Fabric Manager shared-nvswitch virtualization model support

Makefile

@@ -30,7 +30,7 @@ MAKE_TARGETS := build check fmt test check-vendor update-pcidb clean $(CHECK_TAR
 TARGETS := $(MAKE_TARGETS)
 
 DOCKER_TARGETS := $(patsubst %,docker-%, $(TARGETS))
-.PHONY: $(TARGETS) $(DOCKER_TARGETS)
+.PHONY: $(TARGETS) $(DOCKER_TARGETS) docker-test
 
 GOOS ?= linux
 
@@ -91,3 +91,12 @@ update-pcidb:
 
 clean:
 	rm -rf nvidia-kubevirt-gpu-device-plugin && rm -rf coverage.out
+
+docker-test:
+	$(DOCKER) build --target builder \
+		--build-arg GOLANG_VERSION="$(GOLANG_VERSION)" \
+		--build-arg DRIVER_VERSION="$(DRIVER_VERSION)" \
+		-t kubevirt-gpu-test \
+		-f deployments/container/Dockerfile.distroless .
+	$(DOCKER) run --rm kubevirt-gpu-test \
+		bash -c "CGO_ENABLED=1 CGO_CPPFLAGS='-I/usr/include' CGO_LDFLAGS='-L/usr/lib -lnvfm' go test -tags=nvfm -coverprofile=coverage.out.with-mocks $(MODULE)/..."

deployments/container/Dockerfile.distroless

@@ -28,9 +28,19 @@
 FROM nvcr.io/nvidia/cuda:13.1.1-base-ubi9 AS builder
 
 ARG GOLANG_VERSION
+ARG DRIVER_VERSION
 
 RUN yum install -y wget tar gzip make gcc glibc-devel
 
+# Install Fabric Manager SDK for NVLink partition support
+# Setup NVIDIA network repository and install FM devel package
+RUN dnf config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel9/x86_64/cuda-rhel9.repo \
+  && dnf clean expire-cache \
+  && DRIVER_STREAM=$(echo "${DRIVER_VERSION}" | cut -d. -f1) \
+  && dnf module enable -y nvidia-driver:${DRIVER_STREAM}-open/default \
+  && dnf install -y nvidia-fabric-manager-devel-${DRIVER_VERSION} \
+  && dnf clean all
+
 RUN set -eux; \
     \
     arch="$(uname -m)"; \
@@ -58,7 +68,9 @@ RUN go mod download
 COPY . .
 
 RUN CGO_ENABLED=1 \
-    go build -trimpath -o nvidia-kubevirt-gpu-device-plugin ./cmd
+  CGO_CPPFLAGS="-I/usr/include" \
+  CGO_LDFLAGS="-L/usr/lib -lnvfm" \
+  go build -trimpath -tags=nvfm -o nvidia-kubevirt-gpu-device-plugin ./cmd
 
 FROM nvcr.io/nvidia/distroless/go:v4.0.2
 
@@ -74,6 +86,9 @@ LABEL description="See summary"
 
 COPY --from=builder /workspace/nvidia-kubevirt-gpu-device-plugin /usr/bin/
 COPY --from=builder /workspace/utils/pci.ids /usr/pci.ids
+COPY --from=builder /usr/lib64/libnvfm* /usr/lib64/
+
+ENV LD_LIBRARY_PATH=/usr/lib64:$LD_LIBRARY_PATH
 
 USER 0:0
 

deployments/container/Makefile

@@ -25,7 +25,7 @@ endif
 
 GOPROXY ?= https://proxy.golang.org,direct
 
-IMAGE_VERSION := $(VERSION)
+IMAGE_VERSION := $(VERSION)-$(DRIVER_VERSION)
 
 IMAGE = $(IMAGE_NAME):$(IMAGE_VERSION)
 
@@ -63,6 +63,7 @@ $(BUILD_TARGETS): build-%:
 		--build-arg VERSION="$(VERSION)" \
 		--build-arg GIT_COMMIT="$(GIT_COMMIT)" \
 		--build-arg GOPROXY="$(GOPROXY)" \
+		--build-arg DRIVER_VERSION="$(DRIVER_VERSION)" \
 		--file $(DOCKERFILE) \
 		$(CURDIR)
 ifeq ($(PUSH_ON_BUILD),true)

manifests/nvidia-kubevirt-gpu-device-plugin.yaml

@@ -21,6 +21,13 @@ spec:
       containers:
       - name: nvidia-kubevirt-gpu-dp-ctr
         image: nvcr.io/nvidia/kubevirt-gpu-device-plugin:v1.5.0
+        env:
+        # Fabric Manager Integration (Tech Preview)
+        # Set to "true" to enable fabric manager partition coordination
+        # Requires fabric manager daemon running on the node
+        # Defaults to "false" - device plugin works normally without it
+        - name: ENABLE_FABRIC_MANAGER
+          value: "false"
         securityContext:
           allowPrivilegeEscalation: false
           capabilities: