go-nvtrust

A Go library for NVIDIA GPU and NVSwitch confidential computing attestation, providing a Go implementation inspired by nvidia/nvtrust. This library enables secure attestation of NVIDIA GPUs (Hopper, Blackwell) and NVSwitch devices with Confidential Computing capabilities.

Overview

go-nvtrust provides a comprehensive solution for:

• Collecting attestation evidence from NVIDIA GPUs and NVSwitch devices
• Verifying attestation reports through NVIDIA Remote Attestation Service (NRAS)
• Enabling confidential computing workflows
• Go bindings for libnvidia-nscq (NVSwitch attestation library)

The library leverages:

• NVIDIA's NVML (NVIDIA Management Library) through go-nvml for GPU attestation
• NVIDIA's libnvidia-nscq for NVSwitch attestation

Supported Hardware

• GPUs: NVIDIA Hopper (H100), Blackwell architectures
• NVSwitch: LS10 architecture

Installation

go get github.com/confidentsecurity/go-nvtrust

Quick Start

Attestation

    ctx := context.Background()
    nonce := make([]byte, 32)
    if _, err := rand.Read(nonce); err != nil {
        log.Fatalf("Failed to generate nonce: %v", err)
    }

    nrasClient := nras.NewNRASClient(http.DefaultClient)

    // Create GPU admin and attester
    gpuAdmin, err := gpu.NewNvmlGPUAdmin(nil)
    if err != nil {
        log.Fatalf("Failed to create GPU admin: %v", err)
    }
    defer gpuAdmin.Shutdown()

    // Attest GPUs
    attester := gonvtrust.NewRemoteAttester(gpuAdmin, nrasClient)
    result, err := attester.Attest(ctx, nonce)
    if err != nil {
        log.Fatalf("Failed to attest: %v", err)
    }
    if result.Result {
        fmt.Println("GPU attestation successful - GPUs are trusted")
        fmt.Printf("Verified %d GPU(s)\n", len(result.DevicesTokens))
    } else {
        fmt.Println("GPU attestation failed")
    }

    // Attest NVSwitches
    attester = gonvtrust.NewRemoteAttester(switchAdmin, nrasClient)
    result, err = attester.Attest(ctx, nonce)
    if err != nil {
        log.Fatalf("Failed to attest: %v", err)
    }

    if result.Result {
        fmt.Println("NVSwitch attestation successful - switches are trusted")
        fmt.Printf("Verified %d NVSwitch(es)\n", len(result.DevicesTokens))
    } else {
        fmt.Println("NVSwitch attestation failed")
    }

API Reference

Core Attestation API

RemoteAttester[T DeviceInfo]

Generic attester for both GPU and NVSwitch devices.

• NewRemoteAttester[T DeviceInfo](admin DeviceAdmin[T], verifier RemoteVerifier) *RemoteAttester[T] - Creates a new remote attester
• Attest(ctx context.Context, nonce []byte) (*AttestationResult, error) - Collects evidence and verifies it remotely

AttestationResult

type AttestationResult struct {
    Result        bool              // Overall attestation result
    JWTToken      *jwt.Token        // JWT token from NRAS
    DevicesTokens map[string]string // Individual device tokens
}

GPU Administration

NvmlGPUAdmin

Manages NVIDIA GPU attestation through NVML.

• NewNvmlGPUAdmin(handler NvmlHandler) (*NvmlGPUAdmin, error) - Creates a new GPU admin
• CollectEvidence(nonce []byte) ([]GPUDevice, error) - Collects attestation evidence from all GPUs
• IsConfidentialComputeEnabled() (bool, error) - Checks if confidential compute is enabled
• IsGPUReadyStateEnabled() (bool, error) - Checks if GPU ready state is enabled
• EnableGPUReadyState() error - Enables GPU ready state
• AllGPUInPersistenceMode() (bool, error) - Checks if all GPUs are in persistence mode
• Shutdown() error - Shuts down the NVML library

NVSwitch Administration

NscqSwitchAdmin

Manages NVSwitch attestation through libnvidia-nscq.

• NewNscqSwitchAdmin(handler NvSwitchHandler) (*NscqSwitchAdmin, error) - Creates a new switch admin
• CollectEvidence(nonce []byte) ([]SwitchDevice, error) - Collects attestation evidence from all switches
• Shutdown() error - Shuts down the NSCQ library

NRAS Client

NRASClient

Client for communicating with NVIDIA Remote Attestation Service.

• NewNRASClient(httpClient *http.Client) *NRASClient - Creates a new NRAS client
• AttestGPU(ctx context.Context, request *AttestationRequest) (*AttestationResponse, error) - Attests GPU evidence
• AttestSwitch(ctx context.Context, request *AttestationRequest) (*AttestationResponse, error) - Attests switch evidence
• VerifyJWT(ctx context.Context, signedToken string) (*jwt.Token, error) - Verifies JWT token from NRAS

Testing

Unit Tests

Run the standard test suite:

go test -v ./pkg/...

Integration Tests

Run integration tests (requires compatible GPU hardware):

go test -tags=gpu_integration -v ./pkg/...