Guest

The vmm app enables booting a guest operating system using the Zircon hypervisor. The hypervisor and VMM are collectively referred to as “Machina”.

These instructions will guide you through creating minimal Zircon and Linux guests. For instructions on building a more comprehensive Linux guest system see the debian_guest package.

These instructions assume a general familiarity with how to netboot the target device.

Build host system with the guest package

Configure, build, and boot the guest package as follows:

fx set core.${ARCH} --with-base //src/virtualization
fx build

Where ${ARCH} is one of x64 or arm64.

Note for external developers

(Googlers: You don't need to do this, the Linux images are downloaded from CIPD by jiri.)

The linux_guest package expects the Linux kernel binaries to be in prebuilt/virtualization/packages/linux_guest. You should create them before running fx build by running the following scripts:

./src/virtualization/packages/linux_guest/mklinux.sh \
  -l /tmp/linux/source \
  -o prebuilt/virtualization/packages/linux_guest/images/${ARCH}/Image \
  -b machina-4.18 \
  ${ARCH}
./src/virtualization/packages/linux_guest/mksysroot.sh \
  -o  prebuilt/virtualization/packages/linux_guest/images/${ARCH}/disk.img \
  -d /tmp/toybox \
  -s /tmp/dash \
  -u \
  ${ARCH}

Note: -b specifies the branch of zircon_guest to use. You can modify this value if you need a different version or omit it to use a local version.

Running guests

After netbooting the target device, to run Zircon:

guest launch zircon_guest

Likewise, to launch a Linux guest:

guest launch linux_guest

Running on QEMU

Running a guest on QEMU on x64 requires kvm (i.e. pass -k to fx qemu):

fx qemu -k

You may also need to enable nested KVM on your host machine. The following instructions assume a Linux host machine with an Intel processor.

To check whether nested virtualization is enabled, run the following command:

cat /sys/module/kvm_intel/parameters/nested

An output of Y indicates nested virtualization is enabled, 0 or N indicates not enabled.

To enable nested virtualization until the next reboot:

modprobe -r kvm_intel
modprobe kvm_intel nested=1

To make the change permanent add the following line to /etc/modprobe.d/kvm.conf:

options kvm_intel nested=1

Running an arm64 guest on QEMU requires either using GICv2 (pass -G 2).

fx qemu -G 2

Or using a more recent version of QEMU (try 2.12.0). Older versions of QEMU do not correctly emulate GICv3 when running with multiple guest VCPUs.

fx qemu -q /path/to/recent/qemu/aarch64-softmmu
...
guest launch (linux_guest|zircon_guest)

Running from Workstation

To run from Workstation, configure the guest package as follows:

fx set workstation.x64 --with-base //src/virtualization

After booting the guest packages can be launched from the system launcher as linux_guest and zircon_guest.

Integration tests

Machina has a set of integration tests that launch Zircon and Debian guests to test the VMM, hypervisor, and each of the virtio devices. To run the tests:

fx set core.${ARCH} --with-base //src/virtualization
fx run-test guest_integration_tests

Guest Configuration

Guest systems can be configured by including a config file inside the guest package:

{
    "type": "object",
    "properties": {
        "kernel": {
            "type": "string"
        },
        "ramdisk": {
            "type": "string"
        },
        "block": {
            "type": "string"
        },
        "cmdline": {
            "type": "string"
        }
    }
}

Monitor Guest exit statistics

“kstats -v” can print the number of Guest exits per second and the reason. For arm64:

cpu vm_entry vm_exit inst_abt data_abt wfx_inst sys_inst smc_inst ints 0 43 43 0 6 27 1 0 9 1 226 225 0 111 3 17 0 94 2 109 109 0 60 8 7 0 35 3 58 58 0 21 12 2 0 23

Fields:

• inst_abt: The amount of instruction abort exit.
• data_abt: The amount of data abort exit.
• wfx_inst: The amount of instruction wfe/wfi exit.
• sys_inst: The amount of systen register access exit.
• smc_inst: The amount of instruction smc exit.
• ints : The amount of interrupt exit.

For x86_64: cpu vm_entry vm_exit ints ints_win ept ctrl_reg msr(rd wr) inst(io hlt cpuid ple vmcall xsetbv) 0 40 40 10 0 0 0 10 10 5 5 0 0 0 0

Fields:

• ints : The amount of interrupt exit.
• ints_win : The amount of interrupt window exit.
• ept : The amount of EPT violation exit.
• ctrl_reg : The amount of control register(CRx) access exit.
• msr : The amount of MSR register access exit (rd/wr is read/write).
• inst : the amount of some kinds of instruction exit.