[no-relnote] Refactor CDI version extraction

Signed-off-by: Evan Lezar <[email protected]>

pkg/nvcdi/common-nvml.go

@@ -24,7 +24,7 @@ import (
 
 // newCommonNVMLDiscoverer returns a discoverer for entities that are not associated with a specific CDI device.
 // This includes driver libraries and meta devices, for example.
-func (l *nvmllib) newCommonNVMLDiscoverer() (discover.Discover, error) {
+func (l *nvmllib) newCommonNVMLDiscoverer(version string) (discover.Discover, error) {
 	metaDevices := discover.NewCharDeviceDiscoverer(
 		l.logger,
 		l.devRoot,
@@ -41,7 +41,7 @@ func (l *nvmllib) newCommonNVMLDiscoverer() (discover.Discover, error) {
 		l.logger.Warningf("failed to create discoverer for graphics mounts: %v", err)
 	}
 
-	driverFiles, err := NewDriverDiscoverer(l.logger, l.driver, l.nvidiaCDIHookPath, l.ldconfigPath, l.nvmllib)
+	driverFiles, err := NewDriverDiscoverer(l.logger, l.driver, l.nvidiaCDIHookPath, l.ldconfigPath, version)
 	if err != nil {
 		return nil, fmt.Errorf("failed to create discoverer for driver files: %v", err)
 	}

pkg/nvcdi/driver-nvml.go

@@ -22,7 +22,6 @@ import (
 	"path/filepath"
 	"strings"
 
-	"github.com/NVIDIA/go-nvml/pkg/nvml"
 	"golang.org/x/sys/unix"
 
 	"github.com/NVIDIA/nvidia-container-toolkit/internal/discover"
@@ -34,21 +33,7 @@ import (
 
 // NewDriverDiscoverer creates a discoverer for the libraries and binaries associated with a driver installation.
 // The supplied NVML Library is used to query the expected driver version.
-func NewDriverDiscoverer(logger logger.Interface, driver *root.Driver, nvidiaCDIHookPath string, ldconfigPath string, nvmllib nvml.Interface) (discover.Discover, error) {
-	if r := nvmllib.Init(); r != nvml.SUCCESS {
-		return nil, fmt.Errorf("failed to initialize NVML: %v", r)
-	}
-	defer func() {
-		if r := nvmllib.Shutdown(); r != nvml.SUCCESS {
-			logger.Warningf("failed to shutdown NVML: %v", r)
-		}
-	}()
-
-	version, r := nvmllib.SystemGetDriverVersion()
-	if r != nvml.SUCCESS {
-		return nil, fmt.Errorf("failed to determine driver version: %v", r)
-	}
-
+func NewDriverDiscoverer(logger logger.Interface, driver *root.Driver, nvidiaCDIHookPath string, ldconfigPath string, version string) (discover.Discover, error) {
 	return newDriverVersionDiscoverer(logger, driver, nvidiaCDIHookPath, ldconfigPath, version)
 }
 

pkg/nvcdi/lib-nvml.go

@@ -83,7 +83,12 @@ func (l *nvmllib) GetAllDeviceSpecs() ([]specs.Device, error) {
 
 // GetCommonEdits generates a CDI specification that can be used for ANY devices
 func (l *nvmllib) GetCommonEdits() (*cdi.ContainerEdits, error) {
-	common, err := l.newCommonNVMLDiscoverer()
+	version, err := (*nvcdilib)(l).getDriverVersion()
+	if err != nil {
+		return nil, fmt.Errorf("failed to get driver version: %v", err)
+	}
+
+	common, err := l.newCommonNVMLDiscoverer(version)
 	if err != nil {
 		return nil, fmt.Errorf("failed to create discoverer for common entities: %v", err)
 	}

pkg/nvcdi/lib.go

@@ -18,6 +18,8 @@ package nvcdi
 
 import (
 	"fmt"
+	"path/filepath"
+	"strings"
 
 	"github.com/NVIDIA/go-nvlib/pkg/nvlib/device"
 	"github.com/NVIDIA/go-nvlib/pkg/nvlib/info"
@@ -26,6 +28,7 @@ import (
 
 	"github.com/NVIDIA/nvidia-container-toolkit/internal/config/image"
 	"github.com/NVIDIA/nvidia-container-toolkit/internal/logger"
+	"github.com/NVIDIA/nvidia-container-toolkit/internal/lookup/cuda"
 	"github.com/NVIDIA/nvidia-container-toolkit/internal/lookup/root"
 	"github.com/NVIDIA/nvidia-container-toolkit/internal/nvsandboxutils"
 	"github.com/NVIDIA/nvidia-container-toolkit/internal/platform-support/tegra/csv"
@@ -228,18 +231,36 @@ func (l *nvcdilib) resolveMode() (rmode string) {
 	return ModeNvml
 }
 
-// getCudaVersion returns the CUDA version of the current system.
-func (l *nvcdilib) getCudaVersion() (string, error) {
-	version, err := l.getCudaVersionNvsandboxutils()
-	if err == nil {
+// getDriverVersion returns the driver version of the current system.
+func (l *nvcdilib) getDriverVersion() (string, error) {
+	if version, err := l.getDriverVersionNvsandboxutils(); err == nil && version != "" {
 		return version, err
 	}
 
 	// Fallback to NVML
-	return l.getCudaVersionNvml()
+	if version, err := l.getDriverVersionNvml(); err == nil && version != "" {
+		return version, err
+	}
+
+	// Fallback to getting the version from the libcuda.so suffix.
+	return l.getDriverVersionLibcudaSo()
+}
+
+func (l *nvcdilib) getDriverVersionLibcudaSo() (string, error) {
+	libCudaPaths, err := cuda.New(
+		l.driver.Libraries(),
+	).Locate(".*.*")
+	if err != nil {
+		return "", fmt.Errorf("failed to locate libcuda.so: %v", err)
+	}
+	libCudaPath := libCudaPaths[0]
+
+	version := strings.TrimPrefix(filepath.Base(libCudaPath), "libcuda.so.")
+
+	return version, nil
 }
 
-func (l *nvcdilib) getCudaVersionNvml() (string, error) {
+func (l *nvcdilib) getDriverVersionNvml() (string, error) {
 	if hasNVML, reason := l.infolib.HasNvml(); !hasNVML {
 		return "", fmt.Errorf("nvml not detected: %v", reason)
 	}
@@ -263,7 +284,7 @@ func (l *nvcdilib) getCudaVersionNvml() (string, error) {
 	return version, nil
 }
 
-func (l *nvcdilib) getCudaVersionNvsandboxutils() (string, error) {
+func (l *nvcdilib) getDriverVersionNvsandboxutils() (string, error) {
 	if l.nvsandboxutilslib == nil {
 		return "", fmt.Errorf("libnvsandboxutils is not available")
 	}

pkg/nvcdi/management.go

@@ -27,7 +27,6 @@ import (
 
 	"github.com/NVIDIA/nvidia-container-toolkit/internal/discover"
 	"github.com/NVIDIA/nvidia-container-toolkit/internal/edits"
-	"github.com/NVIDIA/nvidia-container-toolkit/internal/lookup/cuda"
 	"github.com/NVIDIA/nvidia-container-toolkit/internal/nvsandboxutils"
 	"github.com/NVIDIA/nvidia-container-toolkit/pkg/nvcdi/spec"
 )
@@ -75,7 +74,7 @@ func (m *managementlib) GetCommonEdits() (*cdi.ContainerEdits, error) {
 		}()
 	}
 
-	version, err := m.getCudaVersion()
+	version, err := (*nvcdilib)(m).getDriverVersion()
 	if err != nil {
 		return nil, fmt.Errorf("failed to get CUDA version: %v", err)
 	}
@@ -93,27 +92,6 @@ func (m *managementlib) GetCommonEdits() (*cdi.ContainerEdits, error) {
 	return edits, nil
 }
 
-// getCudaVersion returns the CUDA version for use in managementlib containers.
-func (m *managementlib) getCudaVersion() (string, error) {
-	version, err := (*nvcdilib)(m).getCudaVersion()
-	if err == nil {
-		return version, nil
-	}
-
-	libCudaPaths, err := cuda.New(
-		m.driver.Libraries(),
-	).Locate(".*.*")
-	if err != nil {
-		return "", fmt.Errorf("failed to locate libcuda.so: %v", err)
-	}
-
-	libCudaPath := libCudaPaths[0]
-
-	version = strings.TrimPrefix(filepath.Base(libCudaPath), "libcuda.so.")
-
-	return version, nil
-}
-
 type managementDiscoverer struct {
 	discover.Discover
 }