1============== 2Memory Hotplug 3============== 4 5Created: Jul 28 2007 6Add description of notifier of memory hotplug Oct 11 2007 7 8This document is about memory hotplug including how-to-use and current status. 9Because Memory Hotplug is still under development, contents of this text will 10be changed often. 11 121. Introduction 13 1.1 purpose of memory hotplug 14 1.2. Phases of memory hotplug 15 1.3. Unit of Memory online/offline operation 162. Kernel Configuration 173. sysfs files for memory hotplug 184. Physical memory hot-add phase 19 4.1 Hardware(Firmware) Support 20 4.2 Notify memory hot-add event by hand 215. Logical Memory hot-add phase 22 5.1. State of memory 23 5.2. How to online memory 246. Logical memory remove 25 6.1 Memory offline and ZONE_MOVABLE 26 6.2. How to offline memory 277. Physical memory remove 288. Memory hotplug event notifier 299. Future Work List 30 31Note(1): x86_64's has special implementation for memory hotplug. 32 This text does not describe it. 33Note(2): This text assumes that sysfs is mounted at /sys. 34 35 36--------------- 371. Introduction 38--------------- 39 401.1 purpose of memory hotplug 41------------ 42Memory Hotplug allows users to increase/decrease the amount of memory. 43Generally, there are two purposes. 44 45(A) For changing the amount of memory. 46 This is to allow a feature like capacity on demand. 47(B) For installing/removing DIMMs or NUMA-nodes physically. 48 This is to exchange DIMMs/NUMA-nodes, reduce power consumption, etc. 49 50(A) is required by highly virtualized environments and (B) is required by 51hardware which supports memory power management. 52 53Linux memory hotplug is designed for both purpose. 54 55 561.2. Phases of memory hotplug 57--------------- 58There are 2 phases in Memory Hotplug. 59 1) Physical Memory Hotplug phase 60 2) Logical Memory Hotplug phase. 61 62The First phase is to communicate hardware/firmware and make/erase 63environment for hotplugged memory. Basically, this phase is necessary 64for the purpose (B), but this is good phase for communication between 65highly virtualized environments too. 66 67When memory is hotplugged, the kernel recognizes new memory, makes new memory 68management tables, and makes sysfs files for new memory's operation. 69 70If firmware supports notification of connection of new memory to OS, 71this phase is triggered automatically. ACPI can notify this event. If not, 72"probe" operation by system administration is used instead. 73(see Section 4.). 74 75Logical Memory Hotplug phase is to change memory state into 76available/unavailable for users. Amount of memory from user's view is 77changed by this phase. The kernel makes all memory in it as free pages 78when a memory range is available. 79 80In this document, this phase is described as online/offline. 81 82Logical Memory Hotplug phase is triggered by write of sysfs file by system 83administrator. For the hot-add case, it must be executed after Physical Hotplug 84phase by hand. 85(However, if you writes udev's hotplug scripts for memory hotplug, these 86 phases can be execute in seamless way.) 87 88 891.3. Unit of Memory online/offline operation 90------------ 91Memory hotplug uses SPARSEMEM memory model. SPARSEMEM divides the whole memory 92into chunks of the same size. The chunk is called a "section". The size of 93a section is architecture dependent. For example, power uses 16MiB, ia64 uses 941GiB. The unit of online/offline operation is "one section". (see Section 3.) 95 96To determine the size of sections, please read this file: 97 98/sys/devices/system/memory/block_size_bytes 99 100This file shows the size of sections in byte. 101 102----------------------- 1032. Kernel Configuration 104----------------------- 105To use memory hotplug feature, kernel must be compiled with following 106config options. 107 108- For all memory hotplug 109 Memory model -> Sparse Memory (CONFIG_SPARSEMEM) 110 Allow for memory hot-add (CONFIG_MEMORY_HOTPLUG) 111 112- To enable memory removal, the followings are also necessary 113 Allow for memory hot remove (CONFIG_MEMORY_HOTREMOVE) 114 Page Migration (CONFIG_MIGRATION) 115 116- For ACPI memory hotplug, the followings are also necessary 117 Memory hotplug (under ACPI Support menu) (CONFIG_ACPI_HOTPLUG_MEMORY) 118 This option can be kernel module. 119 120- As a related configuration, if your box has a feature of NUMA-node hotplug 121 via ACPI, then this option is necessary too. 122 ACPI0004,PNP0A05 and PNP0A06 Container Driver (under ACPI Support menu) 123 (CONFIG_ACPI_CONTAINER). 124 This option can be kernel module too. 125 126-------------------------------- 1274 sysfs files for memory hotplug 128-------------------------------- 129All sections have their device information in sysfs. Each section is part of 130a memory block under /sys/devices/system/memory as 131 132/sys/devices/system/memory/memoryXXX 133(XXX is the section id.) 134 135Now, XXX is defined as (start_address_of_section / section_size) of the first 136section contained in the memory block. The files 'phys_index' and 137'end_phys_index' under each directory report the beginning and end section id's 138for the memory block covered by the sysfs directory. It is expected that all 139memory sections in this range are present and no memory holes exist in the 140range. Currently there is no way to determine if there is a memory hole, but 141the existence of one should not affect the hotplug capabilities of the memory 142block. 143 144For example, assume 1GiB section size. A device for a memory starting at 1450x100000000 is /sys/device/system/memory/memory4 146(0x100000000 / 1Gib = 4) 147This device covers address range [0x100000000 ... 0x140000000) 148 149Under each section, you can see 4 or 5 files, the end_phys_index file being 150a recent addition and not present on older kernels. 151 152/sys/devices/system/memory/memoryXXX/start_phys_index 153/sys/devices/system/memory/memoryXXX/end_phys_index 154/sys/devices/system/memory/memoryXXX/phys_device 155/sys/devices/system/memory/memoryXXX/state 156/sys/devices/system/memory/memoryXXX/removable 157 158'phys_index' : read-only and contains section id of the first section 159 in the memory block, same as XXX. 160'end_phys_index' : read-only and contains section id of the last section 161 in the memory block. 162'state' : read-write 163 at read: contains online/offline state of memory. 164 at write: user can specify "online_kernel", 165 "online_movable", "online", "offline" command 166 which will be performed on al sections in the block. 167'phys_device' : read-only: designed to show the name of physical memory 168 device. This is not well implemented now. 169'removable' : read-only: contains an integer value indicating 170 whether the memory block is removable or not 171 removable. A value of 1 indicates that the memory 172 block is removable and a value of 0 indicates that 173 it is not removable. A memory block is removable only if 174 every section in the block is removable. 175 176NOTE: 177 These directories/files appear after physical memory hotplug phase. 178 179If CONFIG_NUMA is enabled the memoryXXX/ directories can also be accessed 180via symbolic links located in the /sys/devices/system/node/node* directories. 181 182For example: 183/sys/devices/system/node/node0/memory9 -> ../../memory/memory9 184 185A backlink will also be created: 186/sys/devices/system/memory/memory9/node0 -> ../../node/node0 187 188-------------------------------- 1894. Physical memory hot-add phase 190-------------------------------- 191 1924.1 Hardware(Firmware) Support 193------------ 194On x86_64/ia64 platform, memory hotplug by ACPI is supported. 195 196In general, the firmware (ACPI) which supports memory hotplug defines 197memory class object of _HID "PNP0C80". When a notify is asserted to PNP0C80, 198Linux's ACPI handler does hot-add memory to the system and calls a hotplug udev 199script. This will be done automatically. 200 201But scripts for memory hotplug are not contained in generic udev package(now). 202You may have to write it by yourself or online/offline memory by hand. 203Please see "How to online memory", "How to offline memory" in this text. 204 205If firmware supports NUMA-node hotplug, and defines an object _HID "ACPI0004", 206"PNP0A05", or "PNP0A06", notification is asserted to it, and ACPI handler 207calls hotplug code for all of objects which are defined in it. 208If memory device is found, memory hotplug code will be called. 209 210 2114.2 Notify memory hot-add event by hand 212------------ 213In some environments, especially virtualized environment, firmware will not 214notify memory hotplug event to the kernel. For such environment, "probe" 215interface is supported. This interface depends on CONFIG_ARCH_MEMORY_PROBE. 216 217Now, CONFIG_ARCH_MEMORY_PROBE is supported only by powerpc but it does not 218contain highly architecture codes. Please add config if you need "probe" 219interface. 220 221Probe interface is located at 222/sys/devices/system/memory/probe 223 224You can tell the physical address of new memory to the kernel by 225 226% echo start_address_of_new_memory > /sys/devices/system/memory/probe 227 228Then, [start_address_of_new_memory, start_address_of_new_memory + section_size) 229memory range is hot-added. In this case, hotplug script is not called (in 230current implementation). You'll have to online memory by yourself. 231Please see "How to online memory" in this text. 232 233 234 235------------------------------ 2365. Logical Memory hot-add phase 237------------------------------ 238 2395.1. State of memory 240------------ 241To see (online/offline) state of memory section, read 'state' file. 242 243% cat /sys/device/system/memory/memoryXXX/state 244 245 246If the memory section is online, you'll read "online". 247If the memory section is offline, you'll read "offline". 248 249 2505.2. How to online memory 251------------ 252Even if the memory is hot-added, it is not at ready-to-use state. 253For using newly added memory, you have to "online" the memory section. 254 255For onlining, you have to write "online" to the section's state file as: 256 257% echo online > /sys/devices/system/memory/memoryXXX/state 258 259This onlining will not change the ZONE type of the target memory section, 260If the memory section is in ZONE_NORMAL, you can change it to ZONE_MOVABLE: 261 262% echo online_movable > /sys/devices/system/memory/memoryXXX/state 263(NOTE: current limit: this memory section must be adjacent to ZONE_MOVABLE) 264 265And if the memory section is in ZONE_MOVABLE, you can change it to ZONE_NORMAL: 266 267% echo online_kernel > /sys/devices/system/memory/memoryXXX/state 268(NOTE: current limit: this memory section must be adjacent to ZONE_NORMAL) 269 270After this, section memoryXXX's state will be 'online' and the amount of 271available memory will be increased. 272 273Currently, newly added memory is added as ZONE_NORMAL (for powerpc, ZONE_DMA). 274This may be changed in future. 275 276 277 278------------------------ 2796. Logical memory remove 280------------------------ 281 2826.1 Memory offline and ZONE_MOVABLE 283------------ 284Memory offlining is more complicated than memory online. Because memory offline 285has to make the whole memory section be unused, memory offline can fail if 286the section includes memory which cannot be freed. 287 288In general, memory offline can use 2 techniques. 289 290(1) reclaim and free all memory in the section. 291(2) migrate all pages in the section. 292 293In the current implementation, Linux's memory offline uses method (2), freeing 294all pages in the section by page migration. But not all pages are 295migratable. Under current Linux, migratable pages are anonymous pages and 296page caches. For offlining a section by migration, the kernel has to guarantee 297that the section contains only migratable pages. 298 299Now, a boot option for making a section which consists of migratable pages is 300supported. By specifying "kernelcore=" or "movablecore=" boot option, you can 301create ZONE_MOVABLE...a zone which is just used for movable pages. 302(See also Documentation/kernel-parameters.txt) 303 304Assume the system has "TOTAL" amount of memory at boot time, this boot option 305creates ZONE_MOVABLE as following. 306 3071) When kernelcore=YYYY boot option is used, 308 Size of memory not for movable pages (not for offline) is YYYY. 309 Size of memory for movable pages (for offline) is TOTAL-YYYY. 310 3112) When movablecore=ZZZZ boot option is used, 312 Size of memory not for movable pages (not for offline) is TOTAL - ZZZZ. 313 Size of memory for movable pages (for offline) is ZZZZ. 314 315 316Note) Unfortunately, there is no information to show which section belongs 317to ZONE_MOVABLE. This is TBD. 318 319 3206.2. How to offline memory 321------------ 322You can offline a section by using the same sysfs interface that was used in 323memory onlining. 324 325% echo offline > /sys/devices/system/memory/memoryXXX/state 326 327If offline succeeds, the state of the memory section is changed to be "offline". 328If it fails, some error core (like -EBUSY) will be returned by the kernel. 329Even if a section does not belong to ZONE_MOVABLE, you can try to offline it. 330If it doesn't contain 'unmovable' memory, you'll get success. 331 332A section under ZONE_MOVABLE is considered to be able to be offlined easily. 333But under some busy state, it may return -EBUSY. Even if a memory section 334cannot be offlined due to -EBUSY, you can retry offlining it and may be able to 335offline it (or not). 336(For example, a page is referred to by some kernel internal call and released 337 soon.) 338 339Consideration: 340Memory hotplug's design direction is to make the possibility of memory offlining 341higher and to guarantee unplugging memory under any situation. But it needs 342more work. Returning -EBUSY under some situation may be good because the user 343can decide to retry more or not by himself. Currently, memory offlining code 344does some amount of retry with 120 seconds timeout. 345 346------------------------- 3477. Physical memory remove 348------------------------- 349Need more implementation yet.... 350 - Notification completion of remove works by OS to firmware. 351 - Guard from remove if not yet. 352 353-------------------------------- 3548. Memory hotplug event notifier 355-------------------------------- 356Memory hotplug has event notifier. There are 6 types of notification. 357 358MEMORY_GOING_ONLINE 359 Generated before new memory becomes available in order to be able to 360 prepare subsystems to handle memory. The page allocator is still unable 361 to allocate from the new memory. 362 363MEMORY_CANCEL_ONLINE 364 Generated if MEMORY_GOING_ONLINE fails. 365 366MEMORY_ONLINE 367 Generated when memory has successfully brought online. The callback may 368 allocate pages from the new memory. 369 370MEMORY_GOING_OFFLINE 371 Generated to begin the process of offlining memory. Allocations are no 372 longer possible from the memory but some of the memory to be offlined 373 is still in use. The callback can be used to free memory known to a 374 subsystem from the indicated memory section. 375 376MEMORY_CANCEL_OFFLINE 377 Generated if MEMORY_GOING_OFFLINE fails. Memory is available again from 378 the section that we attempted to offline. 379 380MEMORY_OFFLINE 381 Generated after offlining memory is complete. 382 383A callback routine can be registered by 384 hotplug_memory_notifier(callback_func, priority) 385 386The second argument of callback function (action) is event types of above. 387The third argument is passed by pointer of struct memory_notify. 388 389struct memory_notify { 390 unsigned long start_pfn; 391 unsigned long nr_pages; 392 int status_change_nid_normal; 393 int status_change_nid_high; 394 int status_change_nid; 395} 396 397start_pfn is start_pfn of online/offline memory. 398nr_pages is # of pages of online/offline memory. 399status_change_nid_normal is set node id when N_NORMAL_MEMORY of nodemask 400is (will be) set/clear, if this is -1, then nodemask status is not changed. 401status_change_nid_high is set node id when N_HIGH_MEMORY of nodemask 402is (will be) set/clear, if this is -1, then nodemask status is not changed. 403status_change_nid is set node id when N_MEMORY of nodemask is (will be) 404set/clear. It means a new(memoryless) node gets new memory by online and a 405node loses all memory. If this is -1, then nodemask status is not changed. 406If status_changed_nid* >= 0, callback should create/discard structures for the 407node if necessary. 408 409-------------- 4109. Future Work 411-------------- 412 - allowing memory hot-add to ZONE_MOVABLE. maybe we need some switch like 413 sysctl or new control file. 414 - showing memory section and physical device relationship. 415 - showing memory section is under ZONE_MOVABLE or not 416 - test and make it better memory offlining. 417 - support HugeTLB page migration and offlining. 418 - memmap removing at memory offline. 419 - physical remove memory. 420 421