-
Notifications
You must be signed in to change notification settings - Fork 61
WIP: Add rhel9 base for vGPU-manager containers #486
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Open
mvalsecchi-nv
wants to merge
3
commits into
NVIDIA:main
Choose a base branch
from
mvalsecchi-nv:issue/453
base: main
Could not load branches
Branch not found: {{ refName }}
Loading
Could not load tags
Nothing to show
Loading
Are you sure you want to change the base?
Some commits from the old base branch may be removed from the timeline,
and old review comments may become outdated.
Open
Changes from all commits
Commits
Show all changes
3 commits
Select commit
Hold shift + click to select a range
File filter
Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,34 @@ | ||
| FROM nvcr.io/nvidia/cuda:13.0.1-base-ubi9 | ||
|
|
||
| ARG DRIVER_VERSION | ||
| ENV DRIVER_VERSION=$DRIVER_VERSION | ||
| ARG DRIVER_ARCH=x86_64 | ||
| ENV DRIVER_ARCH=$DRIVER_ARCH | ||
|
|
||
| RUN mkdir -p /driver | ||
| WORKDIR /driver | ||
| COPY NVIDIA-Linux-${DRIVER_ARCH}-${DRIVER_VERSION}-vgpu-kvm.run . | ||
| RUN chmod +x NVIDIA-Linux-${DRIVER_ARCH}-${DRIVER_VERSION}-vgpu-kvm.run | ||
|
|
||
| COPY nvidia-driver /usr/local/bin | ||
| COPY ocp_dtk_entrypoint /usr/local/bin | ||
|
|
||
| LABEL io.k8s.display-name="NVIDIA vGPU Manager Container" | ||
| LABEL name="NVIDIA vGPU Manager Container" | ||
| LABEL vendor="NVIDIA" | ||
| LABEL version="${DRIVER_VERSION}" | ||
| LABEL release="N/A" | ||
| LABEL summary="Provision the NVIDIA vGPU Manager through containers" | ||
| LABEL description="See summary" | ||
|
|
||
| # Install / upgrade packages here that are required to resolve CVEs | ||
| ARG CVE_UPDATES | ||
| RUN if [ -n "${CVE_UPDATES}" ]; then \ | ||
| yum update -y ${CVE_UPDATES} && \ | ||
| rm -rf /var/cache/yum/*; \ | ||
| fi | ||
|
|
||
| # Add NGC DL license from the CUDA image | ||
| RUN mkdir /licenses && mv /NGC-DL-CONTAINER-LICENSE /licenses/NGC-DL-CONTAINER-LICENSE | ||
|
|
||
| ENTRYPOINT ["nvidia-driver", "init"] | ||
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,252 @@ | ||
| #!/bin/bash | ||
| # Copyright (c) 2022, NVIDIA CORPORATION. All rights reserved. | ||
|
|
||
| set -xe | ||
|
|
||
| DRIVER_VERSION=${DRIVER_VERSION:?"Missing driver version"} | ||
| DRIVER_RESET_RETRIES=10 | ||
| DELAY_BEFORE_VF_CREATION=${DELAY_BEFORE_VF_CREATION:-15} | ||
| RUN_DIR=/run/nvidia | ||
|
|
||
| # Mount the driver rootfs into the run directory with the exception of sysfs. | ||
| _mount_rootfs() { | ||
| echo "Mounting NVIDIA driver rootfs..." | ||
| mount --make-runbindable /sys | ||
| mount --make-private /sys | ||
| mkdir -p ${RUN_DIR}/driver | ||
| mount --rbind / ${RUN_DIR}/driver | ||
|
|
||
| echo "Change device files security context for selinux compatibility" | ||
| chcon -R -t container_file_t ${RUN_DIR}/driver/dev | ||
| } | ||
|
|
||
| # Unmount the driver rootfs from the run directory. | ||
| _unmount_rootfs() { | ||
| echo "Unmounting NVIDIA driver rootfs..." | ||
| if findmnt -r -o TARGET | grep "${RUN_DIR}/driver" > /dev/null; then | ||
| umount -l -R ${RUN_DIR}/driver | ||
| fi | ||
| } | ||
|
|
||
| # Create /dev/char directory if it doesn't exist inside the container. | ||
| # Without this directory, nvidia-vgpu-mgr will fail to create symlinks | ||
| # under /dev/char for new devices nodes. | ||
| _create_dev_char_directory() { | ||
| if [ ! -d "/dev/char" ]; then | ||
| echo "Creating '/dev/char' directory" | ||
| mkdir -p /dev/char | ||
| fi | ||
| } | ||
|
|
||
| _set_fw_search_path() { | ||
| local nv_fw_search_path="$RUN_DIR/driver/lib/firmware" | ||
| local fw_path_config_file="/sys/module/firmware_class/parameters/path" | ||
|
|
||
| if [[ ! -z $(grep '[^[:space:]]' $fw_path_config_file) ]]; then | ||
| echo "WARNING: A search path is already configured in $fw_path_config_file" | ||
| echo " Retaining the current configuration. Note, GSP firmware may not be found and thus won't be used by the NVIDIA driver." | ||
| return | ||
| fi | ||
|
|
||
| echo "Configuring the following firmware search path in '$fw_path_config_file': $nv_fw_search_path" | ||
| echo -n "$nv_fw_search_path" > $fw_path_config_file | ||
| } | ||
|
|
||
| _install_driver() { | ||
| local tmp_dir=$(mktemp -d) | ||
|
|
||
| sh NVIDIA-Linux-${DRIVER_ARCH}-${DRIVER_VERSION}-vgpu-kvm.run --ui=none --no-questions --tmpdir ${tmp_dir} --no-systemd | ||
| } | ||
|
|
||
| # Currently _install_driver() takes care of loading nvidia modules. Just need to start necessary vgpu daemons | ||
| _load_driver() { | ||
| /usr/bin/nvidia-vgpud | ||
| /usr/bin/nvidia-vgpu-mgr & | ||
|
|
||
| # check nvidia drivers are loaded | ||
| if [ ! -f /sys/module/nvidia_vgpu_vfio/refcnt ] || [ ! -f /sys/module/nvidia/refcnt ]; then | ||
| echo "Failed to load nvidia driver" | ||
| return 1 | ||
| fi | ||
| return 0 | ||
| } | ||
|
|
||
| # Enable virtual functions for all physical GPUs on the node that support SR-IOV. | ||
| # Retry logic is to account for when the driver is busy (i.e. during driver initialization) | ||
| _enable_vfs() { | ||
| # Wait before attempting to create VFs to ensure the driver has finished initializing. | ||
| # This is a WAR for a bug in vGPU 17.2 where sriov-manage does not return a non-zero | ||
| # exit code even though VF creation fails. | ||
| sleep $DELAY_BEFORE_VF_CREATION | ||
|
|
||
| local retry | ||
| for ((retry = 0 ; retry <= $DRIVER_RESET_RETRIES ; retry++)); do | ||
| if /usr/lib/nvidia/sriov-manage -e ALL; then | ||
| return 0 | ||
| fi | ||
| if [ $retry == $DRIVER_RESET_RETRIES ]; then | ||
| echo "Failed to enable VFs" | ||
| fi | ||
| done | ||
| return 1 | ||
| } | ||
|
|
||
| # Disable virtual functions for all physical GPUs on the node that support SR-IOV. | ||
| # Retry logic is to account for when the driver is busy (i.e. during driver initialization) | ||
| _disable_vfs() { | ||
| local retry | ||
| for ((retry = 0 ; retry <= $DRIVER_RESET_RETRIES ; retry++)); do | ||
| if /usr/lib/nvidia/sriov-manage -d ALL; then | ||
| return 0 | ||
| fi | ||
| if [ $retry == $DRIVER_RESET_RETRIES ]; then | ||
| echo "Failed to disable VFs" | ||
| fi | ||
| done | ||
| return 1 | ||
| } | ||
|
|
||
| _unload_driver() { | ||
| local rmmod_args=() | ||
| local nvidia_deps=0 | ||
| local nvidia_refs=0 | ||
| local nvidia_vgpu_vfio_refs=0 | ||
|
|
||
| if [ -f /var/run/nvidia-vgpu-mgr/nvidia-vgpu-mgr.pid ]; then | ||
| echo "Stopping NVIDIA vGPU Manager..." | ||
| local pid=$(< /var/run/nvidia-vgpu-mgr/nvidia-vgpu-mgr.pid) | ||
|
|
||
| kill -TERM "${pid}" | ||
| for i in $(seq 1 50); do | ||
| kill -0 "${pid}" 2> /dev/null || break | ||
| sleep 0.1 | ||
| done | ||
| if [ $i -eq 50 ]; then | ||
| echo "Could not stop NVIDIA vGPU Manager" >&2 | ||
| return 1 | ||
| fi | ||
| fi | ||
|
|
||
| echo "Unloading NVIDIA driver kernel modules..." | ||
| if [ -f /sys/module/nvidia_vgpu_vfio/refcnt ]; then | ||
| nvidia_vgpu_vfio_refs=$(< /sys/module/nvidia_vgpu_vfio/refcnt) | ||
| rmmod_args+=("nvidia_vgpu_vfio") | ||
| ((++nvidia_deps)) | ||
| fi | ||
| if [ -f /sys/module/nvidia/refcnt ]; then | ||
| nvidia_refs=$(< /sys/module/nvidia/refcnt) | ||
| rmmod_args+=("nvidia") | ||
| fi | ||
|
|
||
| # TODO: check if nvidia module is in use by checking refcnt | ||
|
|
||
| if [ ${#rmmod_args[@]} -gt 0 ]; then | ||
| rmmod ${rmmod_args[@]} | ||
| if [ "$?" != "0" ]; then | ||
| return 1 | ||
| fi | ||
| fi | ||
| return 0 | ||
| } | ||
|
|
||
| _shutdown() { | ||
| if _disable_vfs && _unload_driver; then | ||
| _unmount_rootfs | ||
| return 0 | ||
| fi | ||
| echo "Failed to cleanup driver" | ||
| return 1 | ||
| } | ||
|
|
||
| build() { | ||
| echo "build() not implemented" | ||
| } | ||
|
|
||
| load() { | ||
| echo "load() not implemented" | ||
| } | ||
|
|
||
| update() { | ||
| echo "update() not implemented" | ||
| } | ||
|
|
||
| init() { | ||
| trap "echo 'Caught signal'; exit 1" HUP INT QUIT PIPE TERM | ||
| trap "_shutdown" EXIT | ||
|
|
||
| if ! _unload_driver; then | ||
| echo "Previous NVIDIA driver installation cannot be removed. Exiting" | ||
| exit 1 | ||
| fi | ||
| _unmount_rootfs | ||
| _create_dev_char_directory | ||
| _set_fw_search_path | ||
| _install_driver | ||
| _load_driver || exit 1 | ||
| _mount_rootfs | ||
| _enable_vfs | ||
|
|
||
| # In certain scenarios, /sys/class/mdev_bus is not populated with the correct list of devices (PFs and possible VFs) at this point. | ||
| # Re-run nvdidia-vgpud to ensure /sys/class/mdev_bus is populated correctly. And restart nvidia-vgpu-mgr if previously killed. | ||
| nvidia-vgpud & | ||
| pgrep nvidia-vgpu-mgr >/dev/null || (echo "Restarting nvidia-vgpu-mgr after previously killed" && nvidia-vgpu-mgr &) | ||
|
|
||
| set +x | ||
| echo "Done, now waiting for signal" | ||
| trap "echo 'Caught signal'; _shutdown; trap - EXIT; exit" HUP INT QUIT PIPE TERM | ||
|
|
||
| while true; do | ||
| sleep 15 | ||
| pgrep nvidia-vgpu-mgr >/dev/null || (echo "ERROR: nvidia-vgpu-mgr daemon is no longer running. Exiting." && exit 1) | ||
| done | ||
| } | ||
|
|
||
|
|
||
| usage() { | ||
| cat >&2 <<EOF | ||
| Usage: $0 COMMAND [ARG...] | ||
|
|
||
| Commands: | ||
| init [-a | --accept-license] | ||
| build [-a | --accept-license] | ||
| load | ||
| update [-k | --kernel VERSION] [-s | --sign KEYID] [-t | --tag TAG] | ||
| EOF | ||
| exit 1 | ||
| } | ||
|
|
||
| if [ $# -eq 0 ]; then | ||
| usage | ||
| fi | ||
| command=$1; shift | ||
| case "${command}" in | ||
| init) options=$(getopt -l accept-license -o a -- "$@") ;; | ||
| build) options=$(getopt -l accept-license,tag: -o a:t -- "$@") ;; | ||
| load) options="" ;; | ||
| update) options=$(getopt -l kernel:,sign:,tag: -o k:s:t: -- "$@") ;; | ||
| *) usage ;; | ||
| esac | ||
| if [ $? -ne 0 ]; then | ||
| usage | ||
| fi | ||
| eval set -- "${options}" | ||
|
|
||
| ACCEPT_LICENSE="" | ||
| KERNEL_VERSION=$(uname -r) | ||
| PRIVATE_KEY="" | ||
| PACKAGE_TAG="" | ||
|
|
||
| for opt in ${options}; do | ||
| case "$opt" in | ||
| -a | --accept-license) ACCEPT_LICENSE="yes"; shift 1 ;; | ||
| -k | --kernel) KERNEL_VERSION=$2; shift 2 ;; | ||
| -s | --sign) PRIVATE_KEY=$2; shift 2 ;; | ||
| -t | --tag) PACKAGE_TAG=$2; shift 2 ;; | ||
| --) shift; break ;; | ||
| esac | ||
| done | ||
| if [ $# -ne 0 ]; then | ||
| usage | ||
| fi | ||
|
|
||
| $command |
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Uh oh!
There was an error while loading. Please reload this page.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Unfortunately symlinks (from
vgpu-manager/rhel8tovgpu-manager/rhel9) would not cut it, as we pass the subdir, making those files (insiderhel8unreachable from any other sibling folder).Let me see if I can come up with a cleaner way, rather than duplicating all the files in
vgpu-manager/rhel*There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It seems most directories do indeed have copies of the same files, so I'll leave the duplicates inside
vpu-manager/rhel*folders instead of refactoring.