mirror of
https://github.com/cloud-hypervisor/cloud-hypervisor.git
synced 2024-12-22 13:45:20 +00:00
bb6c93d5df
Signed-off-by: Henry Wang <Henry.Wang@arm.com>
218 lines
9.5 KiB
Markdown
218 lines
9.5 KiB
Markdown
# Device Model
|
|
|
|
This document describes the device model supported by `cloud-hypervisor`.
|
|
|
|
## Summary
|
|
|
|
| Device | Build configurable | Enabled by default | Runtime configurable |
|
|
| :----: | :----: | :----: | :----: |
|
|
| Serial port | :negative_squared_cross_mark: | :negative_squared_cross_mark: | :heavy_check_mark: |
|
|
| RTC/CMOS | :heavy_check_mark: | :heavy_check_mark: | :negative_squared_cross_mark: |
|
|
| I/O APIC | :negative_squared_cross_mark: | :negative_squared_cross_mark: | :heavy_check_mark: |
|
|
| i8042 shutdown/reboot | :negative_squared_cross_mark: | :negative_squared_cross_mark: | :negative_squared_cross_mark: |
|
|
| ACPI shutdown/reboot | :negative_squared_cross_mark: | :heavy_check_mark: | :negative_squared_cross_mark: |
|
|
| virtio-blk | :negative_squared_cross_mark: | :negative_squared_cross_mark: | :heavy_check_mark: |
|
|
| virtio-console | :negative_squared_cross_mark: | :negative_squared_cross_mark: | :heavy_check_mark: |
|
|
| virtio-iommu | :negative_squared_cross_mark: | :negative_squared_cross_mark: | :heavy_check_mark: |
|
|
| virtio-net | :negative_squared_cross_mark: | :negative_squared_cross_mark: | :heavy_check_mark: |
|
|
| virtio-pmem | :negative_squared_cross_mark: | :negative_squared_cross_mark: | :heavy_check_mark: |
|
|
| virtio-rng | :negative_squared_cross_mark: | :negative_squared_cross_mark: | :heavy_check_mark: |
|
|
| virtio-vsock | :negative_squared_cross_mark: | :negative_squared_cross_mark: | :heavy_check_mark: |
|
|
| vhost-user-blk | :negative_squared_cross_mark: | :negative_squared_cross_mark: | :heavy_check_mark: |
|
|
| vhost-user-fs | :negative_squared_cross_mark: | :negative_squared_cross_mark: | :heavy_check_mark: |
|
|
| vhost-user-net | :negative_squared_cross_mark: | :negative_squared_cross_mark: | :heavy_check_mark: |
|
|
| VFIO | :heavy_check_mark: | :negative_squared_cross_mark: | :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.
|