18:16:54 -070027#include <asm/uaccess.h>
28
Nathan Fontenot2938ffb2010-10-19 12:45:24 -050029static DEFINE_MUTEX(mem_sysfs_mutex);
30
Dave Hansen3947be12005-10-29 18:16:54 -070031#define MEMORY_CLASS_NAME "memory"
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -060032
33static int sections_per_block;
34
35static inline int base_memory_block_id(int section_nr)
36{
37 return section_nr / sections_per_block;
38}
Dave Hansen3947be12005-10-29 18:16:54 -070039
Kay Sievers10fbcf42011-12-21 14:48:43 -080040static struct bus_type memory_subsys = {
Kay Sieversaf5ca3f2007-12-20 02:09:39 +010041 .name = MEMORY_CLASS_NAME,
Kay Sievers10fbcf42011-12-21 14:48:43 -080042 .dev_name = MEMORY_CLASS_NAME,
Dave Hansen3947be12005-10-29 18:16:54 -070043};
44
Alan Sterne041c682006-03-27 01:16:30 -080045static BLOCKING_NOTIFIER_HEAD(memory_chain);
Dave Hansen3947be12005-10-29 18:16:54 -070046
Andy Whitcroft98a38eb2006-01-06 00:10:35 -080047int register_memory_notifier(struct notifier_block *nb)
Dave Hansen3947be12005-10-29 18:16:54 -070048{
Alan Sterne041c682006-03-27 01:16:30 -080049 return blocking_notifier_chain_register(&memory_chain, nb);
Dave Hansen3947be12005-10-29 18:16:54 -070050}
Hannes Hering3c82c302008-05-07 14:43:01 +020051EXPORT_SYMBOL(register_memory_notifier);
Dave Hansen3947be12005-10-29 18:16:54 -070052
Andy Whitcroft98a38eb2006-01-06 00:10:35 -080053void unregister_memory_notifier(struct notifier_block *nb)
Dave Hansen3947be12005-10-29 18:16:54 -070054{
Alan Sterne041c682006-03-27 01:16:30 -080055 blocking_notifier_chain_unregister(&memory_chain, nb);
Dave Hansen3947be12005-10-29 18:16:54 -070056}
Hannes Hering3c82c302008-05-07 14:43:01 +020057EXPORT_SYMBOL(unregister_memory_notifier);
Dave Hansen3947be12005-10-29 18:16:54 -070058
Robert Jennings925cc712009-12-17 14:44:38 +000059static ATOMIC_NOTIFIER_HEAD(memory_isolate_chain);
60
61int register_memory_isolate_notifier(struct notifier_block *nb)
62{
63 return atomic_notifier_chain_register(&memory_isolate_chain, nb);
64}
65EXPORT_SYMBOL(register_memory_isolate_notifier);
66
67void unregister_memory_isolate_notifier(struct notifier_block *nb)
68{
69 atomic_notifier_chain_unregister(&memory_isolate_chain, nb);
70}
71EXPORT_SYMBOL(unregister_memory_isolate_notifier);
72
Dave Hansen3947be12005-10-29 18:16:54 -070073/*
74 * register_memory - Setup a sysfs device for a memory block
75 */
Badari Pulavarty00a41db2008-02-11 09:23:18 -080076static
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -060077int register_memory(struct memory_block *memory)
Dave Hansen3947be12005-10-29 18:16:54 -070078{
79 int error;
80
Kay Sievers10fbcf42011-12-21 14:48:43 -080081 memory->dev.bus = &memory_subsys;
82 memory->dev.id = memory->start_section_nr / sections_per_block;
Dave Hansen3947be12005-10-29 18:16:54 -070083
Kay Sievers10fbcf42011-12-21 14:48:43 -080084 error = device_register(&memory->dev);
Dave Hansen3947be12005-10-29 18:16:54 -070085 return error;
86}
87
88static void
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -060089unregister_memory(struct memory_block *memory)
Dave Hansen3947be12005-10-29 18:16:54 -070090{
Kay Sievers10fbcf42011-12-21 14:48:43 -080091 BUG_ON(memory->dev.bus != &memory_subsys);
Dave Hansen3947be12005-10-29 18:16:54 -070092
Badari Pulavarty00a41db2008-02-11 09:23:18 -080093 /* drop the ref. we got in remove_memory_block() */
Kay Sievers10fbcf42011-12-21 14:48:43 -080094 kobject_put(&memory->dev.kobj);
95 device_unregister(&memory->dev);
Dave Hansen3947be12005-10-29 18:16:54 -070096}
97
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -060098unsigned long __weak memory_block_size_bytes(void)
99{
100 return MIN_MEMORY_BLOCK_SIZE;
101}
102
103static unsigned long get_memory_block_size(void)
104{
105 unsigned long block_sz;
106
107 block_sz = memory_block_size_bytes();
108
109 /* Validate blk_sz is a power of 2 and not less than section size */
110 if ((block_sz & (block_sz - 1)) || (block_sz < MIN_MEMORY_BLOCK_SIZE)) {
111 WARN_ON(1);
112 block_sz = MIN_MEMORY_BLOCK_SIZE;
113 }
114
115 return block_sz;
116}
117
Dave Hansen3947be12005-10-29 18:16:54 -0700118/*
119 * use this as the physical section index that this memsection
120 * uses.
121 */
122
Kay Sievers10fbcf42011-12-21 14:48:43 -0800123static ssize_t show_mem_start_phys_index(struct device *dev,
124 struct device_attribute *attr, char *buf)
Dave Hansen3947be12005-10-29 18:16:54 -0700125{
126 struct memory_block *mem =
Kay Sievers10fbcf42011-12-21 14:48:43 -0800127 container_of(dev, struct memory_block, dev);
Nathan Fontenotd3360162011-01-20 10:44:29 -0600128 unsigned long phys_index;
129
130 phys_index = mem->start_section_nr / sections_per_block;
131 return sprintf(buf, "%08lx\n", phys_index);
132}
133
Kay Sievers10fbcf42011-12-21 14:48:43 -0800134static ssize_t show_mem_end_phys_index(struct device *dev,
135 struct device_attribute *attr, char *buf)
Nathan Fontenotd3360162011-01-20 10:44:29 -0600136{
137 struct memory_block *mem =
Kay Sievers10fbcf42011-12-21 14:48:43 -0800138 container_of(dev, struct memory_block, dev);
Nathan Fontenotd3360162011-01-20 10:44:29 -0600139 unsigned long phys_index;
140
141 phys_index = mem->end_section_nr / sections_per_block;
142 return sprintf(buf, "%08lx\n", phys_index);
Dave Hansen3947be12005-10-29 18:16:54 -0700143}
144
145/*
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700146 * Show whether the section of memory is likely to be hot-removable
147 */
Kay Sievers10fbcf42011-12-21 14:48:43 -0800148static ssize_t show_mem_removable(struct device *dev,
149 struct device_attribute *attr, char *buf)
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700150{
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600151 unsigned long i, pfn;
152 int ret = 1;
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700153 struct memory_block *mem =
Kay Sievers10fbcf42011-12-21 14:48:43 -0800154 container_of(dev, struct memory_block, dev);
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700155
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600156 for (i = 0; i < sections_per_block; i++) {
Nathan Fontenotd3360162011-01-20 10:44:29 -0600157 pfn = section_nr_to_pfn(mem->start_section_nr + i);
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600158 ret &= is_mem_section_removable(pfn, PAGES_PER_SECTION);
159 }
160
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700161 return sprintf(buf, "%d\n", ret);
162}
163
164/*
Dave Hansen3947be12005-10-29 18:16:54 -0700165 * online, offline, going offline, etc.
166 */
Kay Sievers10fbcf42011-12-21 14:48:43 -0800167static ssize_t show_mem_state(struct device *dev,
168 struct device_attribute *attr, char *buf)
Dave Hansen3947be12005-10-29 18:16:54 -0700169{
170 struct memory_block *mem =
Kay Sievers10fbcf42011-12-21 14:48:43 -0800171 container_of(dev, struct memory_block, dev);
Dave Hansen3947be12005-10-29 18:16:54 -0700172 ssize_t len = 0;
173
174 /*
175 * We can probably put these states in a nice little array
176 * so that they're not open-coded
177 */
178 switch (mem->state) {
179 case MEM_ONLINE:
180 len = sprintf(buf, "online\n");
181 break;
182 case MEM_OFFLINE:
183 len = sprintf(buf, "offline\n");
184 break;
185 case MEM_GOING_OFFLINE:
186 len = sprintf(buf, "going-offline\n");
187 break;
188 default:
189 len = sprintf(buf, "ERROR-UNKNOWN-%ld\n",
190 mem->state);
191 WARN_ON(1);
192 break;
193 }
194
195 return len;
196}
197
Yasunori Goto7b78d332007-10-21 16:41:36 -0700198int memory_notify(unsigned long val, void *v)
Dave Hansen3947be12005-10-29 18:16:54 -0700199{
Alan Sterne041c682006-03-27 01:16:30 -0800200 return blocking_notifier_call_chain(&memory_chain, val, v);
Dave Hansen3947be12005-10-29 18:16:54 -0700201}
202
Robert Jennings925cc712009-12-17 14:44:38 +0000203int memory_isolate_notify(unsigned long val, void *v)
204{
205 return atomic_notifier_call_chain(&memory_isolate_chain, val, v);
206}
207
Dave Hansen3947be12005-10-29 18:16:54 -0700208/*
Mel Gorman2bbcb8782011-10-17 16:38:20 +0200209 * The probe routines leave the pages reserved, just as the bootmem code does.
210 * Make sure they're still that way.
211 */
212static bool pages_correctly_reserved(unsigned long start_pfn,
213 unsigned long nr_pages)
214{
215 int i, j;
216 struct page *page;
217 unsigned long pfn = start_pfn;
218
219 /*
220 * memmap between sections is not contiguous except with
221 * SPARSEMEM_VMEMMAP. We lookup the page once per section
222 * and assume memmap is contiguous within each section
223 */
224 for (i = 0; i < sections_per_block; i++, pfn += PAGES_PER_SECTION) {
225 if (WARN_ON_ONCE(!pfn_valid(pfn)))
226 return false;
227 page = pfn_to_page(pfn);
228
229 for (j = 0; j < PAGES_PER_SECTION; j++) {
230 if (PageReserved(page + j))
231 continue;
232
233 printk(KERN_WARNING "section number %ld page number %d "
234 "not reserved, was it already online?\n",
235 pfn_to_section_nr(pfn), j);
236
237 return false;
238 }
239 }
240
241 return true;
242}
243
244/*
Dave Hansen3947be12005-10-29 18:16:54 -0700245 * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is
246 * OK to have direct references to sparsemem variables in here.
247 */
248static int
Anton Blanchard5409d2c2011-05-11 17:25:14 +1000249memory_block_action(unsigned long phys_index, unsigned long action)
Dave Hansen3947be12005-10-29 18:16:54 -0700250{
Dave Hansen3947be12005-10-29 18:16:54 -0700251 unsigned long start_pfn, start_paddr;
Anton Blanchard5409d2c2011-05-11 17:25:14 +1000252 unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
Greg Kroah-Hartmande0ed362011-10-18 14:00:57 -0700253 struct page *first_page;
Dave Hansen3947be12005-10-29 18:16:54 -0700254 int ret;
Dave Hansen3947be12005-10-29 18:16:54 -0700255
Greg Kroah-Hartmande0ed362011-10-18 14:00:57 -0700256 first_page = pfn_to_page(phys_index << PFN_SECTION_SHIFT);
257
Dave Hansen3947be12005-10-29 18:16:54 -0700258 switch (action) {
259 case MEM_ONLINE:
Greg Kroah-Hartmande0ed362011-10-18 14:00:57 -0700260 start_pfn = page_to_pfn(first_page);
Mel Gorman2bbcb8782011-10-17 16:38:20 +0200261
262 if (!pages_correctly_reserved(start_pfn, nr_pages))
263 return -EBUSY;
264
Anton Blanchard5409d2c2011-05-11 17:25:14 +1000265 ret = online_pages(start_pfn, nr_pages);
Dave Hansen3947be12005-10-29 18:16:54 -0700266 break;
267 case MEM_OFFLINE:
Greg Kroah-Hartmande0ed362011-10-18 14:00:57 -0700268 start_paddr = page_to_pfn(first_page) << PAGE_SHIFT;
Dave Hansen3947be12005-10-29 18:16:54 -0700269 ret = remove_memory(start_paddr,
Anton Blanchard5409d2c2011-05-11 17:25:14 +1000270 nr_pages << PAGE_SHIFT);
Dave Hansen3947be12005-10-29 18:16:54 -0700271 break;
272 default:
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600273 WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: "
274 "%ld\n", __func__, phys_index, action, action);
Dave Hansen3947be12005-10-29 18:16:54 -0700275 ret = -EINVAL;
276 }
Dave Hansen3947be12005-10-29 18:16:54 -0700277
278 return ret;
279}
280
281static int memory_block_change_state(struct memory_block *mem,
282 unsigned long to_state, unsigned long from_state_req)
283{
Greg Kroah-Hartmande0ed362011-10-18 14:00:57 -0700284 int ret = 0;
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600285
Daniel Walkerda19cbc2008-02-04 23:35:47 -0800286 mutex_lock(&mem->state_mutex);
Dave Hansen3947be12005-10-29 18:16:54 -0700287
288 if (mem->state != from_state_req) {
289 ret = -EINVAL;
290 goto out;
291 }
292
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600293 if (to_state == MEM_OFFLINE)
294 mem->state = MEM_GOING_OFFLINE;
295
Anton Blanchard5409d2c2011-05-11 17:25:14 +1000296 ret = memory_block_action(mem->start_section_nr, to_state);
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600297
Michael Holzheuf5138e42012-01-12 17:20:23 -0800298 if (ret) {
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600299 mem->state = from_state_req;
Michael Holzheuf5138e42012-01-12 17:20:23 -0800300 goto out;
301 }
Dave Hansen3947be12005-10-29 18:16:54 -0700302
Michael Holzheuf5138e42012-01-12 17:20:23 -0800303 mem->state = to_state;
304 switch (mem->state) {
305 case MEM_OFFLINE:
306 kobject_uevent(&mem->dev.kobj, KOBJ_OFFLINE);
307 break;
308 case MEM_ONLINE:
309 kobject_uevent(&mem->dev.kobj, KOBJ_ONLINE);
310 break;
311 default:
312 break;
313 }
Dave Hansen3947be12005-10-29 18:16:54 -0700314out:
Daniel Walkerda19cbc2008-02-04 23:35:47 -0800315 mutex_unlock(&mem->state_mutex);
Dave Hansen3947be12005-10-29 18:16:54 -0700316 return ret;
317}
318
319static ssize_t
Kay Sievers10fbcf42011-12-21 14:48:43 -0800320store_mem_state(struct device *dev,
321 struct device_attribute *attr, const char *buf, size_t count)
Dave Hansen3947be12005-10-29 18:16:54 -0700322{
323 struct memory_block *mem;
Dave Hansen3947be12005-10-29 18:16:54 -0700324 int ret = -EINVAL;
325
Kay Sievers10fbcf42011-12-21 14:48:43 -0800326 mem = container_of(dev, struct memory_block, dev);
Dave Hansen3947be12005-10-29 18:16:54 -0700327
328 if (!strncmp(buf, "online", min((int)count, 6)))
329 ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
330 else if(!strncmp(buf, "offline", min((int)count, 7)))
331 ret = memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE);
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600332
Dave Hansen3947be12005-10-29 18:16:54 -0700333 if (ret)
334 return ret;
335 return count;
336}
337
338/*
339 * phys_device is a bad name for this. What I really want
340 * is a way to differentiate between memory ranges that
341 * are part of physical devices that constitute
342 * a complete removable unit or fru.
343 * i.e. do these ranges belong to the same physical device,
344 * s.t. if I offline all of these sections I can then
345 * remove the physical device?
346 */
Kay Sievers10fbcf42011-12-21 14:48:43 -0800347static ssize_t show_phys_device(struct device *dev,
348 struct device_attribute *attr, char *buf)
Dave Hansen3947be12005-10-29 18:16:54 -0700349{
350 struct memory_block *mem =
Kay Sievers10fbcf42011-12-21 14:48:43 -0800351 container_of(dev, struct memory_block, dev);
Dave Hansen3947be12005-10-29 18:16:54 -0700352 return sprintf(buf, "%d\n", mem->phys_device);
353}
354
Kay Sievers10fbcf42011-12-21 14:48:43 -0800355static DEVICE_ATTR(phys_index, 0444, show_mem_start_phys_index, NULL);
356static DEVICE_ATTR(end_phys_index, 0444, show_mem_end_phys_index, NULL);
357static DEVICE_ATTR(state, 0644, show_mem_state, store_mem_state);
358static DEVICE_ATTR(phys_device, 0444, show_phys_device, NULL);
359static DEVICE_ATTR(removable, 0444, show_mem_removable, NULL);
Dave Hansen3947be12005-10-29 18:16:54 -0700360
361#define mem_create_simple_file(mem, attr_name) \
Kay Sievers10fbcf42011-12-21 14:48:43 -0800362 device_create_file(&mem->dev, &dev_attr_##attr_name)
Dave Hansen3947be12005-10-29 18:16:54 -0700363#define mem_remove_simple_file(mem, attr_name) \
Kay Sievers10fbcf42011-12-21 14:48:43 -0800364 device_remove_file(&mem->dev, &dev_attr_##attr_name)
Dave Hansen3947be12005-10-29 18:16:54 -0700365
366/*
367 * Block size attribute stuff
368 */
369static ssize_t
Kay Sievers10fbcf42011-12-21 14:48:43 -0800370print_block_size(struct device *dev, struct device_attribute *attr,
Andi Kleen8564a6c2010-01-05 12:48:06 +0100371 char *buf)
Dave Hansen3947be12005-10-29 18:16:54 -0700372{
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600373 return sprintf(buf, "%lx\n", get_memory_block_size());
Dave Hansen3947be12005-10-29 18:16:54 -0700374}
375
Kay Sievers10fbcf42011-12-21 14:48:43 -0800376static DEVICE_ATTR(block_size_bytes, 0444, print_block_size, NULL);
Dave Hansen3947be12005-10-29 18:16:54 -0700377
378static int block_size_init(void)
379{
Kay Sievers10fbcf42011-12-21 14:48:43 -0800380 return device_create_file(memory_subsys.dev_root,
381 &dev_attr_block_size_bytes);
Dave Hansen3947be12005-10-29 18:16:54 -0700382}
383
384/*
385 * Some architectures will have custom drivers to do this, and
386 * will not need to do it from userspace. The fake hot-add code
387 * as well as ppc64 will do all of their discovery in userspace
388 * and will require this interface.
389 */
390#ifdef CONFIG_ARCH_MEMORY_PROBE
391static ssize_t
Kay Sievers10fbcf42011-12-21 14:48:43 -0800392memory_probe_store(struct device *dev, struct device_attribute *attr,
Andi Kleen28812fe2010-01-05 12:48:07 +0100393 const char *buf, size_t count)
Dave Hansen3947be12005-10-29 18:16:54 -0700394{
395 u64 phys_addr;
Yasunori Gotobc02af92006-06-27 02:53:30 -0700396 int nid;
Nathan Fontenot6add7cd2011-01-31 10:55:23 -0600397 int i, ret;
Anton Blanchard61b94fe2011-09-15 06:26:15 +1000398 unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block;
Dave Hansen3947be12005-10-29 18:16:54 -0700399
400 phys_addr = simple_strtoull(buf, NULL, 0);
401
Anton Blanchard61b94fe2011-09-15 06:26:15 +1000402 if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1))
403 return -EINVAL;
404
Nathan Fontenot6add7cd2011-01-31 10:55:23 -0600405 for (i = 0; i < sections_per_block; i++) {
406 nid = memory_add_physaddr_to_nid(phys_addr);
407 ret = add_memory(nid, phys_addr,
408 PAGES_PER_SECTION << PAGE_SHIFT);
409 if (ret)
Nikanth Karthikesan9f0af692011-03-24 11:46:18 +0530410 goto out;
Nathan Fontenot6add7cd2011-01-31 10:55:23 -0600411
412 phys_addr += MIN_MEMORY_BLOCK_SIZE;
413 }
Dave Hansen3947be12005-10-29 18:16:54 -0700414
Nikanth Karthikesan9f0af692011-03-24 11:46:18 +0530415 ret = count;
416out:
417 return ret;
Dave Hansen3947be12005-10-29 18:16:54 -0700418}
Kay Sievers10fbcf42011-12-21 14:48:43 -0800419static DEVICE_ATTR(probe, S_IWUSR, NULL, memory_probe_store);
Dave Hansen3947be12005-10-29 18:16:54 -0700420
421static int memory_probe_init(void)
422{
Kay Sievers10fbcf42011-12-21 14:48:43 -0800423 return device_create_file(memory_subsys.dev_root, &dev_attr_probe);
Dave Hansen3947be12005-10-29 18:16:54 -0700424}
425#else
Andrew Morton28ec24e2006-12-06 20:37:29 -0800426static inline int memory_probe_init(void)
427{
428 return 0;
429}
Dave Hansen3947be12005-10-29 18:16:54 -0700430#endif
431
Andi Kleenfacb6012009-12-16 12:20:00 +0100432#ifdef CONFIG_MEMORY_FAILURE
433/*
434 * Support for offlining pages of memory
435 */
436
437/* Soft offline a page */
438static ssize_t
Kay Sievers10fbcf42011-12-21 14:48:43 -0800439store_soft_offline_page(struct device *dev,
440 struct device_attribute *attr,
Andi Kleen28812fe2010-01-05 12:48:07 +0100441 const char *buf, size_t count)
Andi Kleenfacb6012009-12-16 12:20:00 +0100442{
443 int ret;
444 u64 pfn;
445 if (!capable(CAP_SYS_ADMIN))
446 return -EPERM;
447 if (strict_strtoull(buf, 0, &pfn) < 0)
448 return -EINVAL;
449 pfn >>= PAGE_SHIFT;
450 if (!pfn_valid(pfn))
451 return -ENXIO;
452 ret = soft_offline_page(pfn_to_page(pfn), 0);
453 return ret == 0 ? count : ret;
454}
455
456/* Forcibly offline a page, including killing processes. */
457static ssize_t
Kay Sievers10fbcf42011-12-21 14:48:43 -0800458store_hard_offline_page(struct device *dev,
459 struct device_attribute *attr,
Andi Kleen28812fe2010-01-05 12:48:07 +0100460 const char *buf, size_t count)
Andi Kleenfacb6012009-12-16 12:20:00 +0100461{
462 int ret;
463 u64 pfn;
464 if (!capable(CAP_SYS_ADMIN))
465 return -EPERM;
466 if (strict_strtoull(buf, 0, &pfn) < 0)
467 return -EINVAL;
468 pfn >>= PAGE_SHIFT;
469 ret = __memory_failure(pfn, 0, 0);
470 return ret ? ret : count;
471}
472
Kay Sievers10fbcf42011-12-21 14:48:43 -0800473static DEVICE_ATTR(soft_offline_page, 0644, NULL, store_soft_offline_page);
474static DEVICE_ATTR(hard_offline_page, 0644, NULL, store_hard_offline_page);
Andi Kleenfacb6012009-12-16 12:20:00 +0100475
476static __init int memory_fail_init(void)
477{
478 int err;
479
Kay Sievers10fbcf42011-12-21 14:48:43 -0800480 err = device_create_file(memory_subsys.dev_root,
481 &dev_attr_soft_offline_page);
Andi Kleenfacb6012009-12-16 12:20:00 +0100482 if (!err)
Kay Sievers10fbcf42011-12-21 14:48:43 -0800483 err = device_create_file(memory_subsys.dev_root,
484 &dev_attr_hard_offline_page);
Andi Kleenfacb6012009-12-16 12:20:00 +0100485 return err;
486}
487#else
488static inline int memory_fail_init(void)
489{
490 return 0;
491}
492#endif
493
Dave Hansen3947be12005-10-29 18:16:54 -0700494/*
495 * Note that phys_device is optional. It is here to allow for
496 * differentiation between which *physical* devices each
497 * section belongs to...
498 */
Heiko Carstensbc32df02010-03-15 00:35:03 -0400499int __weak arch_get_memory_phys_device(unsigned long start_pfn)
500{
501 return 0;
502}
Dave Hansen3947be12005-10-29 18:16:54 -0700503
Kay Sievers10fbcf42011-12-21 14:48:43 -0800504/*
505 * A reference for the returned object is held and the reference for the
506 * hinted object is released.
507 */
Robin Holt98383032010-09-29 14:00:55 -0500508struct memory_block *find_memory_block_hinted(struct mem_section *section,
509 struct memory_block *hint)
510{
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600511 int block_id = base_memory_block_id(__section_nr(section));
Kay Sievers10fbcf42011-12-21 14:48:43 -0800512 struct device *hintdev = hint ? &hint->dev : NULL;
513 struct device *dev;
Robin Holt98383032010-09-29 14:00:55 -0500514
Kay Sievers10fbcf42011-12-21 14:48:43 -0800515 dev = subsys_find_device_by_id(&memory_subsys, block_id, hintdev);
516 if (hint)
517 put_device(&hint->dev);
518 if (!dev)
Robin Holt98383032010-09-29 14:00:55 -0500519 return NULL;
Kay Sievers10fbcf42011-12-21 14:48:43 -0800520 return container_of(dev, struct memory_block, dev);
Robin Holt98383032010-09-29 14:00:55 -0500521}
522
Dave Hansen3947be12005-10-29 18:16:54 -0700523/*
524 * For now, we have a linear search to go find the appropriate
525 * memory_block corresponding to a particular phys_index. If
526 * this gets to be a real problem, we can always use a radix
527 * tree or something here.
528 *
Kay Sievers10fbcf42011-12-21 14:48:43 -0800529 * This could be made generic for all device subsystems.
Dave Hansen3947be12005-10-29 18:16:54 -0700530 */
Gary Hadec04fc582009-01-06 14:39:14 -0800531struct memory_block *find_memory_block(struct mem_section *section)
Dave Hansen3947be12005-10-29 18:16:54 -0700532{
Robin Holt98383032010-09-29 14:00:55 -0500533 return find_memory_block_hinted(section, NULL);
Dave Hansen3947be12005-10-29 18:16:54 -0700534}
535
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600536static int init_memory_block(struct memory_block **memory,
537 struct mem_section *section, unsigned long state)
Nathan Fontenote4619c82010-10-19 12:44:20 -0500538{
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600539 struct memory_block *mem;
Nathan Fontenote4619c82010-10-19 12:44:20 -0500540 unsigned long start_pfn;
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600541 int scn_nr;
Nathan Fontenote4619c82010-10-19 12:44:20 -0500542 int ret = 0;
543
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600544 mem = kzalloc(sizeof(*mem), GFP_KERNEL);
Nathan Fontenote4619c82010-10-19 12:44:20 -0500545 if (!mem)
546 return -ENOMEM;
547
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600548 scn_nr = __section_nr(section);
Nathan Fontenotd3360162011-01-20 10:44:29 -0600549 mem->start_section_nr =
550 base_memory_block_id(scn_nr) * sections_per_block;
551 mem->end_section_nr = mem->start_section_nr + sections_per_block - 1;
Nathan Fontenote4619c82010-10-19 12:44:20 -0500552 mem->state = state;
Nathan Fontenot07681212010-10-19 12:46:19 -0500553 mem->section_count++;
Nathan Fontenote4619c82010-10-19 12:44:20 -0500554 mutex_init(&mem->state_mutex);
Nathan Fontenotd3360162011-01-20 10:44:29 -0600555 start_pfn = section_nr_to_pfn(mem->start_section_nr);
Nathan Fontenote4619c82010-10-19 12:44:20 -0500556 mem->phys_device = arch_get_memory_phys_device(start_pfn);
557
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600558 ret = register_memory(mem);
Nathan Fontenote4619c82010-10-19 12:44:20 -0500559 if (!ret)
560 ret = mem_create_simple_file(mem, phys_index);
561 if (!ret)
Nathan Fontenotd3360162011-01-20 10:44:29 -0600562 ret = mem_create_simple_file(mem, end_phys_index);
563 if (!ret)
Nathan Fontenote4619c82010-10-19 12:44:20 -0500564 ret = mem_create_simple_file(mem, state);
565 if (!ret)
566 ret = mem_create_simple_file(mem, phys_device);
567 if (!ret)
568 ret = mem_create_simple_file(mem, removable);
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600569
570 *memory = mem;
571 return ret;
572}
573
574static int add_memory_section(int nid, struct mem_section *section,
Yinghai Lu321bf4e2012-01-30 13:57:12 -0800575 struct memory_block **mem_p,
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600576 unsigned long state, enum mem_add_context context)
577{
Yinghai Lu321bf4e2012-01-30 13:57:12 -0800578 struct memory_block *mem = NULL;
579 int scn_nr = __section_nr(section);
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600580 int ret = 0;
581
582 mutex_lock(&mem_sysfs_mutex);
583
Yinghai Lu321bf4e2012-01-30 13:57:12 -0800584 if (context == BOOT) {
585 /* same memory block ? */
586 if (mem_p && *mem_p)
587 if (scn_nr >= (*mem_p)->start_section_nr &&
588 scn_nr <= (*mem_p)->end_section_nr) {
589 mem = *mem_p;
590 kobject_get(&mem->dev.kobj);
591 }
592 } else
593 mem = find_memory_block(section);
594
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600595 if (mem) {
596 mem->section_count++;
Kay Sievers10fbcf42011-12-21 14:48:43 -0800597 kobject_put(&mem->dev.kobj);
Yinghai Lu321bf4e2012-01-30 13:57:12 -0800598 } else {
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600599 ret = init_memory_block(&mem, section, state);
Yinghai Lu321bf4e2012-01-30 13:57:12 -0800600 /* store memory_block pointer for next loop */
601 if (!ret && context == BOOT)
602 if (mem_p)
603 *mem_p = mem;
604 }
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600605
Nathan Fontenote4619c82010-10-19 12:44:20 -0500606 if (!ret) {
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600607 if (context == HOTPLUG &&
608 mem->section_count == sections_per_block)
Nathan Fontenote4619c82010-10-19 12:44:20 -0500609 ret = register_mem_sect_under_node(mem, nid);
610 }
611
Nathan Fontenot2938ffb2010-10-19 12:45:24 -0500612 mutex_unlock(&mem_sysfs_mutex);
Nathan Fontenote4619c82010-10-19 12:44:20 -0500613 return ret;
614}
615
Dave Hansen3947be12005-10-29 18:16:54 -0700616int remove_memory_block(unsigned long node_id, struct mem_section *section,
617 int phys_device)
618{
619 struct memory_block *mem;
620
Nathan Fontenot2938ffb2010-10-19 12:45:24 -0500621 mutex_lock(&mem_sysfs_mutex);
Dave Hansen3947be12005-10-29 18:16:54 -0700622 mem = find_memory_block(section);
Nathan Fontenotd3360162011-01-20 10:44:29 -0600623 unregister_mem_sect_under_nodes(mem, __section_nr(section));
Nathan Fontenot07681212010-10-19 12:46:19 -0500624
625 mem->section_count--;
626 if (mem->section_count == 0) {
Nathan Fontenot07681212010-10-19 12:46:19 -0500627 mem_remove_simple_file(mem, phys_index);
Nathan Fontenotd3360162011-01-20 10:44:29 -0600628 mem_remove_simple_file(mem, end_phys_index);
Nathan Fontenot07681212010-10-19 12:46:19 -0500629 mem_remove_simple_file(mem, state);
630 mem_remove_simple_file(mem, phys_device);
631 mem_remove_simple_file(mem, removable);
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600632 unregister_memory(mem);
633 kfree(mem);
634 } else
Kay Sievers10fbcf42011-12-21 14:48:43 -0800635 kobject_put(&mem->dev.kobj);
Dave Hansen3947be12005-10-29 18:16:54 -0700636
Nathan Fontenot2938ffb2010-10-19 12:45:24 -0500637 mutex_unlock(&mem_sysfs_mutex);
Dave Hansen3947be12005-10-29 18:16:54 -0700638 return 0;
639}
640
641/*
642 * need an interface for the VM to add new memory regions,
643 * but without onlining it.
644 */
Gary Hadec04fc582009-01-06 14:39:14 -0800645int register_new_memory(int nid, struct mem_section *section)
Dave Hansen3947be12005-10-29 18:16:54 -0700646{
Yinghai Lu321bf4e2012-01-30 13:57:12 -0800647 return add_memory_section(nid, section, NULL, MEM_OFFLINE, HOTPLUG);
Dave Hansen3947be12005-10-29 18:16:54 -0700648}
649
650int unregister_memory_section(struct mem_section *section)
651{
Andy Whitcroft540557b2007-10-16 01:24:11 -0700652 if (!present_section(section))
Dave Hansen3947be12005-10-29 18:16:54 -0700653 return -EINVAL;
654
655 return remove_memory_block(0, section, 0);
656}
657
658/*
659 * Initialize the sysfs support for memory devices...
660 */
661int __init memory_dev_init(void)
662{
663 unsigned int i;
664 int ret;
Andrew Morton28ec24e2006-12-06 20:37:29 -0800665 int err;
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600666 unsigned long block_sz;
Yinghai Lu321bf4e2012-01-30 13:57:12 -0800667 struct memory_block *mem = NULL;
Dave Hansen3947be12005-10-29 18:16:54 -0700668
Kay Sievers10fbcf42011-12-21 14:48:43 -0800669 ret = subsys_system_register(&memory_subsys, NULL);
Andrew Morton28ec24e2006-12-06 20:37:29 -0800670 if (ret)
671 goto out;
Dave Hansen3947be12005-10-29 18:16:54 -0700672
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600673 block_sz = get_memory_block_size();
674 sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE;
675
Dave Hansen3947be12005-10-29 18:16:54 -0700676 /*
677 * Create entries for memory sections that were found
678 * during boot and have been initialized
679 */
680 for (i = 0; i < NR_MEM_SECTIONS; i++) {
Andy Whitcroft540557b2007-10-16 01:24:11 -0700681 if (!present_section_nr(i))
Dave Hansen3947be12005-10-29 18:16:54 -0700682 continue;
Yinghai Lu321bf4e2012-01-30 13:57:12 -0800683 /* don't need to reuse memory_block if only one per block */
684 err = add_memory_section(0, __nr_to_section(i),
685 (sections_per_block == 1) ? NULL : &mem,
686 MEM_ONLINE,
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600687 BOOT);
Andrew Morton28ec24e2006-12-06 20:37:29 -0800688 if (!ret)
689 ret = err;
Dave Hansen3947be12005-10-29 18:16:54 -0700690 }
691
Andrew Morton28ec24e2006-12-06 20:37:29 -0800692 err = memory_probe_init();
693 if (!ret)
694 ret = err;
Andi Kleenfacb6012009-12-16 12:20:00 +0100695 err = memory_fail_init();
696 if (!ret)
697 ret = err;
Andrew Morton28ec24e2006-12-06 20:37:29 -0800698 err = block_size_init();
699 if (!ret)
700 ret = err;
701out:
702 if (ret)
Harvey Harrison2b3a3022008-03-04 16:41:05 -0800703 printk(KERN_ERR "%s() failed: %d\n", __func__, ret);
Dave Hansen3947be12005-10-29 18:16:54 -0700704 return ret;
705}