Merge branch 'use_upstream_cc_tools' into 'main'

Use upstream gpu_cc_tool from nvtrust

See merge request nvidia/cloud-native/k8s-cc-manager!2

Author: shivamerla

deployments/container/Dockerfile.ubi8

@@ -29,7 +29,6 @@ ARG GIT_COMMIT="unknown"
 
 RUN go build -o /artifacts/k8s-cc-manager ./cmd
 RUN cp ./scripts/cc-manager.sh /artifacts/cc-manager.sh
-#RUN cp ./scripts/nvidia-gpu-tools.py /artifacts/nvidia-gpu-tools.py
 
 RUN OS_ARCH=${TARGETARCH/x86_64/amd64} && OS_ARCH=${OS_ARCH/aarch64/arm64} && curl -o /artifacts/kubectl -L "https://dl.k8s.io/release/$(curl -L -s https://dl.k8s.io/release/stable.txt)/bin/linux/${OS_ARCH}/kubectl";
 RUN chmod a+x /artifacts/kubectl
@@ -50,10 +49,13 @@ LABEL description="See summary"
 RUN mkdir /licenses && mv /NGC-DL-CONTAINER-LICENSE /licenses/NGC-DL-CONTAINER-LICENSE
 
 COPY --from=build /artifacts/k8s-cc-manager /usr/bin/k8s-cc-manager
-COPY --from=build /artifacts/cc-manager.sh /usr/bin/cc-manager.sh
-#COPY --from=build /artifacts/nvidia-gpu-tools.py /usr/bin/nvidia-gpu-tools.py
 COPY --from=build /artifacts/kubectl /usr/bin/kubectl
 
+# Setup CC tools and licenses
+COPY --from=build /artifacts/cc-manager.sh /usr/bin/cc-manager.sh
+RUN curl -o /usr/bin/gpu_cc_tool.py -L https://raw.githubusercontent.com/NVIDIA/nvtrust/main/host_tools/python/gpu_cc_tool.py && chmod a+x /usr/bin/gpu_cc_tool.py
+RUN curl -o /licenses/LICENSE.txt -L https://raw.githubusercontent.com/NVIDIA/nvtrust/main/host_tools/python/LICENSE.txt
+
 # Install / upgrade packages here that are required to resolve CVEs
 ARG CVE_UPDATES
 RUN if [ -n "${CVE_UPDATES}" ]; then \

scripts/cc-manager.sh

@@ -28,7 +28,7 @@ _populate_cc_capable_device_ids() {
 
 _reset_gpu_after_cc_mode() {
     local gpu=$1
-    python3 /usr/bin/nvidia-gpu-tools.py --reset-after-cc-mode-switch --gpu-bdf=$gpu
+    python3 /usr/bin/gpu_cc_tool.py --reset-after-cc-mode-switch --gpu-bdf=$gpu
     if [ $? -ne 0 ]; then
         echo "unable to reset gpu $gpu for cc mode switch, output"
         return 1
@@ -149,7 +149,7 @@ _assert_gpu_cc_mode() {
     local gpu=$1
     local mode=$2
 
-    output=$(python3 /usr/bin/nvidia-gpu-tools.py --query-cc-mode --gpu-bdf=$gpu 2>&1)
+    output=$(python3 /usr/bin/gpu_cc_tool.py --query-cc-mode --gpu-bdf=$gpu 2>&1)
     if [ $? -ne 0 ]; then
         _exit_failed
     fi
@@ -386,7 +386,7 @@ set_gpu_cc_mode() {
     local mode=$CC_MODE
 
     if ! _assert_gpu_cc_mode $gpu $mode; then
-        output=$(python3 /usr/bin/nvidia-gpu-tools.py --set-cc-mode=$mode --reset-after-cc-mode-switch --gpu-bdf=$gpu 2>&1)
+        output=$(python3 /usr/bin/gpu_cc_tool.py --set-cc-mode=$mode --reset-after-cc-mode-switch --gpu-bdf=$gpu 2>&1)
         if [ $? -ne 0 ]; then
             echo "unable to set cc mode of gpu $gpu to $mode, output $output"
             return 1
@@ -434,7 +434,7 @@ get_cc_mode() {
 
 get_gpu_cc_mode() {
     local gpu=$1
-    output=$(python3 /usr/bin/nvidia-gpu-tools.py --query-cc-mode --gpu-bdf=$gpu 2>&1)
+    output=$(python3 /usr/bin/gpu_cc_tool.py --query-cc-mode --gpu-bdf=$gpu 2>&1)
     if [ $? -ne 0 ]; then
         echo "unable to get cc mode of gpu $gpu, output $output"
         return 1