Merge pull request #1461 from elezar/dgpu-on-nvgpu

Handle multiple GPUs in CDI spec generation from CSV

Author: elezar

go.mod

@@ -3,7 +3,7 @@ module github.com/NVIDIA/nvidia-container-toolkit
 go 1.25.0
 
 require (
-	github.com/NVIDIA/go-nvlib v0.9.0
+	github.com/NVIDIA/go-nvlib v0.9.1-0.20251202135446-d0f42ba016dd
 	github.com/NVIDIA/go-nvml v0.13.0-1
 	github.com/google/uuid v1.6.0
 	github.com/moby/sys/mountinfo v0.7.2

go.sum

@@ -1,7 +1,7 @@
 cyphar.com/go-pathrs v0.2.1 h1:9nx1vOgwVvX1mNBWDu93+vaceedpbsDqo+XuBGL40b8=
 cyphar.com/go-pathrs v0.2.1/go.mod h1:y8f1EMG7r+hCuFf/rXsKqMJrJAUoADZGNh5/vZPKcGc=
-github.com/NVIDIA/go-nvlib v0.9.0 h1:GKLIvLJ0uhCtTLLZp2Q8QIDRxOYH45MM4Y5OO3U5Rho=
-github.com/NVIDIA/go-nvlib v0.9.0/go.mod h1:7mzx9FSdO9fXWP9NKuZmWkCwhkEcSWQFe2tmFwtLb9c=
+github.com/NVIDIA/go-nvlib v0.9.1-0.20251202135446-d0f42ba016dd h1:MC1w/VYuo9Zt0se4SSx9BVid4a46ai+voN3knRvVWjE=
+github.com/NVIDIA/go-nvlib v0.9.1-0.20251202135446-d0f42ba016dd/go.mod h1:7mzx9FSdO9fXWP9NKuZmWkCwhkEcSWQFe2tmFwtLb9c=
 github.com/NVIDIA/go-nvml v0.13.0-1 h1:OLX8Jq3dONuPOQPC7rndB6+iDmDakw0XTYgzMxObkEw=
 github.com/NVIDIA/go-nvml v0.13.0-1/go.mod h1:+KNA7c7gIBH7SKSJ1ntlwkfN80zdx8ovl4hrK3LmPt4=
 github.com/blang/semver/v4 v4.0.0 h1:1PFHFE6yCCTv8C1TeyNNarDzntLi7wMI5i/pzqYIsAM=

internal/info/auto_test.go

@@ -215,13 +215,10 @@ func TestResolveAutoMode(t *testing.T) {
 				HasDXCoreFunc: func() (bool, string) {
 					return tc.info["dxcore"], "dxcore"
 				},
-				IsTegraSystemFunc: func() (bool, string) {
-					return tc.info["tegra"], "tegra"
-				},
 				HasTegraFilesFunc: func() (bool, string) {
 					return tc.info["tegra"], "tegra"
 				},
-				HasOnlyIntegratedGPUsFunc: func() (bool, string) {
+				HasAnIntegratedGPUFunc: func() (bool, string) {
 					return tc.info["nvgpu"], "nvgpu"
 				},
 			}

internal/platform-support/tegra/csv.go

@@ -25,17 +25,13 @@ import (
 	"github.com/NVIDIA/nvidia-container-toolkit/internal/platform-support/tegra/csv"
 )
 
-// newDiscovererFromCSVFiles creates a discoverer for the specified CSV files. A logger is also supplied.
-// The constructed discoverer is comprised of a list, with each element in the list being associated with a
-// single CSV files.
-func (o tegraOptions) newDiscovererFromCSVFiles() (discover.Discover, error) {
-	if len(o.csvFiles) == 0 {
-		o.logger.Warningf("No CSV files specified")
+// newDiscovererFromMountSpecs creates a discoverer for the specified mount specs.
+func (o options) newDiscovererFromMountSpecs(targetsByType MountSpecPathsByType) (discover.Discover, error) {
+	if len(targetsByType) == 0 {
+		o.logger.Warningf("No mount specs specified")
 		return discover.None{}, nil
 	}
 
-	targetsByType := getTargetsFromCSVFiles(o.logger, o.csvFiles)
-
 	devices := discover.NewCharDeviceDiscoverer(
 		o.logger,
 		o.devRoot,
@@ -64,15 +60,13 @@ func (o tegraOptions) newDiscovererFromCSVFiles() (discover.Discover, error) {
 	)
 
 	// We process the explicitly requested symlinks.
-	symlinkTargets := o.ignorePatterns.Apply(targetsByType[csv.MountSpecSym]...)
-	o.logger.Debugf("Filtered symlink targets: %v", symlinkTargets)
 	symlinks := discover.NewMounts(
 		o.logger,
 		o.symlinkLocator,
 		o.driverRoot,
-		symlinkTargets,
+		targetsByType[csv.MountSpecSym],
 	)
-	createSymlinks := o.createCSVSymlinkHooks(symlinkTargets)
+	createSymlinks := o.createCSVSymlinkHooks(targetsByType[csv.MountSpecSym])
 
 	d := discover.Merge(
 		devices,
@@ -85,23 +79,24 @@ func (o tegraOptions) newDiscovererFromCSVFiles() (discover.Discover, error) {
 	return d, nil
 }
 
-// getTargetsFromCSVFiles returns the list of mount specs from the specified CSV files.
-// These are aggregated by mount spec type.
-// TODO: We use a function variable here to allow this to be overridden for testing.
-// This should be properly mocked.
-var getTargetsFromCSVFiles = func(logger logger.Interface, files []string) map[csv.MountSpecType][]string {
-	targetsByType := make(map[csv.MountSpecType][]string)
-	for _, filename := range files {
+// MountSpecsFromCSVFiles returns a MountSpecPathsByTyper for the specified list
+// of CSV files.
+func MountSpecsFromCSVFiles(logger logger.Interface, csvFilePaths ...string) MountSpecPathsByType {
+	var mountSpecs mountSpecPathsByTypers
+
+	for _, filename := range csvFilePaths {
 		targets, err := loadCSVFile(logger, filename)
 		if err != nil {
 			logger.Warningf("Skipping CSV file %v: %v", filename, err)
 			continue
 		}
+		targetsByType := make(MountSpecPathsByType)
 		for _, t := range targets {
 			targetsByType[t.Type] = append(targetsByType[t.Type], t.Path)
 		}
+		mountSpecs = append(mountSpecs, targetsByType)
 	}
-	return targetsByType
+	return mountSpecs.MountSpecPathsByType()
 }
 
 // loadCSVFile loads the specified CSV file and returns the list of mount specs

internal/platform-support/tegra/csv_test.go

@@ -24,17 +24,14 @@ import (
 	"github.com/stretchr/testify/require"
 
 	"github.com/NVIDIA/nvidia-container-toolkit/internal/discover"
-	"github.com/NVIDIA/nvidia-container-toolkit/internal/logger"
 	"github.com/NVIDIA/nvidia-container-toolkit/internal/lookup"
-
-	"github.com/NVIDIA/nvidia-container-toolkit/internal/platform-support/tegra/csv"
 )
 
 func TestDiscovererFromCSVFiles(t *testing.T) {
 	logger, _ := testlog.NewNullLogger()
 	testCases := []struct {
 		description         string
-		moutSpecs           map[csv.MountSpecType][]string
+		moutSpecs           MountSpecPathsByType
 		ignorePatterns      []string
 		symlinkLocator      lookup.Locator
 		symlinkChainLocator lookup.Locator
@@ -49,7 +46,7 @@ func TestDiscovererFromCSVFiles(t *testing.T) {
 			// TODO: This current resolves to two mounts that are the same.
 			// These are deduplicated at a later stage. We could consider deduplicating earlier in the pipeline.
 			description: "symlink is resolved to target; mounts and symlink are created",
-			moutSpecs: map[csv.MountSpecType][]string{
+			moutSpecs: MountSpecPathsByType{
 				"lib": {"/usr/lib/aarch64-linux-gnu/tegra/libv4l2_nvargus.so"},
 				"sym": {"/usr/lib/aarch64-linux-gnu/libv4l/plugins/nv/libv4l2_nvargus.so"},
 			},
@@ -105,7 +102,7 @@ func TestDiscovererFromCSVFiles(t *testing.T) {
 			// TODO: This current resolves to two mounts that are the same.
 			// These are deduplicated at a later stage. We could consider deduplicating earlier in the pipeline.
 			description: "single glob filter does not remove symlink mounts",
-			moutSpecs: map[csv.MountSpecType][]string{
+			moutSpecs: MountSpecPathsByType{
 				"lib": {"/usr/lib/aarch64-linux-gnu/tegra/libv4l2_nvargus.so"},
 				"sym": {"/usr/lib/aarch64-linux-gnu/libv4l/plugins/nv/libv4l2_nvargus.so"},
 			},
@@ -160,7 +157,7 @@ func TestDiscovererFromCSVFiles(t *testing.T) {
 		},
 		{
 			description: "** filter removes symlink mounts",
-			moutSpecs: map[csv.MountSpecType][]string{
+			moutSpecs: MountSpecPathsByType{
 				"lib": {"/usr/lib/aarch64-linux-gnu/tegra/libv4l2_nvargus.so"},
 				"sym": {"/usr/lib/aarch64-linux-gnu/libv4l/plugins/nv/libv4l2_nvargus.so"},
 			},
@@ -186,19 +183,20 @@ func TestDiscovererFromCSVFiles(t *testing.T) {
 	hookCreator := discover.NewHookCreator()
 	for _, tc := range testCases {
 		t.Run(tc.description, func(t *testing.T) {
-			defer setGetTargetsFromCSVFiles(tc.moutSpecs)()
-
-			o := tegraOptions{
+			o := options{
 				logger:              logger,
 				hookCreator:         hookCreator,
-				csvFiles:            []string{"dummy"},
-				ignorePatterns:      tc.ignorePatterns,
 				symlinkLocator:      tc.symlinkLocator,
 				symlinkChainLocator: tc.symlinkChainLocator,
 				resolveSymlink:      tc.symlinkResolver,
+
+				mountSpecs: Transform(
+					tc.moutSpecs,
+					IgnoreSymlinkMountSpecsByPattern(tc.ignorePatterns...),
+				),
 			}
 
-			d, err := o.newDiscovererFromCSVFiles()
+			d, err := o.newDiscovererFromMountSpecs(o.mountSpecs.MountSpecPathsByType())
 			require.ErrorIs(t, err, tc.expectedError)
 
 			hooks, err := d.Hooks()
@@ -212,14 +210,3 @@ func TestDiscovererFromCSVFiles(t *testing.T) {
 		})
 	}
 }
-
-func setGetTargetsFromCSVFiles(override map[csv.MountSpecType][]string) func() {
-	original := getTargetsFromCSVFiles
-	getTargetsFromCSVFiles = func(logger logger.Interface, files []string) map[csv.MountSpecType][]string {
-		return override
-	}
-
-	return func() {
-		getTargetsFromCSVFiles = original
-	}
-}

internal/platform-support/tegra/filter.go

@@ -19,11 +19,13 @@ package tegra
 import (
 	"path/filepath"
 	"strings"
+
+	"github.com/NVIDIA/nvidia-container-toolkit/internal/platform-support/tegra/csv"
 )
 
-type ignoreMountSpecPatterns []string
+type ignoreSymlinkMountSpecPatterns []string
 
-func (d ignoreMountSpecPatterns) Match(name string) bool {
+func (d ignoreSymlinkMountSpecPatterns) match(name string) bool {
 	for _, pattern := range d {
 		target := name
 		if strings.HasPrefix(pattern, "**/") {
@@ -37,13 +39,109 @@ func (d ignoreMountSpecPatterns) Match(name string) bool {
 	return false
 }
 
-func (d ignoreMountSpecPatterns) Apply(input ...string) []string {
+func (d ignoreSymlinkMountSpecPatterns) filter(input ...string) []string {
 	var filtered []string
 	for _, name := range input {
-		if d.Match(name) {
+		if d.match(name) {
 			continue
 		}
 		filtered = append(filtered, name)
 	}
 	return filtered
 }
+
+func (d ignoreSymlinkMountSpecPatterns) Apply(input MountSpecPathsByTyper) MountSpecPathsByTyper {
+	ms := input.MountSpecPathsByType()
+
+	if symlinks, ok := ms[csv.MountSpecSym]; ok {
+		ms[csv.MountSpecSym] = d.filter(symlinks...)
+	}
+
+	return ms
+}
+
+// A filter removes elements from an input list and returns the remaining
+// elements.
+type filter interface {
+	apply(...string) []string
+}
+
+// A stringMatcher implements the MatchString function.
+type stringMatcher interface {
+	MatchString(string) bool
+}
+
+// A matcherAsFilter is used to ensure that a string matcher can be used as a filter.
+type matcherAsFilter struct {
+	stringMatcher
+}
+
+type filterByMountSpecType map[csv.MountSpecType]filter
+
+type pathPatterns []string
+type pathPattern string
+type basenamePattern string
+
+// MatchString for a set of path patterns returns true if any of the patterns
+// matches against the input string.
+func (d pathPatterns) MatchString(input string) bool {
+	for _, pattern := range d {
+		if match := pathPattern(pattern).MatchString(input); match {
+			return true
+		}
+	}
+	return false
+}
+
+// MatchString attempts to match a path pattern to the specified input string.
+// If the pattern starts with `**/` the input is treated as a path and only
+// the basenames are matched using regular glob rules.
+func (d pathPattern) MatchString(input string) bool {
+	if strings.HasPrefix(string(d), "**/") {
+		return basenamePattern(d).MatchString(input)
+	}
+	match, _ := filepath.Match(string(d), input)
+	return match
+}
+
+// MatchString for a basename pattern applies the specified pattern against the
+// basename of the input.
+// If the pattern starts with **/, this is stripped before attempting to match.
+func (d basenamePattern) MatchString(input string) bool {
+	pattern := strings.TrimPrefix(string(d), "**/")
+	match, _ := filepath.Match(pattern, filepath.Base(input))
+	return match
+}
+
+// Apply the specified per-type filters to the input mount specs.
+func (p filterByMountSpecType) Apply(input MountSpecPathsByTyper) MountSpecPathsByTyper {
+	ms := input.MountSpecPathsByType()
+	for t, filter := range p {
+		if len(ms[t]) == 0 {
+			continue
+		}
+		ms[t] = filter.apply(ms[t]...)
+	}
+	return ms
+}
+
+// apply uses a matcher to filter an input string.
+// Each element in the input that matches is skipped and the remaining elements
+// are returned.
+func (f *matcherAsFilter) apply(input ...string) []string {
+	var filtered []string
+	for _, path := range input {
+		if f.MatchString(path) {
+			continue
+		}
+		filtered = append(filtered, path)
+	}
+	return filtered
+}
+
+// removeAll is a filter that will not return any inputs.
+type removeAll struct{}
+
+func (a removeAll) apply(...string) []string {
+	return nil
+}

internal/platform-support/tegra/filter_test.go

@@ -25,7 +25,7 @@ import (
 func TestIgnorePatterns(t *testing.T) {
 	testCases := []struct {
 		description   string
-		blockedFilter []string
+		blockedFilter pathPatterns
 		input         []string
 		expected      []string
 	}{
@@ -50,7 +50,7 @@ func TestIgnorePatterns(t *testing.T) {
 
 	for _, tc := range testCases {
 		t.Run(tc.description, func(t *testing.T) {
-			filtered := ignoreMountSpecPatterns(tc.blockedFilter).Apply(tc.input...)
+			filtered := ignoreSymlinkMountSpecPatterns(tc.blockedFilter).filter(tc.input...)
 			require.ElementsMatch(t, tc.expected, filtered)
 		})
 	}