cloud-hypervisor/docs/device_model.md

218 lines
8.7 KiB
Markdown
Raw Normal View History

# Device Model
This document describes the device model supported by `cloud-hypervisor`.
## Summary
| Device | Build configurable | Enabled by default | Runtime configurable |
| :----: | :----: | :----: | :----: |
| Serial port | :x: | :x: | :heavy_check_mark: |
| RTC/CMOS | :heavy_check_mark: | :heavy_check_mark: | :x: |
| I/O APIC | :x: | :x: | :heavy_check_mark: |
| i8042 shutdown/reboot | :x: | :x: | :x: |
| ACPI shutdown/reboot | :x: | :heavy_check_mark: | :x: |
| virtio-blk | :x: | :x: | :heavy_check_mark: |
| virtio-console | :x: | :x: | :heavy_check_mark: |
| virtio-iommu | :x: | :x: | :heavy_check_mark: |
| virtio-net | :x: | :x: | :heavy_check_mark: |
| virtio-pmem | :x: | :x: | :heavy_check_mark: |
| virtio-rng | :x: | :x: | :heavy_check_mark: |
| virtio-vsock | :x: | :x: | :heavy_check_mark: |
| vhost-user-blk | :x: | :x: | :heavy_check_mark: |
| vhost-user-fs | :x: | :x: | :heavy_check_mark: |
| vhost-user-net | :x: | :x: | :heavy_check_mark: |
| VFIO | :heavy_check_mark: | :x: | :heavy_check_mark: |
## Legacy devices
### Serial port
Simple emulation of a serial port by reading and writing to specific port I/O
addresses. The serial port can be very useful to gather early logs from the
operating system booted inside the VM.
For x86_64, The default serial port is from an emulated 16550A device. It can
be used as the default console for Linux when booting with the option
`console=ttyS0`. For AArch64, the default serial port is from an emulated
PL011 UART device. The related command line for AArch64 is `console=ttyAMA0`.
This device is always built-in, and it is disabled by default. It can be
enabled with the `--serial` option, as long as its parameter is not `off`.
### RTC/CMOS
For environments such as Windows or EFI which cannot rely on KVM clock, the
emulation of this legacy device makes the platform usable.
This device is built-in by default, but it can be compiled out with Rust
features. When compiled in, it is always enabled, and cannot be disabled
from the command line.
For AArch64 machines, an ARM PrimeCell Real Time Clock(PL031) is implemented.
This device is built-in by default for the AArch64 platform, and it is always
enabled, and cannot be disabled from the command line.
### I/O APIC
`cloud-hypervisor` supports a so-called split IRQ chip implementation by
implementing support for the [IOAPIC](https://wiki.osdev.org/IOAPIC).
By moving part of the IRQ chip implementation from kernel space to user space,
the IRQ chip emulation does not always run in a fully privileged mode.
The device is always built-in, and it is enabled depending on the presence of
the serial port. If the serial port is disabled, and because no other device
would require pin based interrupts (INTx), the I/O APIC is disabled.
### i8042
Simplified PS/2 port since it supports only one key to trigger a reboot or
shutdown, depending on the ACPI support.
This device is always built-in, but it is disabled by default. Because ACPI is
enabled by default, the handling of reboot/shutdown goes through the dedicated
ACPI device. In case ACPI is disabled, this device is enabled to bring to the
VM some reboot/shutdown support.
### ARM PrimeCell General Purpose Input/Output (PL061)
Simplified ARM PrimeCell GPIO (PL061) implementation. Only supports key 3 to
trigger a graceful shutdown of the AArch64 guest.
### ACPI device
This is a dedicated device for handling ACPI shutdown and reboot when ACPI is
enabled.
This device is always built-in, and it is enabled by default since the ACPI
feature is enabled by default.
## Virtio devices
For all virtio devices listed below, only `virtio-pci` transport layer is
supported. Cloud Hypervisor supports multiple PCI segments, and users can
append `,pci_segment=<PCI_segment_number>` to the device flag in the Cloud
Hypervisor command line to assign devices to a specific PCI segment.
### virtio-block
The `virtio-blk` device exposes a block device to the guest. This device is
usually used to boot the operating system running in the VM.
This device is always built-in, and it is enabled based on the presence of the
flag `--disk`.
### virtio-console
`cloud-hypervisor` exposes a `virtio-console` device to the guest. Although
using this device as a guest console can potentially cut some early boot
messages, it can reduce the guest boot time and provides a complete console
implementation.
This device is always built-in, and it is enabled by default to provide a guest
console. It can be disabled, switching back to the legacy serial port by
selecting `--serial tty --console off` from the command line.
### virtio-iommu
As we want to improve our nested guests support, we added support for exposing
a [paravirtualized IOMMU](iommu.md) device through virtio. This allows for a
safer nested virtio and directly assigned devices support.
This device is always built-in, and it is enabled based on the presence of the
parameter `iommu=on` in any of the virtio or VFIO devices. If at least one of
these devices needs to be connected to the paravirtualized IOMMU, the
`virtio-iommu` device will be created.
### virtio-net
The `virtio-net` device provides network connectivity for the guest, as it
creates a network interface connected to a TAP interface automatically created
by the `cloud-hypervisor` on the host.
This device is always built-in, and it is enabled based on the presence of the
flag `--net`.
### virtio-pmem
The `virtio-pmem` implementation emulates a virtual persistent memory device
that `cloud-hypervisor` can e.g. boot from. Booting from a `virtio-pmem` device
allows to bypass the guest page cache and improve the guest memory footprint.
This device is always built-in, and it is enabled based on the presence of the
flag `--pmem`.
### virtio-rng
A VM does not generate entropy like a real machine would, which is an issue
when workloads running in the guest need random numbers to be generated. The
`virtio-rng` device provides entropy to the guest by relying on the generator
that can be found on the host. By default, the chosen source of entropy is
`/dev/urandom`.
This device is always built-in, and it is always enabled. The `--rng` flag can
be used to change the source of entropy.
### virtio-vsock
In order to more efficiently and securely communicate between host and guest,
we added a hybrid implementation of the [VSOCK](http://man7.org/linux/man-pages/man7/vsock.7.html)
socket address family over virtio.
Credits go to the [Firecracker](https://github.com/firecracker-microvm/firecracker/blob/master/docs/vsock.md)
project as our implementation is a copy of theirs.
This device is always built-in, and it is enabled based on the presence of the
flag `--vsock`.
## Vhost-user devices
Vhost-user devices are virtio backends running outside of the VMM, as its own
separate process. They are usually used to bring more flexibility and increased
isolation.
### vhost-user-blk
As part of the general effort to offload paravirtualized I/O to external
processes, we added support for vhost-user-blk backends. This enables
`cloud-hypervisor` users to plug a `vhost-user` based block device (e.g. SPDK)
into the VMM as their virtio block backend.
This device is always built-in, and it is enabled when `vhost_user=true` and
`socket` are provided to the `--disk` parameter.
### vhost-user-fs
`cloud-hypervisor` supports the [virtio-fs](https://virtio-fs.gitlab.io/)
shared file system, allowing for an efficient and reliable way of sharing
a filesystem between the host and the cloud-hypervisor guest.
See our [filesystem sharing](fs.md) documentation for more details on how to
use virtio-fs with cloud-hypervisor.
This device is always built-in, and it is enabled based on the presence of the
flag `--fs`.
### vhost-user-net
As part of the general effort to offload paravirtualized I/O to external
processes, we added support for [vhost-user-net](https://access.redhat.com/solutions/3394851)
backends. This enables `cloud-hypervisor` users to plug a `vhost-user` based
networking device (e.g. DPDK) into the VMM as their virtio network backend.
This device is always built-in, and it is enabled when `vhost_user=true` and
`socket` are provided to the `--net` parameter.
## VFIO
VFIO (Virtual Function I/O) is a kernel framework that exposes direct device
access to userspace. `cloud-hypervisor` uses VFIO to directly assign host
physical devices into its guest.
See our [VFIO documentation](vfio.md) for more details on how to directly
assign host devices to `cloud-hypervisor` guests.
Because VFIO implies `vfio-pci` in the `cloud-hypervisor` context, the VFIO
support is built-in when the `pci` feature is selected. And because the `pci`
feature is built-in by default, VFIO support is also built-in by default.
When VFIO support is built-in, a physical device can be passed through, using
the flag `--device` in order to enable the VFIO code.