vhost-device-gpu: Add Initial Implementation

This program is a vhost-user backend daemon that provides
VIRTIO GPU device emulation as specified in the VIRTIO Spec v.1.2
https://docs.oasis-open.org/virtio/virtio/v1.2/csd01/virtio-v1.2-csd01.html
This crate utilizes the rutabaga crate from crosvm with some
minor modification to rutabaga crate to fix compilation.
This crate depends on this PR[rust-vmm/vhost#239]
that implements support for QEMU's vhost-user-gpu protocol.

This device uses the rutabaga_gfx crate to offer two rendering backends:
1. Virglrenderer:
   - Rutabaga translates OpenGL API and Vulkan calls to an intermediate
     representation and allows for OpenGL acceleration on the host.
2. Gfxstream:
   - GLES and Vulkan calls are forwarded to the host.

These backends can be used by simply changing the `--gpu-mode` command
line option.
This crate also includes some modifications from libkrun virtio-gpu device
https://github.com/containers/libkrun/tree/main/src/devices/src/virtio/gpu

Fixes: #598

Co-authored-by: Dorinda Bassey <[email protected]>
Co-authored-by: Matej Hrica <[email protected]>

Signed-off-by: Dorinda Bassey <[email protected]>
Signed-off-by: Matej Hrica <[email protected]>

staging/Cargo.toml

@@ -1,5 +1,6 @@
 [workspace]
 resolver = "2"
 members = [
+  "vhost-device-gpu",
   "vhost-device-video",
 ]

staging/coverage_config_x86_64.json

@@ -1,5 +1,5 @@
 {
-  "coverage_score": 74.82,
+  "coverage_score": 79.82,
   "exclude_path": "",
   "crate_features": ""
 }

staging/vhost-device-gpu/CHANGELOG.md

@@ -0,0 +1,10 @@
+# Changelog
+## Unreleased
+
+### Added
+
+### Changed
+
+### Fixed
+
+### Deprecated

staging/vhost-device-gpu/Cargo.toml

@@ -0,0 +1,40 @@
+[package]
+name = "vhost-device-gpu"
+version = "0.1.0"
+authors = ["Dorinda Bassey <[email protected]>", "Matej Hrica <[email protected]>"]
+description = "A virtio-gpu device using the vhost-user protocol."
+repository = "https://github.com/rust-vmm/vhost-device"
+readme = "README.md"
+keywords = ["gpu", "vhost", "vhost-user", "virtio"]
+categories = ["multimedia::video", "virtualization"]
+license = "Apache-2.0 OR BSD-3-Clause"
+edition = "2021"
+publish = false
+
+[features]
+xen = ["vm-memory/xen", "vhost/xen", "vhost-user-backend/xen"]
+
+[dependencies]
+clap = { version = "4.4", features = ["derive"] }
+env_logger = "0.11.5"
+libc = "0.2"
+log = "0.4"
+
+[target.'cfg(not(target_env = "musl"))'.dependencies]
+rutabaga_gfx = { version = "0.1.4", features = ["gfxstream", "virgl_renderer"] }
+thiserror = "1.0"
+vhost = { version = "0.12.1", features = ["vhost-user-backend"] }
+vhost-user-backend = "0.16.1"
+virtio-bindings = "0.2.2"
+virtio-queue = "0.13.0"
+vm-memory = "0.15.0"
+vmm-sys-util = "0.12.1"
+zerocopy = "0.6.3"
+
+[dev-dependencies]
+assert_matches = "1.5"
+tempfile = "3.13"
+virtio-queue = { version = "0.13", features = ["test-utils"] }
+vm-memory = { version = "0.15.0", features = ["backend-mmap", "backend-atomic"] }
+mockall = "0.13.0"
+rusty-fork = "0.3.0"

staging/vhost-device-gpu/LICENSE-APACHE

@@ -0,0 +1 @@
+../../LICENSE-APACHE

staging/vhost-device-gpu/LICENSE-BSD-3-Clause

@@ -0,0 +1 @@
+../../LICENSE-BSD-3-Clause

staging/vhost-device-gpu/README.md

@@ -0,0 +1,101 @@
+# vhost-device-gpu - GPU emulation backend daemon
+
+## Synopsis
+```shell
+vhost-device-gpu --socket-path <SOCKET>
+```
+
+## Description
+A virtio-gpu device using the vhost-user protocol.
+
+## Options
+
+```text
+       -s, --socket-path <SOCKET>
+              vhost-user Unix domain socket path
+
+       -h, --help
+              Print help
+
+       -V, --version
+              Print version
+```
+
+## Limitations
+
+We are currently only supporting sharing the display output to QEMU through a
+socket using the transfer_read operation triggered by
+VIRTIO_GPU_CMD_TRANSFER_FROM_HOST_3D to transfer data from and to virtio-gpu 3d
+resources. It'll be nice to have support for directly sharing display output
+resource using dmabuf.
+
+This device does not yet support the VIRTIO_GPU_CMD_RESOURCE_CREATE_BLOB,
+VIRTIO_GPU_CMD_SET_SCANOUT_BLOB and VIRTIO_GPU_CMD_RESOURCE_ASSIGN_UUID features.
+
+Currently this crate requires some necessary bits in order to move the crate out of staging:
+
+- Achieving a minimum of ~87% code coverage in the main vhost-device repository,
+  which requires some additional unit tests to increase code coverage.
+- Addition of CLI arguments to specify the exact number of capsets and use
+  a default capset configuration when no capset is specified rather than using
+  hard-coded capset value.
+
+## Features
+
+The device leverages the [rutabaga_gfx](https://crates.io/crates/rutabaga_gfx) crate
+to provide virglrenderer and gfxstream rendering. With Virglrenderer, Rutabaga
+translates OpenGL API and Vulkan calls to an intermediate representation and allows
+for OpenGL acceleration on the host. With the gfxstream rendering mode, GLES and
+Vulkan calls are forwarded to the host with minimal modification.
+
+## Examples
+
+First start the daemon on the host machine using either of the 2 gpu modes:
+
+1) virgl-renderer
+2) gfxstream
+
+```shell
+host# vhost-device-gpu --socket-path /tmp/gpu.socket --gpu-mode virgl-renderer
+```
+
+With QEMU, there are two device front-ends you can use with this device.
+You can either use `vhost-user-gpu-pci` or `vhost-user-vga`, which also
+implements VGA, that allows you to see boot messages before the guest
+initializes the GPU. You can also use different display outputs (for example
+`gtk` or `dbus`).
+By default, QEMU also adds another VGA output, use `-vga none` to make 
+sure it is disabled.
+
+1) Using `vhost-user-gpu-pci`
+
+Start QEMU with the following flags:
+
+```text
+-chardev socket,id=vgpu,path=/tmp/gpu.socket \
+-device vhost-user-gpu-pci,chardev=vgpu,id=vgpu \
+-object memory-backend-memfd,share=on,id=mem0,size=4G, \
+-machine q35,memory-backend=mem0,accel=kvm \
+-display gtk,gl=on,show-cursor=on \
+-vga none
+```
+
+2) Using `vhost-user-vga`
+
+Start QEMU with the following flags:
+
+```text
+-chardev socket,id=vgpu,path=/tmp/gpu.socket \
+-device vhost-user-vga,chardev=vgpu,id=vgpu \
+-object memory-backend-memfd,share=on,id=mem0,size=4G, \
+-machine q35,memory-backend=mem0,accel=kvm \
+-display gtk,gl=on,show-cursor=on \
+-vga none
+```
+
+## License
+
+This project is licensed under either of
+
+- [Apache License](http://www.apache.org/licenses/LICENSE-2.0), Version 2.0
+- [BSD-3-Clause License](https://opensource.org/licenses/BSD-3-Clause)

staging/vhost-device-gpu/rustfmt.toml

@@ -0,0 +1,7 @@
+edition = "2018"
+format_generated_files = false
+format_code_in_doc_comments = true
+format_strings = true
+imports_granularity = "Crate"
+group_imports = "StdExternalCrate"
+wrap_comments = true