cpu-hotplug.txt 14 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386
  1. CPU hotplug Support in Linux(tm) Kernel
  2. Maintainers:
  3. CPU Hotplug Core:
  4. Rusty Russell <rusty@rustycorp.com.au>
  5. Srivatsa Vaddagiri <vatsa@in.ibm.com>
  6. i386:
  7. Zwane Mwaikambo <zwane@arm.linux.org.uk>
  8. ppc64:
  9. Nathan Lynch <nathanl@austin.ibm.com>
  10. Joel Schopp <jschopp@austin.ibm.com>
  11. ia64/x86_64:
  12. Ashok Raj <ashok.raj@intel.com>
  13. s390:
  14. Heiko Carstens <heiko.carstens@de.ibm.com>
  15. Authors: Ashok Raj <ashok.raj@intel.com>
  16. Lots of feedback: Nathan Lynch <nathanl@austin.ibm.com>,
  17. Joel Schopp <jschopp@austin.ibm.com>
  18. Introduction
  19. Modern advances in system architectures have introduced advanced error
  20. reporting and correction capabilities in processors. CPU architectures permit
  21. partitioning support, where compute resources of a single CPU could be made
  22. available to virtual machine environments. There are couple OEMS that
  23. support NUMA hardware which are hot pluggable as well, where physical
  24. node insertion and removal require support for CPU hotplug.
  25. Such advances require CPUs available to a kernel to be removed either for
  26. provisioning reasons, or for RAS purposes to keep an offending CPU off
  27. system execution path. Hence the need for CPU hotplug support in the
  28. Linux kernel.
  29. A more novel use of CPU-hotplug support is its use today in suspend
  30. resume support for SMP. Dual-core and HT support makes even
  31. a laptop run SMP kernels which didn't support these methods. SMP support
  32. for suspend/resume is a work in progress.
  33. General Stuff about CPU Hotplug
  34. --------------------------------
  35. Command Line Switches
  36. ---------------------
  37. maxcpus=n Restrict boot time cpus to n. Say if you have 4 cpus, using
  38. maxcpus=2 will only boot 2. You can choose to bring the
  39. other cpus later online, read FAQ's for more info.
  40. additional_cpus=n (*) Use this to limit hotpluggable cpus. This option sets
  41. cpu_possible_map = cpu_present_map + additional_cpus
  42. (*) Option valid only for following architectures
  43. - x86_64, ia64, s390
  44. ia64 and x86_64 use the number of disabled local apics in ACPI tables MADT
  45. to determine the number of potentially hot-pluggable cpus. The implementation
  46. should only rely on this to count the # of cpus, but *MUST* not rely on the
  47. apicid values in those tables for disabled apics. In the event BIOS doesn't
  48. mark such hot-pluggable cpus as disabled entries, one could use this
  49. parameter "additional_cpus=x" to represent those cpus in the cpu_possible_map.
  50. s390 uses the number of cpus it detects at IPL time to also the number of bits
  51. in cpu_possible_map. If it is desired to add additional cpus at a later time
  52. the number should be specified using this option or the possible_cpus option.
  53. possible_cpus=n [s390 only] use this to set hotpluggable cpus.
  54. This option sets possible_cpus bits in
  55. cpu_possible_map. Thus keeping the numbers of bits set
  56. constant even if the machine gets rebooted.
  57. This option overrides additional_cpus.
  58. CPU maps and such
  59. -----------------
  60. [More on cpumaps and primitive to manipulate, please check
  61. include/linux/cpumask.h that has more descriptive text.]
  62. cpu_possible_map: Bitmap of possible CPUs that can ever be available in the
  63. system. This is used to allocate some boot time memory for per_cpu variables
  64. that aren't designed to grow/shrink as CPUs are made available or removed.
  65. Once set during boot time discovery phase, the map is static, i.e no bits
  66. are added or removed anytime. Trimming it accurately for your system needs
  67. upfront can save some boot time memory. See below for how we use heuristics
  68. in x86_64 case to keep this under check.
  69. cpu_online_map: Bitmap of all CPUs currently online. Its set in __cpu_up()
  70. after a cpu is available for kernel scheduling and ready to receive
  71. interrupts from devices. Its cleared when a cpu is brought down using
  72. __cpu_disable(), before which all OS services including interrupts are
  73. migrated to another target CPU.
  74. cpu_present_map: Bitmap of CPUs currently present in the system. Not all
  75. of them may be online. When physical hotplug is processed by the relevant
  76. subsystem (e.g ACPI) can change and new bit either be added or removed
  77. from the map depending on the event is hot-add/hot-remove. There are currently
  78. no locking rules as of now. Typical usage is to init topology during boot,
  79. at which time hotplug is disabled.
  80. You really dont need to manipulate any of the system cpu maps. They should
  81. be read-only for most use. When setting up per-cpu resources almost always use
  82. cpu_possible_map/for_each_possible_cpu() to iterate.
  83. Never use anything other than cpumask_t to represent bitmap of CPUs.
  84. #include <linux/cpumask.h>
  85. for_each_possible_cpu - Iterate over cpu_possible_map
  86. for_each_online_cpu - Iterate over cpu_online_map
  87. for_each_present_cpu - Iterate over cpu_present_map
  88. for_each_cpu_mask(x,mask) - Iterate over some random collection of cpu mask.
  89. #include <linux/cpu.h>
  90. lock_cpu_hotplug() and unlock_cpu_hotplug():
  91. The above calls are used to inhibit cpu hotplug operations. While holding the
  92. cpucontrol mutex, cpu_online_map will not change. If you merely need to avoid
  93. cpus going away, you could also use preempt_disable() and preempt_enable()
  94. for those sections. Just remember the critical section cannot call any
  95. function that can sleep or schedule this process away. The preempt_disable()
  96. will work as long as stop_machine_run() is used to take a cpu down.
  97. CPU Hotplug - Frequently Asked Questions.
  98. Q: How to enable my kernel to support CPU hotplug?
  99. A: When doing make defconfig, Enable CPU hotplug support
  100. "Processor type and Features" -> Support for Hotpluggable CPUs
  101. Make sure that you have CONFIG_HOTPLUG, and CONFIG_SMP turned on as well.
  102. You would need to enable CONFIG_HOTPLUG_CPU for SMP suspend/resume support
  103. as well.
  104. Q: What architectures support CPU hotplug?
  105. A: As of 2.6.14, the following architectures support CPU hotplug.
  106. i386 (Intel), ppc, ppc64, parisc, s390, ia64 and x86_64
  107. Q: How to test if hotplug is supported on the newly built kernel?
  108. A: You should now notice an entry in sysfs.
  109. Check if sysfs is mounted, using the "mount" command. You should notice
  110. an entry as shown below in the output.
  111. ....
  112. none on /sys type sysfs (rw)
  113. ....
  114. If this is not mounted, do the following.
  115. #mkdir /sysfs
  116. #mount -t sysfs sys /sys
  117. Now you should see entries for all present cpu, the following is an example
  118. in a 8-way system.
  119. #pwd
  120. #/sys/devices/system/cpu
  121. #ls -l
  122. total 0
  123. drwxr-xr-x 10 root root 0 Sep 19 07:44 .
  124. drwxr-xr-x 13 root root 0 Sep 19 07:45 ..
  125. drwxr-xr-x 3 root root 0 Sep 19 07:44 cpu0
  126. drwxr-xr-x 3 root root 0 Sep 19 07:44 cpu1
  127. drwxr-xr-x 3 root root 0 Sep 19 07:44 cpu2
  128. drwxr-xr-x 3 root root 0 Sep 19 07:44 cpu3
  129. drwxr-xr-x 3 root root 0 Sep 19 07:44 cpu4
  130. drwxr-xr-x 3 root root 0 Sep 19 07:44 cpu5
  131. drwxr-xr-x 3 root root 0 Sep 19 07:44 cpu6
  132. drwxr-xr-x 3 root root 0 Sep 19 07:48 cpu7
  133. Under each directory you would find an "online" file which is the control
  134. file to logically online/offline a processor.
  135. Q: Does hot-add/hot-remove refer to physical add/remove of cpus?
  136. A: The usage of hot-add/remove may not be very consistently used in the code.
  137. CONFIG_HOTPLUG_CPU enables logical online/offline capability in the kernel.
  138. To support physical addition/removal, one would need some BIOS hooks and
  139. the platform should have something like an attention button in PCI hotplug.
  140. CONFIG_ACPI_HOTPLUG_CPU enables ACPI support for physical add/remove of CPUs.
  141. Q: How do i logically offline a CPU?
  142. A: Do the following.
  143. #echo 0 > /sys/devices/system/cpu/cpuX/online
  144. Once the logical offline is successful, check
  145. #cat /proc/interrupts
  146. You should now not see the CPU that you removed. Also online file will report
  147. the state as 0 when a cpu if offline and 1 when its online.
  148. #To display the current cpu state.
  149. #cat /sys/devices/system/cpu/cpuX/online
  150. Q: Why cant i remove CPU0 on some systems?
  151. A: Some architectures may have some special dependency on a certain CPU.
  152. For e.g in IA64 platforms we have ability to sent platform interrupts to the
  153. OS. a.k.a Corrected Platform Error Interrupts (CPEI). In current ACPI
  154. specifications, we didn't have a way to change the target CPU. Hence if the
  155. current ACPI version doesn't support such re-direction, we disable that CPU
  156. by making it not-removable.
  157. In such cases you will also notice that the online file is missing under cpu0.
  158. Q: How do i find out if a particular CPU is not removable?
  159. A: Depending on the implementation, some architectures may show this by the
  160. absence of the "online" file. This is done if it can be determined ahead of
  161. time that this CPU cannot be removed.
  162. In some situations, this can be a run time check, i.e if you try to remove the
  163. last CPU, this will not be permitted. You can find such failures by
  164. investigating the return value of the "echo" command.
  165. Q: What happens when a CPU is being logically offlined?
  166. A: The following happen, listed in no particular order :-)
  167. - A notification is sent to in-kernel registered modules by sending an event
  168. CPU_DOWN_PREPARE
  169. - All process is migrated away from this outgoing CPU to a new CPU
  170. - All interrupts targeted to this CPU is migrated to a new CPU
  171. - timers/bottom half/task lets are also migrated to a new CPU
  172. - Once all services are migrated, kernel calls an arch specific routine
  173. __cpu_disable() to perform arch specific cleanup.
  174. - Once this is successful, an event for successful cleanup is sent by an event
  175. CPU_DEAD.
  176. "It is expected that each service cleans up when the CPU_DOWN_PREPARE
  177. notifier is called, when CPU_DEAD is called its expected there is nothing
  178. running on behalf of this CPU that was offlined"
  179. Q: If i have some kernel code that needs to be aware of CPU arrival and
  180. departure, how to i arrange for proper notification?
  181. A: This is what you would need in your kernel code to receive notifications.
  182. #include <linux/cpu.h>
  183. static int __cpuinit foobar_cpu_callback(struct notifier_block *nfb,
  184. unsigned long action, void *hcpu)
  185. {
  186. unsigned int cpu = (unsigned long)hcpu;
  187. switch (action) {
  188. case CPU_ONLINE:
  189. foobar_online_action(cpu);
  190. break;
  191. case CPU_DEAD:
  192. foobar_dead_action(cpu);
  193. break;
  194. }
  195. return NOTIFY_OK;
  196. }
  197. static struct notifier_block __cpuinitdata foobar_cpu_notifer =
  198. {
  199. .notifier_call = foobar_cpu_callback,
  200. };
  201. You need to call register_cpu_notifier() from your init function.
  202. Init functions could be of two types:
  203. 1. early init (init function called when only the boot processor is online).
  204. 2. late init (init function called _after_ all the CPUs are online).
  205. For the first case, you should add the following to your init function
  206. register_cpu_notifier(&foobar_cpu_notifier);
  207. For the second case, you should add the following to your init function
  208. register_hotcpu_notifier(&foobar_cpu_notifier);
  209. You can fail PREPARE notifiers if something doesn't work to prepare resources.
  210. This will stop the activity and send a following CANCELED event back.
  211. CPU_DEAD should not be failed, its just a goodness indication, but bad
  212. things will happen if a notifier in path sent a BAD notify code.
  213. Q: I don't see my action being called for all CPUs already up and running?
  214. A: Yes, CPU notifiers are called only when new CPUs are on-lined or offlined.
  215. If you need to perform some action for each cpu already in the system, then
  216. for_each_online_cpu(i) {
  217. foobar_cpu_callback(&foobar_cpu_notifier, CPU_UP_PREPARE, i);
  218. foobar_cpu_callback(&foobar_cpu_notifier, CPU_ONLINE, i);
  219. }
  220. Q: If i would like to develop cpu hotplug support for a new architecture,
  221. what do i need at a minimum?
  222. A: The following are what is required for CPU hotplug infrastructure to work
  223. correctly.
  224. - Make sure you have an entry in Kconfig to enable CONFIG_HOTPLUG_CPU
  225. - __cpu_up() - Arch interface to bring up a CPU
  226. - __cpu_disable() - Arch interface to shutdown a CPU, no more interrupts
  227. can be handled by the kernel after the routine
  228. returns. Including local APIC timers etc are
  229. shutdown.
  230. - __cpu_die() - This actually supposed to ensure death of the CPU.
  231. Actually look at some example code in other arch
  232. that implement CPU hotplug. The processor is taken
  233. down from the idle() loop for that specific
  234. architecture. __cpu_die() typically waits for some
  235. per_cpu state to be set, to ensure the processor
  236. dead routine is called to be sure positively.
  237. Q: I need to ensure that a particular cpu is not removed when there is some
  238. work specific to this cpu is in progress.
  239. A: First switch the current thread context to preferred cpu
  240. int my_func_on_cpu(int cpu)
  241. {
  242. cpumask_t saved_mask, new_mask = CPU_MASK_NONE;
  243. int curr_cpu, err = 0;
  244. saved_mask = current->cpus_allowed;
  245. cpu_set(cpu, new_mask);
  246. err = set_cpus_allowed(current, new_mask);
  247. if (err)
  248. return err;
  249. /*
  250. * If we got scheduled out just after the return from
  251. * set_cpus_allowed() before running the work, this ensures
  252. * we stay locked.
  253. */
  254. curr_cpu = get_cpu();
  255. if (curr_cpu != cpu) {
  256. err = -EAGAIN;
  257. goto ret;
  258. } else {
  259. /*
  260. * Do work : But cant sleep, since get_cpu() disables preempt
  261. */
  262. }
  263. ret:
  264. put_cpu();
  265. set_cpus_allowed(current, saved_mask);
  266. return err;
  267. }
  268. Q: How do we determine how many CPUs are available for hotplug.
  269. A: There is no clear spec defined way from ACPI that can give us that
  270. information today. Based on some input from Natalie of Unisys,
  271. that the ACPI MADT (Multiple APIC Description Tables) marks those possible
  272. CPUs in a system with disabled status.
  273. Andi implemented some simple heuristics that count the number of disabled
  274. CPUs in MADT as hotpluggable CPUS. In the case there are no disabled CPUS
  275. we assume 1/2 the number of CPUs currently present can be hotplugged.
  276. Caveat: Today's ACPI MADT can only provide 256 entries since the apicid field
  277. in MADT is only 8 bits.
  278. User Space Notification
  279. Hotplug support for devices is common in Linux today. Its being used today to
  280. support automatic configuration of network, usb and pci devices. A hotplug
  281. event can be used to invoke an agent script to perform the configuration task.
  282. You can add /etc/hotplug/cpu.agent to handle hotplug notification user space
  283. scripts.
  284. #!/bin/bash
  285. # $Id: cpu.agent
  286. # Kernel hotplug params include:
  287. #ACTION=%s [online or offline]
  288. #DEVPATH=%s
  289. #
  290. cd /etc/hotplug
  291. . ./hotplug.functions
  292. case $ACTION in
  293. online)
  294. echo `date` ":cpu.agent" add cpu >> /tmp/hotplug.txt
  295. ;;
  296. offline)
  297. echo `date` ":cpu.agent" remove cpu >>/tmp/hotplug.txt
  298. ;;
  299. *)
  300. debug_mesg CPU $ACTION event not supported
  301. exit 1
  302. ;;
  303. esac