JAILHOUSE
=========

Jailhouse is a partitioning Hypervisor based on Linux. It is able to run
bare-metal applications or (adapted) operating systems besides Linux. For this
purpose it configures CPU and device virtualization features of the hardware
platform in a way that none of these domains, called "cells" here, can
interfere with each other in an unacceptable way.

Jailhouse is optimized for simplicity rather than feature richness. Unlike
full-featured Linux-based hypervisors like KVM or Xen, Jailhouse does not
support overcommitment of resources like CPUs, RAM or devices. It performs no
scheduling and only virtualizes those resources in software, that are essential
for a platform and cannot be partitioned in hardware.

Once Jailhouse is activated, it runs bare-metal, i.e. it takes full control
over the hardware and needs no external support. However, in contrast to other
bare-metal hypervisors, it is loaded and configured by a normal Linux system.
Its management interface is based on Linux infrastructure. So you boot Linux
first, then you enable Jailhouse and finally you split off parts of the
system's resources and assign them to additional cells.


WARNING: This is work in progress! Don't expect things to be complete in any
dimension. Use at your own risk. And keep the reset button in reach.


Community
---------

Project home:   https://github.com/siemens/jailhouse

git:            https://github.com/siemens/jailhouse.git
                git@github.com:siemens/jailhouse.git

Mailing list:   jailhouse-dev@googlegroups.com
 Subscription:  jailhouse-dev+subscribe@googlegroups.com
                https://groups.google.com/forum/#!forum/jailhouse-dev/join
 Archives:      http://news.gmane.org/gmane.linux.jailhouse


Requirements (preliminary)
--------------------------

currently:
 - Intel x86 processor with support for 64-bit and VMX, more precisely
    - EPT (extended page tables)
    - unrestricted guest mode
 - at least 2 logical CPUs
 - x86-64 Linux kernel (tested against >= 3.9)
    - VT-d usage has to be disabled in the Linux kernel, e.g. via command
      line parameters:

        intel_iommu=off intremap=off

upcoming:
 - Intel IOMMU with interrupt remapping support


Build
-----

Simply run make, optionally specifying the target kernel directory:

    make [KERNELDIR=/path/to/kernel/objects]

Note that the command line tool "jailhouse" requires a separate make run from
within the tools/ directory.


Configuration
-------------

Jailhouse requires one configuration file for the complete system and one for
each additional cell beside Linux. The configuration is currently being
defined manually by filling C structures. To study the structure, use
configs/qemu-vm.c for a system configuration and configs/apic-demo.c for a cell
configuration as reference. The build system will pick up every .c file from
the configs/ directory and generate a corresponding .cell file. .cell files can
then be passed to the jailhouse command line tool for enabling the hypervisor
and creating new cells.


Demonstration in QEMU/KVM
-------------------------

The included system configuration qemu-vm.c can be used to run Jailhouse in
QEMU/KVM virtual machine on Intel x86 hosts. Currently it requires kvm.git,
next branch on the host (in order to get support for guest activity state HLT).
3.14 is expected to include all necessary features for this setup. You also
need a Linux guest image with a recent kernel (tested with >= 3.9) and the
ability to build a module for this kernel. Make sure the kvm-intel module was
loaded with nested=1 to enable nested VMX support. Start the virtual machine as
follows:

    qemu-system-x86_64 LinuxInstallation.img -m 1G -enable-kvm -smp 4 \
        -cpu kvm64,-kvm_pv_eoi,-kvm_steal_time,-kvm_asyncpf,-kvmclock,+vmx,+x2apic \
        -serial stdio -serial vc

Inside the VM, make sure that jailhouse.bin, generated by the build process,
is available for firmware loading (typically /lib/firmware).

The hypervisor requires a contiguous piece of RAM for itself and each
additional cell. This currently has to be pre-allocated during boot-up. So you
need to add

    memmap=66M$0x3be00000

as parameter to the command line of the virtual machine's kernel. Reboot the
guest and load jailhouse.ko. Then enable Jailhouse like this:

    jailhouse enable /path/to/qemu-vm.cell

Next you can create a cell with a demonstration application as follows:

    jailhouse cell create /path/to/apic-demo.cell /path/to/apic-demo.bin \
        -l 0xf0000

apic-demo.bin is left by the built process in the inmates/ directory. This
application will program the APIC timer interrupt to fire at 10 Hz, measuring
the jitter against the PM timer and displaying the result on the 
console. Given that this demonstration runs in a virtual machine, obviously
no decent latencies should be expected.

To demonstrate the execution of a second non-Linux cell, issue the following
command:

    jailhouse cell create /path/to/tiny-demo.cell /path/to/tiny-demo.bin \
        -l 0xf0000

The tiny-demo will use the second serial port provided by QEMU. You will find
its output in a virtual console of the QEMU window.

Cell destruction is performed by specifying the configuration file of the
desired cell. This command will destroy the apic-demo:

    jailhouse cell destroy /path/to/apic-demo.cell

Jailhouse is can also be stopped completely again:

    jailhouse disable

All non-Linux cells running at that point will be destroyed, and resources
will be returned to Linux.