gpu-manager: enable kernel module configuration via KernelModuleConfig

[shivakunv]

No description provided.

[tariq1890]

let's wait for PR #420 to get merged. After that, we can update this PR to align with #420

[shivakunv]

let's wait for PR #420 to get merged. After that, we can update this PR to align with #420

adapted similar change for vgpu

[cdesiniotis]

This change did not make sense to me at first. But then I realized that the _install_driver() actually loads the modules (we leverage the nvidia-installer to load the modules, in addition to building+installing the driver).

Question -- If we create the modprobe conf file (/etc/modprobe.d/nvidia.conf) BEFORE calling _install_driver(), what is the observed behavior? Do the modules get loaded with the desired parameters?

[shivakunv]

1. I put the else condition inside _create_module_params_conf and tested it , it is working fine .

 else
        echo "create nvidia.conf file"
        echo "options nvidia NVreg_EnableGpuFirmwareLogs=3" > ${MODPROBE_CONFIG_DIR}/nvidia.conf
    fi

_create_module_params_conf() {
    echo "Parsing kernel module parameters..."
    _get_module_params

    if [ ${#NVIDIA_MODULE_PARAMS[@]} -gt 0 ]; then
        echo "Configuring nvidia module parameters in ${MODPROBE_CONFIG_DIR}/nvidia.conf"
        echo "options nvidia ${NVIDIA_MODULE_PARAMS[@]}" > ${MODPROBE_CONFIG_DIR}/nvidia.conf
    else
        echo "create nvidia.conf file"
        echo "options nvidia NVreg_EnableGpuFirmwareLogs=3" > ${MODPROBE_CONFIG_DIR}/nvidia.conf
    fi
}

2. I created a configmap for the nvdia.conf file and tested it , which is also working fine . ( refer : PR: vgpu-manager: enable kernel module configuration via KernelModuleConfig gpu-operator#1946

If I need to test it in other ways, please list the steps .

[cdesiniotis]

My original comment is questioning whether this code snippet (lines 96-L107) is actually needed. Can you remove this snippet and verify whether custom module parameters can be applied or not?

[shivakunv]

It is needed .
After removal, the custom module parameter does not take effect and required a system reboot to be applied.