1 CPU hotplug Support in Linux(tm) Kernel 2 3 Maintainers: 4 CPU Hotplug Core: 5 Rusty Russell <rusty@rustycorp.com.au> 6 Srivatsa Vaddagiri <vatsa@in.ibm.com> 7 i386: 8 Zwane Mwaikambo <zwane@arm.linux.org.uk> 9 ppc64: 10 Nathan Lynch <nathanl@austin.ibm.com> 11 Joel Schopp <jschopp@austin.ibm.com> 12 ia64/x86_64: 13 Ashok Raj <ashok.raj@intel.com> 14 s390: 15 Heiko Carstens <heiko.carstens@de.ibm.com> 16 17Authors: Ashok Raj <ashok.raj@intel.com> 18Lots of feedback: Nathan Lynch <nathanl@austin.ibm.com>, 19 Joel Schopp <jschopp@austin.ibm.com> 20 21Introduction 22 23Modern advances in system architectures have introduced advanced error 24reporting and correction capabilities in processors. CPU architectures permit 25partitioning support, where compute resources of a single CPU could be made 26available to virtual machine environments. There are couple OEMS that 27support NUMA hardware which are hot pluggable as well, where physical 28node insertion and removal require support for CPU hotplug. 29 30Such advances require CPUs available to a kernel to be removed either for 31provisioning reasons, or for RAS purposes to keep an offending CPU off 32system execution path. Hence the need for CPU hotplug support in the 33Linux kernel. 34 35A more novel use of CPU-hotplug support is its use today in suspend 36resume support for SMP. Dual-core and HT support makes even 37a laptop run SMP kernels which didn't support these methods. SMP support 38for suspend/resume is a work in progress. 39 40General Stuff about CPU Hotplug 41-------------------------------- 42 43Command Line Switches 44--------------------- 45maxcpus=n Restrict boot time cpus to n. Say if you have 4 cpus, using 46 maxcpus=2 will only boot 2. You can choose to bring the 47 other cpus later online, read FAQ's for more info. 48 49additional_cpus=n (*) Use this to limit hotpluggable cpus. This option sets 50 cpu_possible_map = cpu_present_map + additional_cpus 51 52(*) Option valid only for following architectures 53- ia64 54 55ia64 uses the number of disabled local apics in ACPI tables MADT to 56determine the number of potentially hot-pluggable cpus. The implementation 57should only rely on this to count the # of cpus, but *MUST* not rely 58on the apicid values in those tables for disabled apics. In the event 59BIOS doesn't mark such hot-pluggable cpus as disabled entries, one could 60use this parameter "additional_cpus=x" to represent those cpus in the 61cpu_possible_map. 62 63possible_cpus=n [s390,x86_64] use this to set hotpluggable cpus. 64 This option sets possible_cpus bits in 65 cpu_possible_map. Thus keeping the numbers of bits set 66 constant even if the machine gets rebooted. 67 68CPU maps and such 69----------------- 70[More on cpumaps and primitive to manipulate, please check 71include/linux/cpumask.h that has more descriptive text.] 72 73cpu_possible_map: Bitmap of possible CPUs that can ever be available in the 74system. This is used to allocate some boot time memory for per_cpu variables 75that aren't designed to grow/shrink as CPUs are made available or removed. 76Once set during boot time discovery phase, the map is static, i.e no bits 77are added or removed anytime. Trimming it accurately for your system needs 78upfront can save some boot time memory. See below for how we use heuristics 79in x86_64 case to keep this under check. 80 81cpu_online_map: Bitmap of all CPUs currently online. Its set in __cpu_up() 82after a cpu is available for kernel scheduling and ready to receive 83interrupts from devices. Its cleared when a cpu is brought down using 84__cpu_disable(), before which all OS services including interrupts are 85migrated to another target CPU. 86 87cpu_present_map: Bitmap of CPUs currently present in the system. Not all 88of them may be online. When physical hotplug is processed by the relevant 89subsystem (e.g ACPI) can change and new bit either be added or removed 90from the map depending on the event is hot-add/hot-remove. There are currently 91no locking rules as of now. Typical usage is to init topology during boot, 92at which time hotplug is disabled. 93 94You really dont need to manipulate any of the system cpu maps. They should 95be read-only for most use. When setting up per-cpu resources almost always use 96cpu_possible_map/for_each_possible_cpu() to iterate. 97 98Never use anything other than cpumask_t to represent bitmap of CPUs. 99 100 #include <linux/cpumask.h> 101 102 for_each_possible_cpu - Iterate over cpu_possible_map 103 for_each_online_cpu - Iterate over cpu_online_map 104 for_each_present_cpu - Iterate over cpu_present_map 105 for_each_cpu_mask(x,mask) - Iterate over some random collection of cpu mask. 106 107 #include <linux/cpu.h> 108 get_online_cpus() and put_online_cpus(): 109 110The above calls are used to inhibit cpu hotplug operations. While the 111cpu_hotplug.refcount is non zero, the cpu_online_map will not change. 112If you merely need to avoid cpus going away, you could also use 113preempt_disable() and preempt_enable() for those sections. 114Just remember the critical section cannot call any 115function that can sleep or schedule this process away. The preempt_disable() 116will work as long as stop_machine_run() is used to take a cpu down. 117 118CPU Hotplug - Frequently Asked Questions. 119 120Q: How to enable my kernel to support CPU hotplug? 121A: When doing make defconfig, Enable CPU hotplug support 122 123 "Processor type and Features" -> Support for Hotpluggable CPUs 124 125Make sure that you have CONFIG_HOTPLUG, and CONFIG_SMP turned on as well. 126 127You would need to enable CONFIG_HOTPLUG_CPU for SMP suspend/resume support 128as well. 129 130Q: What architectures support CPU hotplug? 131A: As of 2.6.14, the following architectures support CPU hotplug. 132 133i386 (Intel), ppc, ppc64, parisc, s390, ia64 and x86_64 134 135Q: How to test if hotplug is supported on the newly built kernel? 136A: You should now notice an entry in sysfs. 137 138Check if sysfs is mounted, using the "mount" command. You should notice 139an entry as shown below in the output. 140 141 .... 142 none on /sys type sysfs (rw) 143 .... 144 145If this is not mounted, do the following. 146 147 #mkdir /sysfs 148 #mount -t sysfs sys /sys 149 150Now you should see entries for all present cpu, the following is an example 151in a 8-way system. 152 153 #pwd 154 #/sys/devices/system/cpu 155 #ls -l 156 total 0 157 drwxr-xr-x 10 root root 0 Sep 19 07:44 . 158 drwxr-xr-x 13 root root 0 Sep 19 07:45 .. 159 drwxr-xr-x 3 root root 0 Sep 19 07:44 cpu0 160 drwxr-xr-x 3 root root 0 Sep 19 07:44 cpu1 161 drwxr-xr-x 3 root root 0 Sep 19 07:44 cpu2 162 drwxr-xr-x 3 root root 0 Sep 19 07:44 cpu3 163 drwxr-xr-x 3 root root 0 Sep 19 07:44 cpu4 164 drwxr-xr-x 3 root root 0 Sep 19 07:44 cpu5 165 drwxr-xr-x 3 root root 0 Sep 19 07:44 cpu6 166 drwxr-xr-x 3 root root 0 Sep 19 07:48 cpu7 167 168Under each directory you would find an "online" file which is the control 169file to logically online/offline a processor. 170 171Q: Does hot-add/hot-remove refer to physical add/remove of cpus? 172A: The usage of hot-add/remove may not be very consistently used in the code. 173CONFIG_HOTPLUG_CPU enables logical online/offline capability in the kernel. 174To support physical addition/removal, one would need some BIOS hooks and 175the platform should have something like an attention button in PCI hotplug. 176CONFIG_ACPI_HOTPLUG_CPU enables ACPI support for physical add/remove of CPUs. 177 178Q: How do i logically offline a CPU? 179A: Do the following. 180 181 #echo 0 > /sys/devices/system/cpu/cpuX/online 182 183Once the logical offline is successful, check 184 185 #cat /proc/interrupts 186 187You should now not see the CPU that you removed. Also online file will report 188the state as 0 when a cpu if offline and 1 when its online. 189 190 #To display the current cpu state. 191 #cat /sys/devices/system/cpu/cpuX/online 192 193Q: Why cant i remove CPU0 on some systems? 194A: Some architectures may have some special dependency on a certain CPU. 195 196For e.g in IA64 platforms we have ability to sent platform interrupts to the 197OS. a.k.a Corrected Platform Error Interrupts (CPEI). In current ACPI 198specifications, we didn't have a way to change the target CPU. Hence if the 199current ACPI version doesn't support such re-direction, we disable that CPU 200by making it not-removable. 201 202In such cases you will also notice that the online file is missing under cpu0. 203 204Q: How do i find out if a particular CPU is not removable? 205A: Depending on the implementation, some architectures may show this by the 206absence of the "online" file. This is done if it can be determined ahead of 207time that this CPU cannot be removed. 208 209In some situations, this can be a run time check, i.e if you try to remove the 210last CPU, this will not be permitted. You can find such failures by 211investigating the return value of the "echo" command. 212 213Q: What happens when a CPU is being logically offlined? 214A: The following happen, listed in no particular order :-) 215 216- A notification is sent to in-kernel registered modules by sending an event 217 CPU_DOWN_PREPARE or CPU_DOWN_PREPARE_FROZEN, depending on whether or not the 218 CPU is being offlined while tasks are frozen due to a suspend operation in 219 progress 220- All processes are migrated away from this outgoing CPU to new CPUs. 221 The new CPU is chosen from each process' current cpuset, which may be 222 a subset of all online CPUs. 223- All interrupts targeted to this CPU is migrated to a new CPU 224- timers/bottom half/task lets are also migrated to a new CPU 225- Once all services are migrated, kernel calls an arch specific routine 226 __cpu_disable() to perform arch specific cleanup. 227- Once this is successful, an event for successful cleanup is sent by an event 228 CPU_DEAD (or CPU_DEAD_FROZEN if tasks are frozen due to a suspend while the 229 CPU is being offlined). 230 231 "It is expected that each service cleans up when the CPU_DOWN_PREPARE 232 notifier is called, when CPU_DEAD is called its expected there is nothing 233 running on behalf of this CPU that was offlined" 234 235Q: If i have some kernel code that needs to be aware of CPU arrival and 236 departure, how to i arrange for proper notification? 237A: This is what you would need in your kernel code to receive notifications. 238 239 #include <linux/cpu.h> 240 static int __cpuinit foobar_cpu_callback(struct notifier_block *nfb, 241 unsigned long action, void *hcpu) 242 { 243 unsigned int cpu = (unsigned long)hcpu; 244 245 switch (action) { 246 case CPU_ONLINE: 247 case CPU_ONLINE_FROZEN: 248 foobar_online_action(cpu); 249 break; 250 case CPU_DEAD: 251 case CPU_DEAD_FROZEN: 252 foobar_dead_action(cpu); 253 break; 254 } 255 return NOTIFY_OK; 256 } 257 258 static struct notifier_block __cpuinitdata foobar_cpu_notifer = 259 { 260 .notifier_call = foobar_cpu_callback, 261 }; 262 263You need to call register_cpu_notifier() from your init function. 264Init functions could be of two types: 2651. early init (init function called when only the boot processor is online). 2662. late init (init function called _after_ all the CPUs are online). 267 268For the first case, you should add the following to your init function 269 270 register_cpu_notifier(&foobar_cpu_notifier); 271 272For the second case, you should add the following to your init function 273 274 register_hotcpu_notifier(&foobar_cpu_notifier); 275 276You can fail PREPARE notifiers if something doesn't work to prepare resources. 277This will stop the activity and send a following CANCELED event back. 278 279CPU_DEAD should not be failed, its just a goodness indication, but bad 280things will happen if a notifier in path sent a BAD notify code. 281 282Q: I don't see my action being called for all CPUs already up and running? 283A: Yes, CPU notifiers are called only when new CPUs are on-lined or offlined. 284 If you need to perform some action for each cpu already in the system, then 285 286 for_each_online_cpu(i) { 287 foobar_cpu_callback(&foobar_cpu_notifier, CPU_UP_PREPARE, i); 288 foobar_cpu_callback(&foobar_cpu_notifier, CPU_ONLINE, i); 289 } 290 291Q: If i would like to develop cpu hotplug support for a new architecture, 292 what do i need at a minimum? 293A: The following are what is required for CPU hotplug infrastructure to work 294 correctly. 295 296 - Make sure you have an entry in Kconfig to enable CONFIG_HOTPLUG_CPU 297 - __cpu_up() - Arch interface to bring up a CPU 298 - __cpu_disable() - Arch interface to shutdown a CPU, no more interrupts 299 can be handled by the kernel after the routine 300 returns. Including local APIC timers etc are 301 shutdown. 302 - __cpu_die() - This actually supposed to ensure death of the CPU. 303 Actually look at some example code in other arch 304 that implement CPU hotplug. The processor is taken 305 down from the idle() loop for that specific 306 architecture. __cpu_die() typically waits for some 307 per_cpu state to be set, to ensure the processor 308 dead routine is called to be sure positively. 309 310Q: I need to ensure that a particular cpu is not removed when there is some 311 work specific to this cpu is in progress. 312A: First switch the current thread context to preferred cpu 313 314 int my_func_on_cpu(int cpu) 315 { 316 cpumask_t saved_mask, new_mask = CPU_MASK_NONE; 317 int curr_cpu, err = 0; 318 319 saved_mask = current->cpus_allowed; 320 cpu_set(cpu, new_mask); 321 err = set_cpus_allowed(current, new_mask); 322 323 if (err) 324 return err; 325 326 /* 327 * If we got scheduled out just after the return from 328 * set_cpus_allowed() before running the work, this ensures 329 * we stay locked. 330 */ 331 curr_cpu = get_cpu(); 332 333 if (curr_cpu != cpu) { 334 err = -EAGAIN; 335 goto ret; 336 } else { 337 /* 338 * Do work : But cant sleep, since get_cpu() disables preempt 339 */ 340 } 341 ret: 342 put_cpu(); 343 set_cpus_allowed(current, saved_mask); 344 return err; 345 } 346 347 348Q: How do we determine how many CPUs are available for hotplug. 349A: There is no clear spec defined way from ACPI that can give us that 350 information today. Based on some input from Natalie of Unisys, 351 that the ACPI MADT (Multiple APIC Description Tables) marks those possible 352 CPUs in a system with disabled status. 353 354 Andi implemented some simple heuristics that count the number of disabled 355 CPUs in MADT as hotpluggable CPUS. In the case there are no disabled CPUS 356 we assume 1/2 the number of CPUs currently present can be hotplugged. 357 358 Caveat: Today's ACPI MADT can only provide 256 entries since the apicid field 359 in MADT is only 8 bits. 360 361User Space Notification 362 363Hotplug support for devices is common in Linux today. Its being used today to 364support automatic configuration of network, usb and pci devices. A hotplug 365event can be used to invoke an agent script to perform the configuration task. 366 367You can add /etc/hotplug/cpu.agent to handle hotplug notification user space 368scripts. 369 370 #!/bin/bash 371 # $Id: cpu.agent 372 # Kernel hotplug params include: 373 #ACTION=%s [online or offline] 374 #DEVPATH=%s 375 # 376 cd /etc/hotplug 377 . ./hotplug.functions 378 379 case $ACTION in 380 online) 381 echo `date` ":cpu.agent" add cpu >> /tmp/hotplug.txt 382 ;; 383 offline) 384 echo `date` ":cpu.agent" remove cpu >>/tmp/hotplug.txt 385 ;; 386 *) 387 debug_mesg CPU $ACTION event not supported 388 exit 1 389 ;; 390 esac 391