Merge branch 'gdrcopy' into 'main'

Add support for building GDRcopy driver in the OCP DTK sidecar container

See merge request nvidia/container-images/driver!290

rhel8/common.sh

@@ -3,6 +3,7 @@
 
 GPU_DIRECT_RDMA_ENABLED="${GPU_DIRECT_RDMA_ENABLED:-false}"
 GDS_ENABLED="${GDS_ENABLED:-false}"
+GDRCOPY_ENABLED="${GDRCOPY_ENABLED:-false}"
 
 # Check if mellanox devices are present
 _mellanox_devices_present() {
@@ -32,7 +33,15 @@ _gpu_direct_rdma_enabled() {
 # Check if GDS is enabled
 _gpu_direct_storage_enabled() {
     if [ "${GDS_ENABLED}" = "true" ]; then
-            return 0
+        return 0
+    fi
+    return 1
+}
+
+# Check if GDRCopy is enabled
+_gdrcopy_enabled() {
+    if [ "${GDRCOPY_ENABLED}" = "true" ]; then
+        return 0
     fi
     return 1
 }

rhel8/ocp_dtk_entrypoint

@@ -186,6 +186,26 @@ dtk-build-driver() {
         echo "#"
     fi
 
+    if _gdrcopy_enabled; then
+        echo "#"
+        echo "# Executing gdrcopy driver build."
+        echo "#"
+        # Make nvidia driver sources accessible for building nvidia-fs
+        nvidia_src_dir=$(find /usr/src/nvidia-* -name "nv-p2p.c" -print -quit | xargs dirname || echo "NVIDIA_DRIVER_MISSING" 2>/dev/null)
+        if [ "nvidia_src_dir" = "NVIDIA_DRIVER_MISSING" ]; then
+            echo "Failed to find NVIDIA driver source, exiting."
+            return 1
+        fi
+        export NVIDIA_SRC_DIR=$nvidia_src_dir
+
+        # Build gdrdrv kernel module
+        make -C $DRIVER_TOOLKIT_SHARED_DIR/gdrcopy driver
+
+        echo "#"
+        echo "# gdrcopy  build script completed."
+        echo "#"
+    fi
+
     MODULES_SHARED="${DRIVER_TOOLKIT_SHARED_DIR}/modules"
     mkdir -p "${MODULES_SHARED}"
 
@@ -223,6 +243,11 @@ dtk-build-driver() {
         touch "${DRIVER_TOOLKIT_SHARED_DIR}/nvidia_fs_built"
     fi
 
+	if _gdrcopy_enabled; then
+        echo "gdrcopy driver built, inform nvidia-gdrcopy-ctr container about it and sleep forever."
+        touch "${DRIVER_TOOLKIT_SHARED_DIR}/gdrcopy_built"
+    fi
+
     while [ -f "$DRIVER_TOOLKIT_SHARED_DIR/driver_built" ]; do
         sleep 30
     done