An Nvidia GPU device can be passed to a Kata Containers container using GPU passthrough (Nvidia GPU pass-through mode) as well as GPU mediated passthrough (Nvidia vGPU mode).
Nvidia GPU pass-through mode, an entire physical GPU is directly assigned to one VM, bypassing the Nvidia Virtual GPU Manager. In this mode of operation, the GPU is accessed exclusively by the Nvidia driver running in the VM to which it is assigned. The GPU is not shared among VMs.
Nvidia Virtual GPU (vGPU) enables multiple virtual machines (VMs) to have simultaneous, direct access to a single physical GPU, using the same Nvidia graphics drivers that are deployed on non-virtualized operating systems. By doing this, Nvidia vGPU provides VMs with unparalleled graphics performance, compute performance, and application compatibility, together with the cost-effectiveness and scalability brought about by sharing a GPU among multiple workloads.
Technology | Description | Behaviour | Detail |
---|---|---|---|
Nvidia GPU pass-through mode | GPU passthrough | Physical GPU assigned to a single VM | Direct GPU assignment to VM without limitation |
Nvidia vGPU mode | GPU sharing | Physical GPU shared by multiple VMs | Mediated passthrough |
Nvidia GPUs Recommended for Virtualization:
- Nvidia Tesla (T4, M10, P6, V100 or newer)
- Nvidia Quadro RTX 6000/8000
Some hardware requires a larger PCI BARs window, for example, Nvidia Tesla P100, K40m
$ lspci -s 04:00.0 -vv | grep Region
Region 0: Memory at c6000000 (32-bit, non-prefetchable) [size=16M]
Region 1: Memory at 383800000000 (64-bit, prefetchable) [size=16G] #above 4G
Region 3: Memory at 383c00000000 (64-bit, prefetchable) [size=32M]
For large BARs devices, MMIO mapping above 4G address space should be enabled
in the PCI configuration of the BIOS.
Some hardware vendors use different name in BIOS, such as:
- Above 4G Decoding
- Memory Hole for PCI MMIO
- Memory Mapped I/O above 4GB
The following steps outline the workflow for using an Nvidia GPU with Kata.
The following configurations need to be enabled on your host kernel:
CONFIG_VFIO
CONFIG_VFIO_IOMMU_TYPE1
CONFIG_VFIO_MDEV
CONFIG_VFIO_MDEV_DEVICE
CONFIG_VFIO_PCI
Your host kernel needs to be booted with intel_iommu=on
on the kernel command line.
To use non-large BARs devices (for example, Nvidia Tesla T4), you need Kata version 1.3.0 or above. Follow the Kata Containers setup instructions to install the latest version of Kata.
The following configuration in the Kata configuration.toml
file as shown below can work:
machine_type = "pc"
hotplug_vfio_on_root_bus = true
To use large BARs devices (for example, Nvidia Tesla P100), you need Kata version 1.11.0 or above.
The following configuration in the Kata configuration.toml
file as shown below can work:
Hotplug for PCI devices by shpchp
(Linux's SHPC PCI Hotplug driver):
machine_type = "q35"
hotplug_vfio_on_root_bus = false
Hotplug for PCIe devices by pciehp
(Linux's PCIe Hotplug driver):
machine_type = "q35"
hotplug_vfio_on_root_bus = true
pcie_root_port = 1
The default guest kernel installed with Kata Containers does not provide GPU support. To use an Nvidia GPU with Kata Containers, you need to build a kernel with the necessary GPU support.
The following kernel config options need to be enabled:
# Support PCI/PCIe device hotplug (Required for large BARs device)
CONFIG_HOTPLUG_PCI_PCIE=y
CONFIG_HOTPLUG_PCI_SHPC=y
# Support for loading modules (Required for load Nvidia drivers)
CONFIG_MODULES=y
CONFIG_MODULE_UNLOAD=y
# Enable the MMIO access method for PCIe devices (Required for large BARs device)
CONFIG_PCI_MMCONFIG=y
The following kernel config options need to be disabled:
# Disable Open Source Nvidia driver nouveau
# It conflicts with Nvidia official driver
CONFIG_DRM_NOUVEAU=n
Note:
CONFIG_DRM_NOUVEAU
is normally disabled by default. It is worth checking that it is not enabled in your kernel configuration to prevent any conflicts.
Build the Kata Containers kernel with the previous config options, using the instructions described in Building Kata Containers kernel. For further details on building and installing guest kernels, see the developer guide.
There is an easy way to build a guest kernel that supports Nvidia GPU:
## Build guest kernel with https://github.com/kata-containers/packaging/tree/master/kernel
# Prepare (download guest kernel source, generate .config)
$ ./build-kernel.sh -v 4.19.86 -g nvidia -f setup
# Build guest kernel
$ ./build-kernel.sh -v 4.19.86 -g nvidia build
# Install guest kernel
$ sudo -E ./build-kernel.sh -v 4.19.86 -g nvidia install
/usr/share/kata-containers/vmlinux-nvidia-gpu.container -> vmlinux-4.19.86-70-nvidia-gpu
/usr/share/kata-containers/vmlinuz-nvidia-gpu.container -> vmlinuz-4.19.86-70-nvidia-gpu
To build Nvidia Driver in Kata container, kernel-devel
is required.
This is a way to generate rpm packages for kernel-devel
:
$ cd kata-linux-4.19.86-68
$ make rpm-pkg
Output RPMs:
~/rpmbuild/RPMS/x86_64/kernel-devel-4.19.86_nvidia_gpu-1.x86_64.rpm
Note:
kernel-devel
should be installed in Kata container before run Nvidia driver installer.- Run
make deb-pkg
to build the deb package.
Before using the new guest kernel, please update the kernel
parameters in configuration.toml
.
kernel = "/usr/share/kata-containers/vmlinuz-nvidia-gpu.container"
Use the following steps to pass an Nvidia GPU device in pass-through mode with Kata:
-
Find the Bus-Device-Function (BDF) for GPU device on host:
$ sudo lspci -nn -D | grep -i nvidia 0000:04:00.0 3D controller [0302]: NVIDIA Corporation Device [10de:15f8] (rev a1) 0000:84:00.0 3D controller [0302]: NVIDIA Corporation Device [10de:15f8] (rev a1)
PCI address
0000:04:00.0
is assigned to the hardware GPU device.10de:15f8
is the device ID of the hardware GPU device. -
Find the IOMMU group for the GPU device:
$ BDF="0000:04:00.0" $ readlink -e /sys/bus/pci/devices/$BDF/iommu_group /sys/kernel/iommu_groups/45
The previous output shows that the GPU belongs to IOMMU group 45.
-
Check the IOMMU group number under
/dev/vfio
:$ ls -l /dev/vfio total 0 crw------- 1 root root 248, 0 Feb 28 09:57 45 crw------- 1 root root 248, 1 Feb 28 09:57 54 crw-rw-rw- 1 root root 10, 196 Feb 28 09:57 vfio
-
Start a Kata container with GPU device:
$ sudo docker run -it --runtime=kata-runtime --cap-add=ALL --device /dev/vfio/45 centos /bin/bash
-
Run
lspci
within the container to verify the GPU device is seen in the list of the PCI devices. Note the vendor-device id of the GPU (10de:15f8
) in thelspci
output.$ lspci -nn -D | grep '10de:15f8' 0000:01:01.0 3D controller [0302]: NVIDIA Corporation GP100GL [Tesla P100 PCIe 16GB] [10de:15f8] (rev a1)
-
Additionally, you can check the PCI BARs space of the Nvidia GPU device in the container:
$ lspci -s 01:01.0 -vv | grep Region Region 0: Memory at c0000000 (32-bit, non-prefetchable) [disabled] [size=16M] Region 1: Memory at 4400000000 (64-bit, prefetchable) [disabled] [size=16G] Region 3: Memory at 4800000000 (64-bit, prefetchable) [disabled] [size=32M]
Note: If you see a message similar to the above, the BAR space of the Nvidia GPU has been successfully allocated.
Nvidia vGPU is a licensed product on all supported GPU boards. A software license is required to enable all vGPU features within the guest VM.
Note: There is no suitable test environment, so it is not written here.
Download the official Nvidia driver from
https://www.nvidia.com/Download/index.aspx,
for example NVIDIA-Linux-x86_64-418.87.01.run
.
Install the kernel-devel
(generated in the previous steps) for guest kernel:
$ sudo rpm -ivh kernel-devel-4.19.86_gpu-1.x86_64.rpm
Here is an example to extract, compile and install Nvidia driver:
## Extract
$ sh ./NVIDIA-Linux-x86_64-418.87.01.run -x
## Compile and install (It will take some time)
$ cd NVIDIA-Linux-x86_64-418.87.01
$ sudo ./nvidia-installer -a -q --ui=none \
--no-cc-version-check \
--no-opengl-files --no-install-libglvnd \
--kernel-source-path=/usr/src/kernels/`uname -r`
Or just run one command line:
$ sudo sh ./NVIDIA-Linux-x86_64-418.87.01.run -a -q --ui=none \
--no-cc-version-check \
--no-opengl-files --no-install-libglvnd \
--kernel-source-path=/usr/src/kernels/`uname -r`
To view detailed logs of the installer:
$ tail -f /var/log/nvidia-installer.log
Load Nvidia driver module manually
# Optional(generate modules.dep and map files for Nvidia driver)
$ sudo depmod
# Load module
$ sudo modprobe nvidia-drm
# Check module
$ lsmod | grep nvidia
nvidia_drm 45056 0
nvidia_modeset 1093632 1 nvidia_drm
nvidia 18202624 1 nvidia_modeset
drm_kms_helper 159744 1 nvidia_drm
drm 364544 3 nvidia_drm,drm_kms_helper
i2c_core 65536 3 nvidia,drm_kms_helper,drm
ipmi_msghandler 49152 1 nvidia
Check Nvidia device status with nvidia-smi
$ nvidia-smi
Tue Mar 3 00:03:49 2020
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 418.87.01 Driver Version: 418.87.01 CUDA Version: 10.1 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 Tesla P100-PCIE... Off | 00000000:01:01.0 Off | 0 |
| N/A 27C P0 25W / 250W | 0MiB / 16280MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| No running processes found |
+-----------------------------------------------------------------------------+