Skip to content

Commit 108b8c5

Browse files
committed
[no-relnote] Refactor CDI version extraction
Signed-off-by: Evan Lezar <elezar@nvidia.com>
1 parent b0a5fb9 commit 108b8c5

File tree

5 files changed

+38
-49
lines changed

5 files changed

+38
-49
lines changed

pkg/nvcdi/common-nvml.go

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -24,7 +24,7 @@ import (
2424

2525
// newCommonNVMLDiscoverer returns a discoverer for entities that are not associated with a specific CDI device.
2626
// This includes driver libraries and meta devices, for example.
27-
func (l *nvmllib) newCommonNVMLDiscoverer() (discover.Discover, error) {
27+
func (l *nvmllib) newCommonNVMLDiscoverer(version string) (discover.Discover, error) {
2828
metaDevices := discover.NewCharDeviceDiscoverer(
2929
l.logger,
3030
l.devRoot,
@@ -41,7 +41,7 @@ func (l *nvmllib) newCommonNVMLDiscoverer() (discover.Discover, error) {
4141
l.logger.Warningf("failed to create discoverer for graphics mounts: %v", err)
4242
}
4343

44-
driverFiles, err := NewDriverDiscoverer(l.logger, l.driver, l.nvidiaCDIHookPath, l.ldconfigPath, l.nvmllib)
44+
driverFiles, err := NewDriverDiscoverer(l.logger, l.driver, l.nvidiaCDIHookPath, l.ldconfigPath, version)
4545
if err != nil {
4646
return nil, fmt.Errorf("failed to create discoverer for driver files: %v", err)
4747
}

pkg/nvcdi/driver-nvml.go

Lines changed: 1 addition & 16 deletions
Original file line numberDiff line numberDiff line change
@@ -22,7 +22,6 @@ import (
2222
"path/filepath"
2323
"strings"
2424

25-
"github.com/NVIDIA/go-nvml/pkg/nvml"
2625
"golang.org/x/sys/unix"
2726

2827
"github.com/NVIDIA/nvidia-container-toolkit/internal/discover"
@@ -34,21 +33,7 @@ import (
3433

3534
// NewDriverDiscoverer creates a discoverer for the libraries and binaries associated with a driver installation.
3635
// The supplied NVML Library is used to query the expected driver version.
37-
func NewDriverDiscoverer(logger logger.Interface, driver *root.Driver, nvidiaCDIHookPath string, ldconfigPath string, nvmllib nvml.Interface) (discover.Discover, error) {
38-
if r := nvmllib.Init(); r != nvml.SUCCESS {
39-
return nil, fmt.Errorf("failed to initialize NVML: %v", r)
40-
}
41-
defer func() {
42-
if r := nvmllib.Shutdown(); r != nvml.SUCCESS {
43-
logger.Warningf("failed to shutdown NVML: %v", r)
44-
}
45-
}()
46-
47-
version, r := nvmllib.SystemGetDriverVersion()
48-
if r != nvml.SUCCESS {
49-
return nil, fmt.Errorf("failed to determine driver version: %v", r)
50-
}
51-
36+
func NewDriverDiscoverer(logger logger.Interface, driver *root.Driver, nvidiaCDIHookPath string, ldconfigPath string, version string) (discover.Discover, error) {
5237
return newDriverVersionDiscoverer(logger, driver, nvidiaCDIHookPath, ldconfigPath, version)
5338
}
5439

pkg/nvcdi/lib-nvml.go

Lines changed: 6 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -83,7 +83,12 @@ func (l *nvmllib) GetAllDeviceSpecs() ([]specs.Device, error) {
8383

8484
// GetCommonEdits generates a CDI specification that can be used for ANY devices
8585
func (l *nvmllib) GetCommonEdits() (*cdi.ContainerEdits, error) {
86-
common, err := l.newCommonNVMLDiscoverer()
86+
version, err := (*nvcdilib)(l).getDriverVersion()
87+
if err != nil {
88+
return nil, fmt.Errorf("failed to get driver version: %v", err)
89+
}
90+
91+
common, err := l.newCommonNVMLDiscoverer(version)
8792
if err != nil {
8893
return nil, fmt.Errorf("failed to create discoverer for common entities: %v", err)
8994
}

pkg/nvcdi/lib.go

Lines changed: 28 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -18,6 +18,8 @@ package nvcdi
1818

1919
import (
2020
"fmt"
21+
"path/filepath"
22+
"strings"
2123

2224
"github.com/NVIDIA/go-nvlib/pkg/nvlib/device"
2325
"github.com/NVIDIA/go-nvlib/pkg/nvlib/info"
@@ -26,6 +28,7 @@ import (
2628

2729
"github.com/NVIDIA/nvidia-container-toolkit/internal/config/image"
2830
"github.com/NVIDIA/nvidia-container-toolkit/internal/logger"
31+
"github.com/NVIDIA/nvidia-container-toolkit/internal/lookup/cuda"
2932
"github.com/NVIDIA/nvidia-container-toolkit/internal/lookup/root"
3033
"github.com/NVIDIA/nvidia-container-toolkit/internal/nvsandboxutils"
3134
"github.com/NVIDIA/nvidia-container-toolkit/internal/platform-support/tegra/csv"
@@ -228,18 +231,36 @@ func (l *nvcdilib) resolveMode() (rmode string) {
228231
return ModeNvml
229232
}
230233

231-
// getCudaVersion returns the CUDA version of the current system.
232-
func (l *nvcdilib) getCudaVersion() (string, error) {
233-
version, err := l.getCudaVersionNvsandboxutils()
234-
if err == nil {
234+
// getDriverVersion returns the driver version of the current system.
235+
func (l *nvcdilib) getDriverVersion() (string, error) {
236+
if version, err := l.getDriverVersionNvsandboxutils(); err == nil && version != "" {
235237
return version, err
236238
}
237239

238240
// Fallback to NVML
239-
return l.getCudaVersionNvml()
241+
if version, err := l.getDriverVersionNvml(); err == nil && version != "" {
242+
return version, err
243+
}
244+
245+
// Fallback to getting the version from the libcuda.so suffix.
246+
return l.getDriverVersionLibcudaSo()
247+
}
248+
249+
func (l *nvcdilib) getDriverVersionLibcudaSo() (string, error) {
250+
libCudaPaths, err := cuda.New(
251+
l.driver.Libraries(),
252+
).Locate(".*.*")
253+
if err != nil {
254+
return "", fmt.Errorf("failed to locate libcuda.so: %v", err)
255+
}
256+
libCudaPath := libCudaPaths[0]
257+
258+
version := strings.TrimPrefix(filepath.Base(libCudaPath), "libcuda.so.")
259+
260+
return version, nil
240261
}
241262

242-
func (l *nvcdilib) getCudaVersionNvml() (string, error) {
263+
func (l *nvcdilib) getDriverVersionNvml() (string, error) {
243264
if hasNVML, reason := l.infolib.HasNvml(); !hasNVML {
244265
return "", fmt.Errorf("nvml not detected: %v", reason)
245266
}
@@ -263,7 +284,7 @@ func (l *nvcdilib) getCudaVersionNvml() (string, error) {
263284
return version, nil
264285
}
265286

266-
func (l *nvcdilib) getCudaVersionNvsandboxutils() (string, error) {
287+
func (l *nvcdilib) getDriverVersionNvsandboxutils() (string, error) {
267288
if l.nvsandboxutilslib == nil {
268289
return "", fmt.Errorf("libnvsandboxutils is not available")
269290
}

pkg/nvcdi/management.go

Lines changed: 1 addition & 23 deletions
Original file line numberDiff line numberDiff line change
@@ -27,7 +27,6 @@ import (
2727

2828
"github.com/NVIDIA/nvidia-container-toolkit/internal/discover"
2929
"github.com/NVIDIA/nvidia-container-toolkit/internal/edits"
30-
"github.com/NVIDIA/nvidia-container-toolkit/internal/lookup/cuda"
3130
"github.com/NVIDIA/nvidia-container-toolkit/internal/nvsandboxutils"
3231
"github.com/NVIDIA/nvidia-container-toolkit/pkg/nvcdi/spec"
3332
)
@@ -75,7 +74,7 @@ func (m *managementlib) GetCommonEdits() (*cdi.ContainerEdits, error) {
7574
}()
7675
}
7776

78-
version, err := m.getCudaVersion()
77+
version, err := (*nvcdilib)(m).getDriverVersion()
7978
if err != nil {
8079
return nil, fmt.Errorf("failed to get CUDA version: %v", err)
8180
}
@@ -93,27 +92,6 @@ func (m *managementlib) GetCommonEdits() (*cdi.ContainerEdits, error) {
9392
return edits, nil
9493
}
9594

96-
// getCudaVersion returns the CUDA version for use in managementlib containers.
97-
func (m *managementlib) getCudaVersion() (string, error) {
98-
version, err := (*nvcdilib)(m).getCudaVersion()
99-
if err == nil {
100-
return version, nil
101-
}
102-
103-
libCudaPaths, err := cuda.New(
104-
m.driver.Libraries(),
105-
).Locate(".*.*")
106-
if err != nil {
107-
return "", fmt.Errorf("failed to locate libcuda.so: %v", err)
108-
}
109-
110-
libCudaPath := libCudaPaths[0]
111-
112-
version = strings.TrimPrefix(filepath.Base(libCudaPath), "libcuda.so.")
113-
114-
return version, nil
115-
}
116-
11795
type managementDiscoverer struct {
11896
discover.Discover
11997
}

0 commit comments

Comments
 (0)