blob: c5cdd190b7816ce446d6d7b06235c5f995c4c3f2 [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>
Dave Hansen3947be12005-10-29 18:16:54 -070019#include <linux/memory_hotplug.h>
20#include <linux/mm.h>
Daniel Walkerda19cbc2008-02-04 23:35:47 -080021#include <linux/mutex.h>
Shaohua Li9f1b16a2008-10-18 20:27:12 -070022#include <linux/stat.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090023#include <linux/slab.h>
Shaohua Li9f1b16a2008-10-18 20:27:12 -070024
Arun Sharma600634972011-07-26 16:09:06 -070025#include <linux/atomic.h>
Dave Hansen3947be12005-10-29 18:16:54 -070026#include <asm/uaccess.h>
27
Nathan Fontenot2938ffb2010-10-19 12:45:24 -050028static DEFINE_MUTEX(mem_sysfs_mutex);
29
Dave Hansen3947be12005-10-29 18:16:54 -070030#define MEMORY_CLASS_NAME "memory"
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -060031
Gu Zheng7315f0c2013-08-28 14:38:27 +080032#define to_memory_block(dev) container_of(dev, struct memory_block, dev)
33
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -060034static int sections_per_block;
35
36static inline int base_memory_block_id(int section_nr)
37{
38 return section_nr / sections_per_block;
39}
Dave Hansen3947be12005-10-29 18:16:54 -070040
Rafael J. Wysocki4960e052013-05-08 14:18:37 +020041static int memory_subsys_online(struct device *dev);
42static int memory_subsys_offline(struct device *dev);
43
Kay Sievers10fbcf42011-12-21 14:48:43 -080044static struct bus_type memory_subsys = {
Kay Sieversaf5ca3f2007-12-20 02:09:39 +010045 .name = MEMORY_CLASS_NAME,
Kay Sievers10fbcf42011-12-21 14:48:43 -080046 .dev_name = MEMORY_CLASS_NAME,
Rafael J. Wysocki4960e052013-05-08 14:18:37 +020047 .online = memory_subsys_online,
48 .offline = memory_subsys_offline,
Dave Hansen3947be12005-10-29 18:16:54 -070049};
50
Alan Sterne041c682006-03-27 01:16:30 -080051static BLOCKING_NOTIFIER_HEAD(memory_chain);
Dave Hansen3947be12005-10-29 18:16:54 -070052
Andy Whitcroft98a38eb2006-01-06 00:10:35 -080053int register_memory_notifier(struct notifier_block *nb)
Dave Hansen3947be12005-10-29 18:16:54 -070054{
Ioana Ciornei2aeebca2015-03-08 12:48:35 +020055 return blocking_notifier_chain_register(&memory_chain, nb);
Dave Hansen3947be12005-10-29 18:16:54 -070056}
Hannes Hering3c82c302008-05-07 14:43:01 +020057EXPORT_SYMBOL(register_memory_notifier);
Dave Hansen3947be12005-10-29 18:16:54 -070058
Andy Whitcroft98a38eb2006-01-06 00:10:35 -080059void unregister_memory_notifier(struct notifier_block *nb)
Dave Hansen3947be12005-10-29 18:16:54 -070060{
Ioana Ciornei2aeebca2015-03-08 12:48:35 +020061 blocking_notifier_chain_unregister(&memory_chain, nb);
Dave Hansen3947be12005-10-29 18:16:54 -070062}
Hannes Hering3c82c302008-05-07 14:43:01 +020063EXPORT_SYMBOL(unregister_memory_notifier);
Dave Hansen3947be12005-10-29 18:16:54 -070064
Robert Jennings925cc712009-12-17 14:44:38 +000065static ATOMIC_NOTIFIER_HEAD(memory_isolate_chain);
66
67int register_memory_isolate_notifier(struct notifier_block *nb)
68{
69 return atomic_notifier_chain_register(&memory_isolate_chain, nb);
70}
71EXPORT_SYMBOL(register_memory_isolate_notifier);
72
73void unregister_memory_isolate_notifier(struct notifier_block *nb)
74{
75 atomic_notifier_chain_unregister(&memory_isolate_chain, nb);
76}
77EXPORT_SYMBOL(unregister_memory_isolate_notifier);
78
Yasuaki Ishimatsufa7194e2012-12-11 16:00:44 -080079static void memory_block_release(struct device *dev)
80{
Gu Zheng7315f0c2013-08-28 14:38:27 +080081 struct memory_block *mem = to_memory_block(dev);
Yasuaki Ishimatsufa7194e2012-12-11 16:00:44 -080082
83 kfree(mem);
84}
85
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -060086unsigned long __weak memory_block_size_bytes(void)
87{
88 return MIN_MEMORY_BLOCK_SIZE;
89}
90
91static unsigned long get_memory_block_size(void)
92{
93 unsigned long block_sz;
94
95 block_sz = memory_block_size_bytes();
96
97 /* Validate blk_sz is a power of 2 and not less than section size */
98 if ((block_sz & (block_sz - 1)) || (block_sz < MIN_MEMORY_BLOCK_SIZE)) {
99 WARN_ON(1);
100 block_sz = MIN_MEMORY_BLOCK_SIZE;
101 }
102
103 return block_sz;
104}
105
Dave Hansen3947be12005-10-29 18:16:54 -0700106/*
107 * use this as the physical section index that this memsection
108 * uses.
109 */
110
Kay Sievers10fbcf42011-12-21 14:48:43 -0800111static ssize_t show_mem_start_phys_index(struct device *dev,
112 struct device_attribute *attr, char *buf)
Dave Hansen3947be12005-10-29 18:16:54 -0700113{
Gu Zheng7315f0c2013-08-28 14:38:27 +0800114 struct memory_block *mem = to_memory_block(dev);
Nathan Fontenotd3360162011-01-20 10:44:29 -0600115 unsigned long phys_index;
116
117 phys_index = mem->start_section_nr / sections_per_block;
118 return sprintf(buf, "%08lx\n", phys_index);
119}
120
Dave Hansen3947be12005-10-29 18:16:54 -0700121/*
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700122 * Show whether the section of memory is likely to be hot-removable
123 */
Kay Sievers10fbcf42011-12-21 14:48:43 -0800124static ssize_t show_mem_removable(struct device *dev,
125 struct device_attribute *attr, char *buf)
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700126{
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600127 unsigned long i, pfn;
128 int ret = 1;
Gu Zheng7315f0c2013-08-28 14:38:27 +0800129 struct memory_block *mem = to_memory_block(dev);
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700130
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600131 for (i = 0; i < sections_per_block; i++) {
Russ Anderson21ea9f52013-08-28 16:35:18 -0700132 if (!present_section_nr(mem->start_section_nr + i))
133 continue;
Nathan Fontenotd3360162011-01-20 10:44:29 -0600134 pfn = section_nr_to_pfn(mem->start_section_nr + i);
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600135 ret &= is_mem_section_removable(pfn, PAGES_PER_SECTION);
136 }
137
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700138 return sprintf(buf, "%d\n", ret);
139}
140
141/*
Dave Hansen3947be12005-10-29 18:16:54 -0700142 * online, offline, going offline, etc.
143 */
Kay Sievers10fbcf42011-12-21 14:48:43 -0800144static ssize_t show_mem_state(struct device *dev,
145 struct device_attribute *attr, char *buf)
Dave Hansen3947be12005-10-29 18:16:54 -0700146{
Gu Zheng7315f0c2013-08-28 14:38:27 +0800147 struct memory_block *mem = to_memory_block(dev);
Dave Hansen3947be12005-10-29 18:16:54 -0700148 ssize_t len = 0;
149
150 /*
151 * We can probably put these states in a nice little array
152 * so that they're not open-coded
153 */
154 switch (mem->state) {
Ioana Ciornei3d3af6a2015-03-08 12:29:04 +0200155 case MEM_ONLINE:
156 len = sprintf(buf, "online\n");
157 break;
158 case MEM_OFFLINE:
159 len = sprintf(buf, "offline\n");
160 break;
161 case MEM_GOING_OFFLINE:
162 len = sprintf(buf, "going-offline\n");
163 break;
164 default:
165 len = sprintf(buf, "ERROR-UNKNOWN-%ld\n",
166 mem->state);
167 WARN_ON(1);
168 break;
Dave Hansen3947be12005-10-29 18:16:54 -0700169 }
170
171 return len;
172}
173
Yasunori Goto7b78d332007-10-21 16:41:36 -0700174int memory_notify(unsigned long val, void *v)
Dave Hansen3947be12005-10-29 18:16:54 -0700175{
Alan Sterne041c682006-03-27 01:16:30 -0800176 return blocking_notifier_call_chain(&memory_chain, val, v);
Dave Hansen3947be12005-10-29 18:16:54 -0700177}
178
Robert Jennings925cc712009-12-17 14:44:38 +0000179int memory_isolate_notify(unsigned long val, void *v)
180{
181 return atomic_notifier_call_chain(&memory_isolate_chain, val, v);
182}
183
Dave Hansen3947be12005-10-29 18:16:54 -0700184/*
Mel Gorman2bbcb8782011-10-17 16:38:20 +0200185 * The probe routines leave the pages reserved, just as the bootmem code does.
186 * Make sure they're still that way.
187 */
Tang Chen6056d612013-04-29 15:08:40 -0700188static bool pages_correctly_reserved(unsigned long start_pfn)
Mel Gorman2bbcb8782011-10-17 16:38:20 +0200189{
190 int i, j;
191 struct page *page;
192 unsigned long pfn = start_pfn;
193
194 /*
195 * memmap between sections is not contiguous except with
196 * SPARSEMEM_VMEMMAP. We lookup the page once per section
197 * and assume memmap is contiguous within each section
198 */
199 for (i = 0; i < sections_per_block; i++, pfn += PAGES_PER_SECTION) {
200 if (WARN_ON_ONCE(!pfn_valid(pfn)))
201 return false;
202 page = pfn_to_page(pfn);
203
204 for (j = 0; j < PAGES_PER_SECTION; j++) {
205 if (PageReserved(page + j))
206 continue;
207
208 printk(KERN_WARNING "section number %ld page number %d "
209 "not reserved, was it already online?\n",
210 pfn_to_section_nr(pfn), j);
211
212 return false;
213 }
214 }
215
216 return true;
217}
218
219/*
Dave Hansen3947be12005-10-29 18:16:54 -0700220 * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is
221 * OK to have direct references to sparsemem variables in here.
David Rientjes30467e02015-04-14 15:45:11 -0700222 * Must already be protected by mem_hotplug_begin().
Dave Hansen3947be12005-10-29 18:16:54 -0700223 */
224static int
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800225memory_block_action(unsigned long phys_index, unsigned long action, int online_type)
Dave Hansen3947be12005-10-29 18:16:54 -0700226{
Wen Congyanga16cee12012-10-08 16:33:58 -0700227 unsigned long start_pfn;
Anton Blanchard5409d2c2011-05-11 17:25:14 +1000228 unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
Greg Kroah-Hartmande0ed362011-10-18 14:00:57 -0700229 struct page *first_page;
Dave Hansen3947be12005-10-29 18:16:54 -0700230 int ret;
Dave Hansen3947be12005-10-29 18:16:54 -0700231
Sheng Yong19c07d52015-04-14 15:44:54 -0700232 start_pfn = section_nr_to_pfn(phys_index);
Zhang Zhen71fbd552014-12-12 16:55:33 -0800233 first_page = pfn_to_page(start_pfn);
Greg Kroah-Hartmande0ed362011-10-18 14:00:57 -0700234
Dave Hansen3947be12005-10-29 18:16:54 -0700235 switch (action) {
Ioana Ciornei3d3af6a2015-03-08 12:29:04 +0200236 case MEM_ONLINE:
237 if (!pages_correctly_reserved(start_pfn))
238 return -EBUSY;
Mel Gorman2bbcb8782011-10-17 16:38:20 +0200239
Ioana Ciornei3d3af6a2015-03-08 12:29:04 +0200240 ret = online_pages(start_pfn, nr_pages, online_type);
241 break;
242 case MEM_OFFLINE:
243 ret = offline_pages(start_pfn, nr_pages);
244 break;
245 default:
246 WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: "
247 "%ld\n", __func__, phys_index, action, action);
248 ret = -EINVAL;
Dave Hansen3947be12005-10-29 18:16:54 -0700249 }
Dave Hansen3947be12005-10-29 18:16:54 -0700250
251 return ret;
252}
253
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -0700254int memory_block_change_state(struct memory_block *mem,
Seth Jenningsfa2be402013-08-20 16:05:05 -0500255 unsigned long to_state, unsigned long from_state_req)
Dave Hansen3947be12005-10-29 18:16:54 -0700256{
Greg Kroah-Hartmande0ed362011-10-18 14:00:57 -0700257 int ret = 0;
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600258
Rafael J. Wysocki4960e052013-05-08 14:18:37 +0200259 if (mem->state != from_state_req)
260 return -EINVAL;
Dave Hansen3947be12005-10-29 18:16:54 -0700261
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600262 if (to_state == MEM_OFFLINE)
263 mem->state = MEM_GOING_OFFLINE;
264
Seth Jenningsfa2be402013-08-20 16:05:05 -0500265 ret = memory_block_action(mem->start_section_nr, to_state,
266 mem->online_type);
267
Rafael J. Wysockib2c064b2013-05-23 10:38:55 +0200268 mem->state = ret ? from_state_req : to_state;
Seth Jenningsfa2be402013-08-20 16:05:05 -0500269
Rafael J. Wysocki4960e052013-05-08 14:18:37 +0200270 return ret;
271}
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600272
Seth Jenningsfa2be402013-08-20 16:05:05 -0500273/* The device lock serializes operations on memory_subsys_[online|offline] */
Rafael J. Wysocki4960e052013-05-08 14:18:37 +0200274static int memory_subsys_online(struct device *dev)
275{
Gu Zheng7315f0c2013-08-28 14:38:27 +0800276 struct memory_block *mem = to_memory_block(dev);
Rafael J. Wysocki4960e052013-05-08 14:18:37 +0200277 int ret;
Dave Hansen3947be12005-10-29 18:16:54 -0700278
Seth Jenningsfa2be402013-08-20 16:05:05 -0500279 if (mem->state == MEM_ONLINE)
280 return 0;
Rafael J. Wysocki4960e052013-05-08 14:18:37 +0200281
Seth Jenningsfa2be402013-08-20 16:05:05 -0500282 /*
283 * If we are called from store_mem_state(), online_type will be
284 * set >= 0 Otherwise we were called from the device online
285 * attribute and need to set the online_type.
286 */
287 if (mem->online_type < 0)
Tang Chen4f7c6b42014-08-06 16:05:13 -0700288 mem->online_type = MMOP_ONLINE_KEEP;
Rafael J. Wysocki4960e052013-05-08 14:18:37 +0200289
David Rientjes30467e02015-04-14 15:45:11 -0700290 /* Already under protection of mem_hotplug_begin() */
Seth Jenningsfa2be402013-08-20 16:05:05 -0500291 ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
292
293 /* clear online_type */
294 mem->online_type = -1;
295
Rafael J. Wysocki4960e052013-05-08 14:18:37 +0200296 return ret;
297}
298
299static int memory_subsys_offline(struct device *dev)
300{
Gu Zheng7315f0c2013-08-28 14:38:27 +0800301 struct memory_block *mem = to_memory_block(dev);
Rafael J. Wysocki4960e052013-05-08 14:18:37 +0200302
Seth Jenningsfa2be402013-08-20 16:05:05 -0500303 if (mem->state == MEM_OFFLINE)
304 return 0;
Rafael J. Wysocki4960e052013-05-08 14:18:37 +0200305
Seth Jennings26bbe7e2015-12-11 13:40:57 -0800306 /* Can't offline block with non-present sections */
307 if (mem->section_count != sections_per_block)
308 return -EINVAL;
309
Seth Jenningsfa2be402013-08-20 16:05:05 -0500310 return memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE);
Rafael J. Wysocki4960e052013-05-08 14:18:37 +0200311}
312
Dave Hansen3947be12005-10-29 18:16:54 -0700313static ssize_t
Kay Sievers10fbcf42011-12-21 14:48:43 -0800314store_mem_state(struct device *dev,
315 struct device_attribute *attr, const char *buf, size_t count)
Dave Hansen3947be12005-10-29 18:16:54 -0700316{
Gu Zheng7315f0c2013-08-28 14:38:27 +0800317 struct memory_block *mem = to_memory_block(dev);
Seth Jenningsfa2be402013-08-20 16:05:05 -0500318 int ret, online_type;
Dave Hansen3947be12005-10-29 18:16:54 -0700319
Rafael J. Wysocki5e33bc42013-08-28 21:41:01 +0200320 ret = lock_device_hotplug_sysfs();
321 if (ret)
322 return ret;
Rafael J. Wysocki4960e052013-05-08 14:18:37 +0200323
Tang Chen1f6a6cc2014-08-06 16:05:11 -0700324 if (sysfs_streq(buf, "online_kernel"))
Tang Chen4f7c6b42014-08-06 16:05:13 -0700325 online_type = MMOP_ONLINE_KERNEL;
Tang Chen1f6a6cc2014-08-06 16:05:11 -0700326 else if (sysfs_streq(buf, "online_movable"))
Tang Chen4f7c6b42014-08-06 16:05:13 -0700327 online_type = MMOP_ONLINE_MOVABLE;
Tang Chen1f6a6cc2014-08-06 16:05:11 -0700328 else if (sysfs_streq(buf, "online"))
Tang Chen4f7c6b42014-08-06 16:05:13 -0700329 online_type = MMOP_ONLINE_KEEP;
Tang Chen1f6a6cc2014-08-06 16:05:11 -0700330 else if (sysfs_streq(buf, "offline"))
Tang Chen4f7c6b42014-08-06 16:05:13 -0700331 online_type = MMOP_OFFLINE;
Yasuaki Ishimatsua37f8632013-10-11 15:36:25 +0900332 else {
333 ret = -EINVAL;
334 goto err;
335 }
Seth Jenningsfa2be402013-08-20 16:05:05 -0500336
David Rientjes30467e02015-04-14 15:45:11 -0700337 /*
338 * Memory hotplug needs to hold mem_hotplug_begin() for probe to find
339 * the correct memory block to online before doing device_online(dev),
340 * which will take dev->mutex. Take the lock early to prevent an
341 * inversion, memory_subsys_online() callbacks will be implemented by
342 * assuming it's already protected.
343 */
344 mem_hotplug_begin();
345
Seth Jenningsfa2be402013-08-20 16:05:05 -0500346 switch (online_type) {
Tang Chen4f7c6b42014-08-06 16:05:13 -0700347 case MMOP_ONLINE_KERNEL:
348 case MMOP_ONLINE_MOVABLE:
349 case MMOP_ONLINE_KEEP:
Seth Jenningsfa2be402013-08-20 16:05:05 -0500350 mem->online_type = online_type;
351 ret = device_online(&mem->dev);
352 break;
Tang Chen4f7c6b42014-08-06 16:05:13 -0700353 case MMOP_OFFLINE:
Seth Jenningsfa2be402013-08-20 16:05:05 -0500354 ret = device_offline(&mem->dev);
355 break;
356 default:
357 ret = -EINVAL; /* should never happen */
Rafael J. Wysocki4960e052013-05-08 14:18:37 +0200358 }
Rafael J. Wysocki4960e052013-05-08 14:18:37 +0200359
David Rientjes30467e02015-04-14 15:45:11 -0700360 mem_hotplug_done();
Yasuaki Ishimatsua37f8632013-10-11 15:36:25 +0900361err:
Rafael J. Wysocki4960e052013-05-08 14:18:37 +0200362 unlock_device_hotplug();
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600363
Reza Arbabd66ba152016-10-07 17:00:15 -0700364 if (ret < 0)
Dave Hansen3947be12005-10-29 18:16:54 -0700365 return ret;
Reza Arbabd66ba152016-10-07 17:00:15 -0700366 if (ret)
367 return -EINVAL;
368
Dave Hansen3947be12005-10-29 18:16:54 -0700369 return count;
370}
371
372/*
373 * phys_device is a bad name for this. What I really want
374 * is a way to differentiate between memory ranges that
375 * are part of physical devices that constitute
376 * a complete removable unit or fru.
377 * i.e. do these ranges belong to the same physical device,
378 * s.t. if I offline all of these sections I can then
379 * remove the physical device?
380 */
Kay Sievers10fbcf42011-12-21 14:48:43 -0800381static ssize_t show_phys_device(struct device *dev,
382 struct device_attribute *attr, char *buf)
Dave Hansen3947be12005-10-29 18:16:54 -0700383{
Gu Zheng7315f0c2013-08-28 14:38:27 +0800384 struct memory_block *mem = to_memory_block(dev);
Dave Hansen3947be12005-10-29 18:16:54 -0700385 return sprintf(buf, "%d\n", mem->phys_device);
386}
387
Zhang Zhened2f2402014-10-09 15:26:31 -0700388#ifdef CONFIG_MEMORY_HOTREMOVE
389static ssize_t show_valid_zones(struct device *dev,
390 struct device_attribute *attr, char *buf)
391{
392 struct memory_block *mem = to_memory_block(dev);
393 unsigned long start_pfn, end_pfn;
Toshi Kani6cb04972017-02-03 13:13:23 -0800394 unsigned long valid_start, valid_end, valid_pages;
Zhang Zhened2f2402014-10-09 15:26:31 -0700395 unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
Zhang Zhened2f2402014-10-09 15:26:31 -0700396 struct zone *zone;
Reza Arbaba371d9f2016-07-26 15:22:27 -0700397 int zone_shift = 0;
Zhang Zhened2f2402014-10-09 15:26:31 -0700398
399 start_pfn = section_nr_to_pfn(mem->start_section_nr);
400 end_pfn = start_pfn + nr_pages;
Zhang Zhened2f2402014-10-09 15:26:31 -0700401
402 /* The block contains more than one zone can not be offlined. */
Toshi Kani6cb04972017-02-03 13:13:23 -0800403 if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, &valid_end))
Zhang Zhened2f2402014-10-09 15:26:31 -0700404 return sprintf(buf, "none\n");
405
Toshi Kani6cb04972017-02-03 13:13:23 -0800406 zone = page_zone(pfn_to_page(valid_start));
407 valid_pages = valid_end - valid_start;
Zhang Zhened2f2402014-10-09 15:26:31 -0700408
Reza Arbaba371d9f2016-07-26 15:22:27 -0700409 /* MMOP_ONLINE_KEEP */
410 sprintf(buf, "%s", zone->name);
411
412 /* MMOP_ONLINE_KERNEL */
Toshi Kani6cb04972017-02-03 13:13:23 -0800413 zone_can_shift(valid_start, valid_pages, ZONE_NORMAL, &zone_shift);
Reza Arbaba371d9f2016-07-26 15:22:27 -0700414 if (zone_shift) {
415 strcat(buf, " ");
416 strcat(buf, (zone + zone_shift)->name);
Zhang Zhened2f2402014-10-09 15:26:31 -0700417 }
418
Reza Arbaba371d9f2016-07-26 15:22:27 -0700419 /* MMOP_ONLINE_MOVABLE */
Toshi Kani6cb04972017-02-03 13:13:23 -0800420 zone_can_shift(valid_start, valid_pages, ZONE_MOVABLE, &zone_shift);
Reza Arbaba371d9f2016-07-26 15:22:27 -0700421 if (zone_shift) {
422 strcat(buf, " ");
423 strcat(buf, (zone + zone_shift)->name);
Zhang Zhened2f2402014-10-09 15:26:31 -0700424 }
425
Reza Arbaba371d9f2016-07-26 15:22:27 -0700426 strcat(buf, "\n");
427
428 return strlen(buf);
Zhang Zhened2f2402014-10-09 15:26:31 -0700429}
430static DEVICE_ATTR(valid_zones, 0444, show_valid_zones, NULL);
431#endif
432
Kay Sievers10fbcf42011-12-21 14:48:43 -0800433static DEVICE_ATTR(phys_index, 0444, show_mem_start_phys_index, NULL);
Kay Sievers10fbcf42011-12-21 14:48:43 -0800434static DEVICE_ATTR(state, 0644, show_mem_state, store_mem_state);
435static DEVICE_ATTR(phys_device, 0444, show_phys_device, NULL);
436static DEVICE_ATTR(removable, 0444, show_mem_removable, NULL);
Dave Hansen3947be12005-10-29 18:16:54 -0700437
Dave Hansen3947be12005-10-29 18:16:54 -0700438/*
439 * Block size attribute stuff
440 */
441static ssize_t
Kay Sievers10fbcf42011-12-21 14:48:43 -0800442print_block_size(struct device *dev, struct device_attribute *attr,
Andi Kleen8564a6c2010-01-05 12:48:06 +0100443 char *buf)
Dave Hansen3947be12005-10-29 18:16:54 -0700444{
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600445 return sprintf(buf, "%lx\n", get_memory_block_size());
Dave Hansen3947be12005-10-29 18:16:54 -0700446}
447
Kay Sievers10fbcf42011-12-21 14:48:43 -0800448static DEVICE_ATTR(block_size_bytes, 0444, print_block_size, NULL);
Dave Hansen3947be12005-10-29 18:16:54 -0700449
Dave Hansen3947be12005-10-29 18:16:54 -0700450/*
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -0700451 * Memory auto online policy.
452 */
453
454static ssize_t
455show_auto_online_blocks(struct device *dev, struct device_attribute *attr,
456 char *buf)
457{
458 if (memhp_auto_online)
459 return sprintf(buf, "online\n");
460 else
461 return sprintf(buf, "offline\n");
462}
463
464static ssize_t
465store_auto_online_blocks(struct device *dev, struct device_attribute *attr,
466 const char *buf, size_t count)
467{
468 if (sysfs_streq(buf, "online"))
469 memhp_auto_online = true;
470 else if (sysfs_streq(buf, "offline"))
471 memhp_auto_online = false;
472 else
473 return -EINVAL;
474
475 return count;
476}
477
478static DEVICE_ATTR(auto_online_blocks, 0644, show_auto_online_blocks,
479 store_auto_online_blocks);
480
481/*
Dave Hansen3947be12005-10-29 18:16:54 -0700482 * Some architectures will have custom drivers to do this, and
483 * will not need to do it from userspace. The fake hot-add code
484 * as well as ppc64 will do all of their discovery in userspace
485 * and will require this interface.
486 */
487#ifdef CONFIG_ARCH_MEMORY_PROBE
488static ssize_t
Kay Sievers10fbcf42011-12-21 14:48:43 -0800489memory_probe_store(struct device *dev, struct device_attribute *attr,
Andi Kleen28812fe2010-01-05 12:48:07 +0100490 const char *buf, size_t count)
Dave Hansen3947be12005-10-29 18:16:54 -0700491{
492 u64 phys_addr;
John Allencb5490a2016-01-14 15:22:16 -0800493 int nid, ret;
Anton Blanchard61b94fe2011-09-15 06:26:15 +1000494 unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block;
Dave Hansen3947be12005-10-29 18:16:54 -0700495
Zhang Zhenb69deb22014-08-06 16:06:06 -0700496 ret = kstrtoull(buf, 0, &phys_addr);
497 if (ret)
498 return ret;
Dave Hansen3947be12005-10-29 18:16:54 -0700499
Anton Blanchard61b94fe2011-09-15 06:26:15 +1000500 if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1))
501 return -EINVAL;
502
John Allencb5490a2016-01-14 15:22:16 -0800503 nid = memory_add_physaddr_to_nid(phys_addr);
504 ret = add_memory(nid, phys_addr,
505 MIN_MEMORY_BLOCK_SIZE * sections_per_block);
Nathan Fontenot6add7cd2011-01-31 10:55:23 -0600506
John Allencb5490a2016-01-14 15:22:16 -0800507 if (ret)
508 goto out;
Dave Hansen3947be12005-10-29 18:16:54 -0700509
Nikanth Karthikesan9f0af692011-03-24 11:46:18 +0530510 ret = count;
511out:
512 return ret;
Dave Hansen3947be12005-10-29 18:16:54 -0700513}
Dave Hansen3947be12005-10-29 18:16:54 -0700514
Nathan Fontenot96b2c0f2013-06-04 14:42:28 -0500515static DEVICE_ATTR(probe, S_IWUSR, NULL, memory_probe_store);
Dave Hansen3947be12005-10-29 18:16:54 -0700516#endif
517
Andi Kleenfacb6012009-12-16 12:20:00 +0100518#ifdef CONFIG_MEMORY_FAILURE
519/*
520 * Support for offlining pages of memory
521 */
522
523/* Soft offline a page */
524static ssize_t
Kay Sievers10fbcf42011-12-21 14:48:43 -0800525store_soft_offline_page(struct device *dev,
526 struct device_attribute *attr,
Andi Kleen28812fe2010-01-05 12:48:07 +0100527 const char *buf, size_t count)
Andi Kleenfacb6012009-12-16 12:20:00 +0100528{
529 int ret;
530 u64 pfn;
531 if (!capable(CAP_SYS_ADMIN))
532 return -EPERM;
Jingoo Han34da5e62013-07-26 13:10:22 +0900533 if (kstrtoull(buf, 0, &pfn) < 0)
Andi Kleenfacb6012009-12-16 12:20:00 +0100534 return -EINVAL;
535 pfn >>= PAGE_SHIFT;
536 if (!pfn_valid(pfn))
537 return -ENXIO;
538 ret = soft_offline_page(pfn_to_page(pfn), 0);
539 return ret == 0 ? count : ret;
540}
541
542/* Forcibly offline a page, including killing processes. */
543static ssize_t
Kay Sievers10fbcf42011-12-21 14:48:43 -0800544store_hard_offline_page(struct device *dev,
545 struct device_attribute *attr,
Andi Kleen28812fe2010-01-05 12:48:07 +0100546 const char *buf, size_t count)
Andi Kleenfacb6012009-12-16 12:20:00 +0100547{
548 int ret;
549 u64 pfn;
550 if (!capable(CAP_SYS_ADMIN))
551 return -EPERM;
Jingoo Han34da5e62013-07-26 13:10:22 +0900552 if (kstrtoull(buf, 0, &pfn) < 0)
Andi Kleenfacb6012009-12-16 12:20:00 +0100553 return -EINVAL;
554 pfn >>= PAGE_SHIFT;
Tony Luckcd42f4a2011-12-15 10:48:12 -0800555 ret = memory_failure(pfn, 0, 0);
Andi Kleenfacb6012009-12-16 12:20:00 +0100556 return ret ? ret : count;
557}
558
Felipe Balbi74fef7a2013-02-18 21:09:03 +0200559static DEVICE_ATTR(soft_offline_page, S_IWUSR, NULL, store_soft_offline_page);
560static DEVICE_ATTR(hard_offline_page, S_IWUSR, NULL, store_hard_offline_page);
Andi Kleenfacb6012009-12-16 12:20:00 +0100561#endif
562
Dave Hansen3947be12005-10-29 18:16:54 -0700563/*
564 * Note that phys_device is optional. It is here to allow for
565 * differentiation between which *physical* devices each
566 * section belongs to...
567 */
Heiko Carstensbc32df02010-03-15 00:35:03 -0400568int __weak arch_get_memory_phys_device(unsigned long start_pfn)
569{
570 return 0;
571}
Dave Hansen3947be12005-10-29 18:16:54 -0700572
Kay Sievers10fbcf42011-12-21 14:48:43 -0800573/*
574 * A reference for the returned object is held and the reference for the
575 * hinted object is released.
576 */
Robin Holt98383032010-09-29 14:00:55 -0500577struct memory_block *find_memory_block_hinted(struct mem_section *section,
578 struct memory_block *hint)
579{
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600580 int block_id = base_memory_block_id(__section_nr(section));
Kay Sievers10fbcf42011-12-21 14:48:43 -0800581 struct device *hintdev = hint ? &hint->dev : NULL;
582 struct device *dev;
Robin Holt98383032010-09-29 14:00:55 -0500583
Kay Sievers10fbcf42011-12-21 14:48:43 -0800584 dev = subsys_find_device_by_id(&memory_subsys, block_id, hintdev);
585 if (hint)
586 put_device(&hint->dev);
587 if (!dev)
Robin Holt98383032010-09-29 14:00:55 -0500588 return NULL;
Gu Zheng7315f0c2013-08-28 14:38:27 +0800589 return to_memory_block(dev);
Robin Holt98383032010-09-29 14:00:55 -0500590}
591
Dave Hansen3947be12005-10-29 18:16:54 -0700592/*
593 * For now, we have a linear search to go find the appropriate
594 * memory_block corresponding to a particular phys_index. If
595 * this gets to be a real problem, we can always use a radix
596 * tree or something here.
597 *
Kay Sievers10fbcf42011-12-21 14:48:43 -0800598 * This could be made generic for all device subsystems.
Dave Hansen3947be12005-10-29 18:16:54 -0700599 */
Gary Hadec04fc582009-01-06 14:39:14 -0800600struct memory_block *find_memory_block(struct mem_section *section)
Dave Hansen3947be12005-10-29 18:16:54 -0700601{
Robin Holt98383032010-09-29 14:00:55 -0500602 return find_memory_block_hinted(section, NULL);
Dave Hansen3947be12005-10-29 18:16:54 -0700603}
604
Nathan Fontenot96b2c0f2013-06-04 14:42:28 -0500605static struct attribute *memory_memblk_attrs[] = {
606 &dev_attr_phys_index.attr,
Nathan Fontenot96b2c0f2013-06-04 14:42:28 -0500607 &dev_attr_state.attr,
608 &dev_attr_phys_device.attr,
609 &dev_attr_removable.attr,
Zhang Zhened2f2402014-10-09 15:26:31 -0700610#ifdef CONFIG_MEMORY_HOTREMOVE
611 &dev_attr_valid_zones.attr,
612#endif
Nathan Fontenot96b2c0f2013-06-04 14:42:28 -0500613 NULL
614};
615
616static struct attribute_group memory_memblk_attr_group = {
617 .attrs = memory_memblk_attrs,
618};
619
620static const struct attribute_group *memory_memblk_attr_groups[] = {
621 &memory_memblk_attr_group,
622 NULL,
623};
624
625/*
626 * register_memory - Setup a sysfs device for a memory block
627 */
628static
629int register_memory(struct memory_block *memory)
630{
Nathan Fontenot96b2c0f2013-06-04 14:42:28 -0500631 memory->dev.bus = &memory_subsys;
632 memory->dev.id = memory->start_section_nr / sections_per_block;
633 memory->dev.release = memory_block_release;
634 memory->dev.groups = memory_memblk_attr_groups;
Linus Torvaldsf991fae2013-07-03 14:35:40 -0700635 memory->dev.offline = memory->state == MEM_OFFLINE;
Nathan Fontenot96b2c0f2013-06-04 14:42:28 -0500636
Seth Jennings879f1be2013-08-20 12:12:58 -0500637 return device_register(&memory->dev);
Nathan Fontenot96b2c0f2013-06-04 14:42:28 -0500638}
639
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600640static int init_memory_block(struct memory_block **memory,
641 struct mem_section *section, unsigned long state)
Nathan Fontenote4619c82010-10-19 12:44:20 -0500642{
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600643 struct memory_block *mem;
Nathan Fontenote4619c82010-10-19 12:44:20 -0500644 unsigned long start_pfn;
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600645 int scn_nr;
Nathan Fontenote4619c82010-10-19 12:44:20 -0500646 int ret = 0;
647
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600648 mem = kzalloc(sizeof(*mem), GFP_KERNEL);
Nathan Fontenote4619c82010-10-19 12:44:20 -0500649 if (!mem)
650 return -ENOMEM;
651
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600652 scn_nr = __section_nr(section);
Nathan Fontenotd3360162011-01-20 10:44:29 -0600653 mem->start_section_nr =
654 base_memory_block_id(scn_nr) * sections_per_block;
655 mem->end_section_nr = mem->start_section_nr + sections_per_block - 1;
Nathan Fontenote4619c82010-10-19 12:44:20 -0500656 mem->state = state;
Nathan Fontenotd3360162011-01-20 10:44:29 -0600657 start_pfn = section_nr_to_pfn(mem->start_section_nr);
Nathan Fontenote4619c82010-10-19 12:44:20 -0500658 mem->phys_device = arch_get_memory_phys_device(start_pfn);
659
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600660 ret = register_memory(mem);
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600661
662 *memory = mem;
663 return ret;
664}
665
Seth Jenningscb5e39b2013-08-20 12:13:03 -0500666static int add_memory_block(int base_section_nr)
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600667{
Seth Jenningscb5e39b2013-08-20 12:13:03 -0500668 struct memory_block *mem;
669 int i, ret, section_count = 0, section_nr;
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600670
Seth Jenningscb5e39b2013-08-20 12:13:03 -0500671 for (i = base_section_nr;
672 (i < base_section_nr + sections_per_block) && i < NR_MEM_SECTIONS;
673 i++) {
674 if (!present_section_nr(i))
675 continue;
676 if (section_count == 0)
677 section_nr = i;
678 section_count++;
Yinghai Lu321bf4e2012-01-30 13:57:12 -0800679 }
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600680
Seth Jenningscb5e39b2013-08-20 12:13:03 -0500681 if (section_count == 0)
682 return 0;
683 ret = init_memory_block(&mem, __nr_to_section(section_nr), MEM_ONLINE);
684 if (ret)
685 return ret;
686 mem->section_count = section_count;
687 return 0;
Nathan Fontenote4619c82010-10-19 12:44:20 -0500688}
689
Dan Williams260ae3f2016-01-15 16:56:17 -0800690static bool is_zone_device_section(struct mem_section *ms)
691{
692 struct page *page;
693
694 page = sparse_decode_mem_map(ms->section_mem_map, __section_nr(ms));
695 return is_zone_device_page(page);
696}
Seth Jenningscb5e39b2013-08-20 12:13:03 -0500697
David Rientjes4edd7ce2013-04-29 15:08:22 -0700698/*
699 * need an interface for the VM to add new memory regions,
700 * but without onlining it.
701 */
702int register_new_memory(int nid, struct mem_section *section)
703{
Seth Jenningsd7f80532013-08-20 12:13:00 -0500704 int ret = 0;
705 struct memory_block *mem;
Seth Jenningsb1eaef32013-08-20 12:12:57 -0500706
Dan Williams260ae3f2016-01-15 16:56:17 -0800707 if (is_zone_device_section(section))
708 return 0;
709
Seth Jenningsb1eaef32013-08-20 12:12:57 -0500710 mutex_lock(&mem_sysfs_mutex);
Seth Jenningsb1eaef32013-08-20 12:12:57 -0500711
Seth Jenningsd7f80532013-08-20 12:13:00 -0500712 mem = find_memory_block(section);
713 if (mem) {
714 mem->section_count++;
715 put_device(&mem->dev);
716 } else {
717 ret = init_memory_block(&mem, section, MEM_OFFLINE);
718 if (ret)
719 goto out;
Seth Jennings56c6b5d2016-01-14 15:20:21 -0800720 mem->section_count++;
Seth Jenningsd7f80532013-08-20 12:13:00 -0500721 }
722
723 if (mem->section_count == sections_per_block)
724 ret = register_mem_sect_under_node(mem, nid);
725out:
726 mutex_unlock(&mem_sysfs_mutex);
Seth Jenningsb1eaef32013-08-20 12:12:57 -0500727 return ret;
David Rientjes4edd7ce2013-04-29 15:08:22 -0700728}
729
730#ifdef CONFIG_MEMORY_HOTREMOVE
731static void
732unregister_memory(struct memory_block *memory)
733{
734 BUG_ON(memory->dev.bus != &memory_subsys);
735
736 /* drop the ref. we got in remove_memory_block() */
Seth Jenningsdf2b7172013-08-20 12:12:59 -0500737 put_device(&memory->dev);
David Rientjes4edd7ce2013-04-29 15:08:22 -0700738 device_unregister(&memory->dev);
739}
740
Seth Jenningscc292b02016-01-14 15:20:24 -0800741static int remove_memory_section(unsigned long node_id,
David Rientjes4edd7ce2013-04-29 15:08:22 -0700742 struct mem_section *section, int phys_device)
Dave Hansen3947be12005-10-29 18:16:54 -0700743{
744 struct memory_block *mem;
745
Dan Williams260ae3f2016-01-15 16:56:17 -0800746 if (is_zone_device_section(section))
747 return 0;
748
Nathan Fontenot2938ffb2010-10-19 12:45:24 -0500749 mutex_lock(&mem_sysfs_mutex);
Dave Hansen3947be12005-10-29 18:16:54 -0700750 mem = find_memory_block(section);
Nathan Fontenotd3360162011-01-20 10:44:29 -0600751 unregister_mem_sect_under_nodes(mem, __section_nr(section));
Nathan Fontenot07681212010-10-19 12:46:19 -0500752
753 mem->section_count--;
Nathan Fontenot96b2c0f2013-06-04 14:42:28 -0500754 if (mem->section_count == 0)
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600755 unregister_memory(mem);
Nathan Fontenot96b2c0f2013-06-04 14:42:28 -0500756 else
Seth Jenningsdf2b7172013-08-20 12:12:59 -0500757 put_device(&mem->dev);
Dave Hansen3947be12005-10-29 18:16:54 -0700758
Nathan Fontenot2938ffb2010-10-19 12:45:24 -0500759 mutex_unlock(&mem_sysfs_mutex);
Dave Hansen3947be12005-10-29 18:16:54 -0700760 return 0;
761}
762
Dave Hansen3947be12005-10-29 18:16:54 -0700763int unregister_memory_section(struct mem_section *section)
764{
Andy Whitcroft540557b2007-10-16 01:24:11 -0700765 if (!present_section(section))
Dave Hansen3947be12005-10-29 18:16:54 -0700766 return -EINVAL;
767
Seth Jenningscc292b02016-01-14 15:20:24 -0800768 return remove_memory_section(0, section, 0);
Dave Hansen3947be12005-10-29 18:16:54 -0700769}
David Rientjes4edd7ce2013-04-29 15:08:22 -0700770#endif /* CONFIG_MEMORY_HOTREMOVE */
Dave Hansen3947be12005-10-29 18:16:54 -0700771
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -0800772/* return true if the memory block is offlined, otherwise, return false */
773bool is_memblock_offlined(struct memory_block *mem)
774{
775 return mem->state == MEM_OFFLINE;
776}
777
Nathan Fontenot96b2c0f2013-06-04 14:42:28 -0500778static struct attribute *memory_root_attrs[] = {
779#ifdef CONFIG_ARCH_MEMORY_PROBE
780 &dev_attr_probe.attr,
781#endif
782
783#ifdef CONFIG_MEMORY_FAILURE
784 &dev_attr_soft_offline_page.attr,
785 &dev_attr_hard_offline_page.attr,
786#endif
787
788 &dev_attr_block_size_bytes.attr,
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -0700789 &dev_attr_auto_online_blocks.attr,
Nathan Fontenot96b2c0f2013-06-04 14:42:28 -0500790 NULL
791};
792
793static struct attribute_group memory_root_attr_group = {
794 .attrs = memory_root_attrs,
795};
796
797static const struct attribute_group *memory_root_attr_groups[] = {
798 &memory_root_attr_group,
799 NULL,
800};
801
Wen Congyange90bdb72012-10-08 16:34:01 -0700802/*
Dave Hansen3947be12005-10-29 18:16:54 -0700803 * Initialize the sysfs support for memory devices...
804 */
805int __init memory_dev_init(void)
806{
807 unsigned int i;
808 int ret;
Andrew Morton28ec24e2006-12-06 20:37:29 -0800809 int err;
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600810 unsigned long block_sz;
Dave Hansen3947be12005-10-29 18:16:54 -0700811
Nathan Fontenot96b2c0f2013-06-04 14:42:28 -0500812 ret = subsys_system_register(&memory_subsys, memory_root_attr_groups);
Andrew Morton28ec24e2006-12-06 20:37:29 -0800813 if (ret)
814 goto out;
Dave Hansen3947be12005-10-29 18:16:54 -0700815
Nathan Fontenot0c2c99b2011-01-20 10:43:34 -0600816 block_sz = get_memory_block_size();
817 sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE;
818
Dave Hansen3947be12005-10-29 18:16:54 -0700819 /*
820 * Create entries for memory sections that were found
821 * during boot and have been initialized
822 */
Seth Jenningsb1eaef32013-08-20 12:12:57 -0500823 mutex_lock(&mem_sysfs_mutex);
Seth Jenningscb5e39b2013-08-20 12:13:03 -0500824 for (i = 0; i < NR_MEM_SECTIONS; i += sections_per_block) {
825 err = add_memory_block(i);
Andrew Morton28ec24e2006-12-06 20:37:29 -0800826 if (!ret)
827 ret = err;
Dave Hansen3947be12005-10-29 18:16:54 -0700828 }
Seth Jenningsb1eaef32013-08-20 12:12:57 -0500829 mutex_unlock(&mem_sysfs_mutex);
Dave Hansen3947be12005-10-29 18:16:54 -0700830
Andrew Morton28ec24e2006-12-06 20:37:29 -0800831out:
832 if (ret)
Harvey Harrison2b3a3022008-03-04 16:41:05 -0800833 printk(KERN_ERR "%s() failed: %d\n", __func__, ret);
Dave Hansen3947be12005-10-29 18:16:54 -0700834 return ret;
835}