A micro hypervisor for RISC-V systems.

git submodule update --init
bazel build //:salus-all

Salus:

• bazelisk see https://github.com/bazelbuild/bazelisk
  • bazelisk will install the proper version of bazel
  • bazel will install all the proper toolchains

QEMU:

• Out-of-tree patches are required; see table below.
• Install libslirp-dev for QEMU to build SLIRP network stack
• Build using QEMU instructions with --target-list=riscv64-softmmu
• Set the QEMU= variable to point to the compiled QEMU tree when using the run_* scripts described below.

Linux kernel:

• Out-of-tree patches are required; see table below.
• Build: ARCH=riscv CROSS_COMPILE=riscv64-unknown-linux-gnu- make defconfig Image
• Set the LINUX= variable to point to the compiled Linux kernel tree when using the linux related run_* scripts described below.

Buildroot:

• Out-of-tree patches are required; see table below.
• Build: make qemu_riscv64_virt_defconfig && make
• Set the BUILDROOT= variable to point to the buildroot source directory while running run_buildroot.sh script described below.

Debian:

• Download and extract a pre-baked riscv64-virt image from https://people.debian.org/~gio/dqib/.
• Set the DEBIAN= variable to point to the extracted archive when using the run_debian.sh script described below.

Latest known-working branches:

What were make targets in the Make/Cargo build are now shell scripts.

From the top level directory, run

scripts/run_tellus.sh

Many of the variable can be overwritten using environment variables on the command line. For example, to use a different version of qemu and 3 cores, you can do the following:

QEMU=/scratch/qemu-salus NCPU=3 scripts/run_tellus.sh

All the other make targets to run salus with linux work analogously.

The scripts/run_linux.sh script will boot a bare Linux kernel as the host VM that will panic upon reaching init due to the lack of a root filesystem.

To boot a more functional Linux VM, use the scripts/run_debian.sh script which will boot a Debian VM with emulated storage and network devices using pre-baked Debian initrd and rootfs images.

Example:

  QEMU=<path-to-qemu-directory> \
  LINUX=<path-to-linux-tree> \
  DEBIAN=<path-to-pre-baked-image> \
  scripts/run_debian.sh

To boot a quick functional Linux VM with busybox based rootfs built from buildroot, use the scripts/run_buildroot.sh script. The above buildroot tree must be compiled to generate the rootfs with networking enabled.

Example:

    QEMU=<path-to-qemu-directory> \
    LINUX=<path-to-linux-tree> \
    BUILDROOT=<path-to-buildroot repo>
    scripts/run_buildroot.sh

Once booted, the VM can be SSH'ed into with root:root at localhost:7722.

Additional emulated devices may be added with the EXTRA_QEMU_ARGS Makefile variable. Note that only PCI devices using MSI/MSI-X will be usable by the VM. virtio-pci devices may also be used with iommu_platform=on,disable-legacy=on flags.

Example:

   EXTRA_QEMU_ARGS="-device virtio-net-pci,iommu_platform=on,disable-legacy=on" \
   ... \
   scripts/run_debian.sh

A pair of test VMs are located in test-workloads.

tellus is a target build with bazel build //test-workloads:tellus_guestvm_rule that runs in VS mode and provides the ability to send test API calls to salus running in HS mode.

guestvm is a test confidential guest. It is started by tellus and used for testing the guest side of the TSM API.

Once it has been build, you can use the command below to run it.

    QEMU=<path-to-qemu-directory> \
    scripts/run_tellus.sh

This will boot salus, tellus, and the guestvm using the specified QEMU.

One important difference between Bazel and Cargo is in the handling of crate dependencies. If you change a dependency, Cargo will pick it up automatically. But with Bazel, you must sync the changes. There is a script provided to help you do that. To repin the changes, you can just run scripts/repin.sh.

  +---U-mode--+ +-----VS-mode-----+ +-VS-mode-+
  |           | |                 | |         |
  |           | | +---VU-mode---+ | |         |
  |   Salus   | | | VMM(crosvm) | | |  Guest  |
  | Delegated | | +-------------+ | |         |
  |   Tasks   | |                 | |         |
  |           | |    Host(linux)  | |         |
  +-----------+ +-----------------+ +---------+
        |                |               |
   TBD syscall   SBI (COVH-API)    SBI(COVG-API)
        |                |               |
  +-------------HS-mode-----------------------+
  |       Salus                               |
  +-------------------------------------------+
                         |
                        SBI
                         |
  +----------M-mode---------------------------+
  |       Firmware(OpenSBI)                   |
  +-------------------------------------------+

Normally Linux, this is the primary operating system for the device running in VS mode.

Responsibilities:

• Scheduling
• Memory allocation (except memory kept by firmware and salus at boot)
• Guest VM start/stop/scheduling via COVH-API provided by salus
• Device drivers and delegation

The virtual machine manager that runs in userspace of the host.

• qemu/kvm or crosvm
• configures memory and devices for guests
• runs any virtualized or para virtualized devices
• runs guests with vcpu_run.

VS-mode operating systems started by the host.

• Can run confidential or shared workloads.
• Uses memory shared from or donated by the host
• scheduled by the host
• can start sub-guests
• Confidential guests use COVG-API for salus/host services

The code in this repository. An HS-mode hypervisor.

• starts the host and guests
• manages stage-2 translations and IOMMU configuration for guest isolation
• delegates some tasks such as attestation to u-mode helpers
• measured by the trusted firmware/RoT

M-mode code.

OpenSBI currently boots salus from the memory (0x80200000) where qemu loader loaded it and passes the device tree to Salus.

The above instructions use OpenSBI inbuilt in Qemu. If OpenSBI needs to be built from scratch, fw_dynamic should be used for -bios argument in the qemu commandline.

Salus is able to detect if the CPU supports the vector extension. The same binary will run on processors with or without the extension, and will enable vector code if it is present.