blob: 5fbcc22c98e944c1ec628bca656ec1eae73cec5e [file] [log] [blame]
Yasunori Goto6867c932007-08-10 13:00:59 -07001==============
2Memory Hotplug
3==============
4
5Last Updated: Jul 28 2007
6
7This document is about memory hotplug including how-to-use and current status.
8Because Memory Hotplug is still under development, contents of this text will
9be changed often.
10
111. Introduction
12 1.1 purpose of memory hotplug
13 1.2. Phases of memory hotplug
14 1.3. Unit of Memory online/offline operation
152. Kernel Configuration
163. sysfs files for memory hotplug
174. Physical memory hot-add phase
18 4.1 Hardware(Firmware) Support
19 4.2 Notify memory hot-add event by hand
205. Logical Memory hot-add phase
21 5.1. State of memory
22 5.2. How to online memory
236. Logical memory remove
24 6.1 Memory offline and ZONE_MOVABLE
25 6.2. How to offline memory
267. Physical memory remove
278. Future Work List
28
29Note(1): x86_64's has special implementation for memory hotplug.
30 This text does not describe it.
31Note(2): This text assumes that sysfs is mounted at /sys.
32
33
34---------------
351. Introduction
36---------------
37
381.1 purpose of memory hotplug
39------------
40Memory Hotplug allows users to increase/decrease the amount of memory.
41Generally, there are two purposes.
42
43(A) For changing the amount of memory.
44 This is to allow a feature like capacity on demand.
45(B) For installing/removing DIMMs or NUMA-nodes physically.
46 This is to exchange DIMMs/NUMA-nodes, reduce power consumption, etc.
47
48(A) is required by highly virtualized environments and (B) is required by
49hardware which supports memory power management.
50
51Linux memory hotplug is designed for both purpose.
52
53
541.2. Phases of memory hotplug
55---------------
56There are 2 phases in Memory Hotplug.
57 1) Physical Memory Hotplug phase
58 2) Logical Memory Hotplug phase.
59
60The First phase is to communicate hardware/firmware and make/erase
61environment for hotplugged memory. Basically, this phase is necessary
62for the purpose (B), but this is good phase for communication between
63highly virtualized environments too.
64
65When memory is hotplugged, the kernel recognizes new memory, makes new memory
66management tables, and makes sysfs files for new memory's operation.
67
68If firmware supports notification of connection of new memory to OS,
69this phase is triggered automatically. ACPI can notify this event. If not,
70"probe" operation by system administration is used instead.
71(see Section 4.).
72
73Logical Memory Hotplug phase is to change memory state into
74avaiable/unavailable for users. Amount of memory from user's view is
75changed by this phase. The kernel makes all memory in it as free pages
76when a memory range is available.
77
78In this document, this phase is described as online/offline.
79
80Logical Memory Hotplug phase is triggred by write of sysfs file by system
81administrator. For the hot-add case, it must be executed after Physical Hotplug
82phase by hand.
83(However, if you writes udev's hotplug scripts for memory hotplug, these
84 phases can be execute in seamless way.)
85
86
871.3. Unit of Memory online/offline operation
88------------
89Memory hotplug uses SPARSEMEM memory model. SPARSEMEM divides the whole memory
90into chunks of the same size. The chunk is called a "section". The size of
91a section is architecture dependent. For example, power uses 16MiB, ia64 uses
921GiB. The unit of online/offline operation is "one section". (see Section 3.)
93
94To determine the size of sections, please read this file:
95
96/sys/devices/system/memory/block_size_bytes
97
98This file shows the size of sections in byte.
99
100-----------------------
1012. Kernel Configuration
102-----------------------
103To use memory hotplug feature, kernel must be compiled with following
104config options.
105
106- For all memory hotplug
107 Memory model -> Sparse Memory (CONFIG_SPARSEMEM)
108 Allow for memory hot-add (CONFIG_MEMORY_HOTPLUG)
109
110- To enable memory removal, the followings are also necessary
111 Allow for memory hot remove (CONFIG_MEMORY_HOTREMOVE)
112 Page Migration (CONFIG_MIGRATION)
113
114- For ACPI memory hotplug, the followings are also necessary
115 Memory hotplug (under ACPI Support menu) (CONFIG_ACPI_HOTPLUG_MEMORY)
116 This option can be kernel module.
117
118- As a related configuration, if your box has a feature of NUMA-node hotplug
119 via ACPI, then this option is necessary too.
120 ACPI0004,PNP0A05 and PNP0A06 Container Driver (under ACPI Support menu)
121 (CONFIG_ACPI_CONTAINER).
122 This option can be kernel module too.
123
124--------------------------------
1253 sysfs files for memory hotplug
126--------------------------------
127All sections have their device information under /sys/devices/system/memory as
128
129/sys/devices/system/memory/memoryXXX
130(XXX is section id.)
131
132Now, XXX is defined as start_address_of_section / section_size.
133
134For example, assume 1GiB section size. A device for a memory starting at
1350x100000000 is /sys/device/system/memory/memory4
136(0x100000000 / 1Gib = 4)
137This device covers address range [0x100000000 ... 0x140000000)
138
139Under each section, you can see 3 files.
140
141/sys/devices/system/memory/memoryXXX/phys_index
142/sys/devices/system/memory/memoryXXX/phys_device
143/sys/devices/system/memory/memoryXXX/state
144
145'phys_index' : read-only and contains section id, same as XXX.
146'state' : read-write
147 at read: contains online/offline state of memory.
148 at write: user can specify "online", "offline" command
149'phys_device': read-only: designed to show the name of physical memory device.
150 This is not well implemented now.
151
152NOTE:
153 These directories/files appear after physical memory hotplug phase.
154
155
156--------------------------------
1574. Physical memory hot-add phase
158--------------------------------
159
1604.1 Hardware(Firmware) Support
161------------
162On x86_64/ia64 platform, memory hotplug by ACPI is supported.
163
164In general, the firmware (ACPI) which supports memory hotplug defines
165memory class object of _HID "PNP0C80". When a notify is asserted to PNP0C80,
166Linux's ACPI handler does hot-add memory to the system and calls a hotplug udev
167script. This will be done automatically.
168
169But scripts for memory hotplug are not contained in generic udev package(now).
170You may have to write it by yourself or online/offline memory by hand.
171Please see "How to online memory", "How to offline memory" in this text.
172
173If firmware supports NUMA-node hotplug, and defines an object _HID "ACPI0004",
174"PNP0A05", or "PNP0A06", notification is asserted to it, and ACPI handler
175calls hotplug code for all of objects which are defined in it.
176If memory device is found, memory hotplug code will be called.
177
178
1794.2 Notify memory hot-add event by hand
180------------
181In some environments, especially virtualized environment, firmware will not
182notify memory hotplug event to the kernel. For such environment, "probe"
183interface is supported. This interface depends on CONFIG_ARCH_MEMORY_PROBE.
184
185Now, CONFIG_ARCH_MEMORY_PROBE is supported only by powerpc but it does not
186contain highly architecture codes. Please add config if you need "probe"
187interface.
188
189Probe interface is located at
190/sys/devices/system/memory/probe
191
192You can tell the physical address of new memory to the kernel by
193
194% echo start_address_of_new_memory > /sys/devices/system/memory/probe
195
196Then, [start_address_of_new_memory, start_address_of_new_memory + section_size)
197memory range is hot-added. In this case, hotplug script is not called (in
198current implementation). You'll have to online memory by yourself.
199Please see "How to online memory" in this text.
200
201
202
203------------------------------
2045. Logical Memory hot-add phase
205------------------------------
206
2075.1. State of memory
208------------
209To see (online/offline) state of memory section, read 'state' file.
210
211% cat /sys/device/system/memory/memoryXXX/state
212
213
214If the memory section is online, you'll read "online".
215If the memory section is offline, you'll read "offline".
216
217
2185.2. How to online memory
219------------
220Even if the memory is hot-added, it is not at ready-to-use state.
221For using newly added memory, you have to "online" the memory section.
222
223For onlining, you have to write "online" to the section's state file as:
224
225% echo online > /sys/devices/system/memory/memoryXXX/state
226
227After this, section memoryXXX's state will be 'online' and the amount of
228available memory will be increased.
229
230Currently, newly added memory is added as ZONE_NORMAL (for powerpc, ZONE_DMA).
231This may be changed in future.
232
233
234
235------------------------
2366. Logical memory remove
237------------------------
238
2396.1 Memory offline and ZONE_MOVABLE
240------------
241Memory offlining is more complicated than memory online. Because memory offline
242has to make the whole memory section be unused, memory offline can fail if
243the section includes memory which cannot be freed.
244
245In general, memory offline can use 2 techniques.
246
247(1) reclaim and free all memory in the section.
248(2) migrate all pages in the section.
249
250In the current implementation, Linux's memory offline uses method (2), freeing
251all pages in the section by page migration. But not all pages are
252migratable. Under current Linux, migratable pages are anonymous pages and
253page caches. For offlining a section by migration, the kernel has to guarantee
254that the section contains only migratable pages.
255
256Now, a boot option for making a section which consists of migratable pages is
257supported. By specifying "kernelcore=" or "movablecore=" boot option, you can
258create ZONE_MOVABLE...a zone which is just used for movable pages.
259(See also Documentation/kernel-parameters.txt)
260
261Assume the system has "TOTAL" amount of memory at boot time, this boot option
262creates ZONE_MOVABLE as following.
263
2641) When kernelcore=YYYY boot option is used,
265 Size of memory not for movable pages (not for offline) is YYYY.
266 Size of memory for movable pages (for offline) is TOTAL-YYYY.
267
2682) When movablecore=ZZZZ boot option is used,
269 Size of memory not for movable pages (not for offline) is TOTAL - ZZZZ.
270 Size of memory for movable pages (for offline) is ZZZZ.
271
272
273Note) Unfortunately, there is no information to show which section belongs
274to ZONE_MOVABLE. This is TBD.
275
276
2776.2. How to offline memory
278------------
279You can offline a section by using the same sysfs interface that was used in
280memory onlining.
281
282% echo offline > /sys/devices/system/memory/memoryXXX/state
283
284If offline succeeds, the state of the memory section is changed to be "offline".
285If it fails, some error core (like -EBUSY) will be returned by the kernel.
286Even if a section does not belong to ZONE_MOVABLE, you can try to offline it.
287If it doesn't contain 'unmovable' memory, you'll get success.
288
289A section under ZONE_MOVABLE is considered to be able to be offlined easily.
290But under some busy state, it may return -EBUSY. Even if a memory section
291cannot be offlined due to -EBUSY, you can retry offlining it and may be able to
292offline it (or not).
293(For example, a page is referred to by some kernel internal call and released
294 soon.)
295
296Consideration:
297Memory hotplug's design direction is to make the possibility of memory offlining
298higher and to guarantee unplugging memory under any situation. But it needs
299more work. Returning -EBUSY under some situation may be good because the user
300can decide to retry more or not by himself. Currently, memory offlining code
301does some amount of retry with 120 seconds timeout.
302
303-------------------------
3047. Physical memory remove
305-------------------------
306Need more implementation yet....
307 - Notification completion of remove works by OS to firmware.
308 - Guard from remove if not yet.
309
310--------------
3118. Future Work
312--------------
313 - allowing memory hot-add to ZONE_MOVABLE. maybe we need some switch like
314 sysctl or new control file.
315 - showing memory section and physical device relationship.
316 - showing memory section and node relationship (maybe good for NUMA)
317 - showing memory section is under ZONE_MOVABLE or not
318 - test and make it better memory offlining.
319 - support HugeTLB page migration and offlining.
320 - memmap removing at memory offline.
321 - physical remove memory.
322