On Mon, 8 Aug 2016 10:28:02 +0800 Dou Liyang <douly.f...@cn.fujitsu.com> wrote:
> This document describes how to use cpu hotplug in QEMU. > > Signed-off-by: Dou Liyang <douly.f...@cn.fujitsu.com> > --- > docs/cpu-hotplug.txt | 110 > +++++++++++++++++++++++++++++++++++++++++++++++++++ > 1 file changed, 110 insertions(+) > create mode 100644 docs/cpu-hotplug.txt > > diff --git a/docs/cpu-hotplug.txt b/docs/cpu-hotplug.txt > new file mode 100644 > index 0000000..d62638e > --- /dev/null > +++ b/docs/cpu-hotplug.txt > @@ -0,0 +1,110 @@ > +QEMU CPU hotplug > +=================== > + > +This document explains how to use the cpu hotplug feature in QEMU, > +which is present since v2.6.0. -device/device_add based CPUs are merged since 2.7 > + > +Guest support is required for cpu hotplug to work. > + > +CPU hotplug > +------------------- > + > +In order to be able to hotplug cpu, QEMU has to be told what is the > +maximum amount of cpus the guest can grow. This is done at startup > +time by means of the -smp command-line option, which has the following > +format: > + > + -smp [cpus=]n[,maxcpus=cpus][,cores=cores][,threads=threads] > + [,sockets=sockets] > + > +Where, > + > + - "cpus" set the number of CPUs to 'n' [default=1] > + - "maxcpus" maximum number of total cpus, including offlineCPUs for > + hotplug, etc s/total// s/offlineCPUs/offline VCPUs/ or threads > + - "cores" number of CPU cores on one socket > + - "threads= number of threads on one CPU core > + - "sockets= number of discrete sockets in the system Above needs additional meaning clarification for SPAPR, CCing David > + > + > +For example, the following command-line: > + > + qemu [...] -smp 3,maxcpus=10,sockets=2,cores=2,threads=2 > + > +Creates a guest with 3 cpus and it support up to 10 cpus. The cpu > +topology is sockets (2) * cores (2) * threads (2) and can't greater > +than maxcpus. When the guest is just booted, the guest will see 3 > +cpus. so there are seven cpus can be hotplugged by using any > +combination of the available sockets,cores and threads topology or > +using apic-id. s/cpus/vcpus/ drop any mentioning of apic-id, it's not supposed to be used by user and socket/core/thread-ids should be used i.e. properties advertised by QEMU via QMP command query-hotpluggable-cpus or corresponding HMP command 'info hotpluggable-cpus' PS: Eduardo should we drop apic-id property before 2.7 is released? So people won't even try to use it, it wan never a public property before as cpus weren't available with -device/device_add. > +cpu hot-plug > +----------------------- > + > +A monitor commands are used to hotplug cpu: > + > + - "device_add": creates a cpu device and inserts it into the > + specific topology as a device > + > +For example, the following commands add a cpu which id is cpu1 to > +the guest discussed earlier: > + > + (qemu) device_add qemu64-x86_64-cpu,id=cpu1,apic-id=3 likewise remove any mentioning of apic-id from document I'd add as the first section here commands that query possible available cpus including topology properties. > + > + - "qemu64-x86_64-cpu" is the cpu modle. > + - "id" is the unique identifier in the device sets. > + - "apic-id" is the hotpluged cpu's physical identification. > + > +Another command uses the cpu topology to add the additional cpu in > +the designated position. > + > +For example, the following commands add a cpu in the last position > +of the guest cpu topology discussed earlier. > + > + (qemu) device_add qemu64-x86_64-cpu,id=cpu1,socket-id=2,core-id=1, > + thread-id=1 > + > +It's also possible to start a guest with cpu cold-plugged into the > +hotpluggable cpu topology. s/hotpluggable cpu topology/specific place (socket,core,thread)/ > + > +In the following command-line example, a guest which has 3 cpus is > +created where one of the cpus comes from the "apic-id", and another > +one comes from "socket-id...". After that, the guest has additional > +seven cpus to be hot-plug when needed: > + > + qemu [...] -smp 1,maxcpus=10,sockets=2,cores=2,threads=2 > + -device qemu64-x86_64-cpu,id=cpu1,apic-id=1 > + -device qemu64-x86_64-cpu,socket-id=2,core-id=1,thread-id=0 > + > +cpu hot-unplug > +------------------------ > + > +In order to be able to hot unplug cpu device, QEMU has two ways > +to remove cpu device. > + 1. Using the ids which were assigned when you hot plugged cpus. > + 2. Using qom_path where the cpu is located in the guest. I wouldn't advertise #2 as it's possible to unplug CPUs created with -smp X with its help and then have broken migration device_del qom_path should even be able to remove a device without ID, it should allow removal of devices below /machine/peripheral only instead of current anywhere. Daniel, is it a bug in current device_del impl.? i.e. device_del /machine/unattached/device[2] is currently allowed while I think it shouldn't. > + > +A monitor commands are used to hot unplug cpus: > + > + - "device_del": deletes a cpu device > + > +For example, assuming that the cpu device with id "cpu1" exists, > +the following commands tries to remove it. > + > + (qemu) device_del cpu1 > + > +If you don't set the ids when you hot plugged cpus. > + > +First, you may need to obtain the cpu's qom_path. The following > +commands list all hotplugable-cpus, you can get the qom_path > +through the cpu topology. > + > + (qemu) info hotpluggable-cpus > + > +Then you can remove it by the qom_path. > + > +For example, assuming that the cpu device with qom_path "/machine > +/unattached/device[0]" exists, the following commands tries to > +remove it. > + > + (qemu) device_del /machine/unattached/device[0] I'd remove qom-path related parts and leave only ID based one.
