fix: DTK shared directory path mismatch due to NFD label sanitization

The DTK flow was failing with path mismatches because the volume mount
path is created by network-operator using NFD labels, which sanitize
kernel version strings by replacing non-alphanumeric characters (except
-._) with underscores.

Example kernel: 5.14.0-570.60.1.el9_6.aarch64+64k
NFD label: feature.node.kubernetes.io/kernel-version.full=5.14.0-570.60.1.el9_6.aarch64_64k

This caused:
- Volume mount path: /mnt/shared-doca-driver-toolkit/5.14.0-570.60.1.el9_6.aarch64_64k/
- Container using: /mnt/shared-doca-driver-toolkit/5.14.0-570.60.1.el9_6.aarch64+64k/

The mismatch led to timeout errors where one container waited for
dtk_start_compile flag in the sanitized path while the other container
created flags in the unsanitized path.

Fix: Apply NFD-compatible sanitization to kernel version when constructing
DTK_OCP_NIC_SHARED_DIR path in both entrypoint.sh and dtk_nic_driver_build.sh.
The FULL_KVER variable remains unchanged for package operations that require
the actual kernel version format.

This matches the sanitization logic in node-feature-discovery:
https://github.com/kubernetes-sigs/node-feature-discovery/blob/master/source/kernel/version.go#L38-L42

Applies to RT kernels (+rt) and 64k hugepage kernels (+64k) on RHEL/RHCOS.

Signed-off-by: Fred Rolland <[email protected]>

Author: rollandf

dtk_nic_driver_build.sh

@@ -4,7 +4,10 @@
 : ${ENTRYPOINT_DEBUG:=false}
 : ${DTK_OCP_NIC_SHARED_DIR:=/mnt/shared-nvidia-nic-driver-toolkit}
 
-DTK_OCP_NIC_SHARED_DIR=$DTK_OCP_NIC_SHARED_DIR/$(uname -r)
+# Sanitize kernel version to match Kubernetes NFD label format used by network-operator for volume paths
+# NFD replaces all non-alphanumeric characters (except -._) with underscore, then trims leading/trailing -._
+DTK_KVER=$(uname -r | sed 's/[^-A-Za-z0-9_.]/_/g' | sed 's/^[-_.]*//;s/[-_.]*$//')
+DTK_OCP_NIC_SHARED_DIR=$DTK_OCP_NIC_SHARED_DIR/$DTK_KVER
 DTK_OCP_START_COMPILE_FLAG=""
 DTK_OCP_DONE_COMPILE_FLAG=""
 DTK_OCP_COMPILED_DRIVER_VER=""

entrypoint.sh

@@ -1360,7 +1360,10 @@ IS_OS_SLES=true; [[ "$(grep -i sles /etc/os-release -c)" == "0" ]] && IS_OS_SLES
 RHEL_MAJOR_VERSION=0
 OPENSHIFT_VERSION=""
 
-DTK_OCP_NIC_SHARED_DIR=${DTK_OCP_NIC_SHARED_DIR}/${FULL_KVER}
+# Sanitize kernel version to match Kubernetes NFD label format used by network-operator for volume paths
+# NFD replaces all non-alphanumeric characters (except -._) with underscore, then trims leading/trailing -._
+DTK_KVER=$(echo "${FULL_KVER}" | sed 's/[^-A-Za-z0-9_.]/_/g' | sed 's/^[-_.]*//;s/[-_.]*$//')
+DTK_OCP_NIC_SHARED_DIR=${DTK_OCP_NIC_SHARED_DIR}/${DTK_KVER}
 DTK_OCP_BUILD_SCRIPT="/root/dtk_nic_driver_build.sh"
 DTK_OCP_START_COMPILE_FLAG=${DTK_OCP_NIC_SHARED_DIR}/dtk_start_compile
 DTK_OCP_DONE_COMPILE_FLAG_PREFIX=${DTK_OCP_NIC_SHARED_DIR}/dtk_done_compile_