[release-1.19] Fix handling of CUDA Compat header on Orin systems

CHANGELOG.md

@@ -9,6 +9,7 @@
 - implement NRI plugin reconnect mechanism
 - ensure the toolkit exits if NRI Plugin init fails
 - feat(nvcdi): Allow IPC sockets to not be discovered
+- fix(cudacompat): Fix handling of CUDA compat on Orin
 
 ## v1.19.0
 - Promote v1.19.0-rc.7 to v1.19.0

cmd/nvidia-cdi-hook/cudacompat/cuda-elf-header.go

@@ -50,6 +50,9 @@ func (h elf32_Nhdr) sizeof() int {
 	return 12
 }
 
+// GetCUDACompatElfHeader returns the elf header for the specified library.
+// This should be equivalent to:
+// readelf -p .note.cuda.fwd_compatibility {{.libraryPath}}
 func GetCUDACompatElfHeader(libraryPath string) (*compatElfHeader, error) {
 	lib, err := elf.Open(libraryPath)
 	if os.IsNotExist(err) {
@@ -99,8 +102,13 @@ func alignUp[T uint32 | uint64, S uint64](size T, to S) int {
 	return int((size + T(to) - 1) &^ (T(to) - 1))
 }
 
-func trim(data []byte, from int, len int) []byte {
-	return bytes.Trim(data[from:from+len], "\x00")
+func trim(data []byte, from int, n int) []byte {
+	if len(data) == 0 {
+		return nil
+	}
+	from = min(len(data)-1, from)
+	to := min(len(data), from+n)
+	return bytes.Trim(data[from:to], "\x00")
 }
 
 func getCUDAFwdCompatibilitySection(lib *elf.File) *elf.Section {

cmd/nvidia-cdi-hook/cudacompat/cuda-elf-header_test.go

@@ -33,12 +33,13 @@ func TestGetCUDACompatElfHeader(t *testing.T) {
 	dataRoot := filepath.Join(moduleRoot, "testdata", "compat")
 
 	testCases := []struct {
-		description string
-		filename    string
-		expected    *compatElfHeader
+		description   string
+		filename      string
+		expected      *compatElfHeader
+		expectedError string
 	}{
 		{
-			description: "wip",
+			description: "libcuda.so.575.57.08",
 			filename:    "libcuda.so.575.57.08",
 			expected: &compatElfHeader{
 				Format:      1,
@@ -48,7 +49,7 @@ func TestGetCUDACompatElfHeader(t *testing.T) {
 			},
 		},
 		{
-			description: "wip",
+			description: "libcuda.so.590.44.01",
 			filename:    "libcuda.so.590.44.01",
 			expected: &compatElfHeader{
 				Format:      1,
@@ -57,13 +58,31 @@ func TestGetCUDACompatElfHeader(t *testing.T) {
 				Device:      []int{1, 2, 7, 8, 9, 10, 11, 12, 13, 14},
 			},
 		},
+		{
+			description:   "invalid json",
+			filename:      "libcuda.invalid.json.so.99.88",
+			expectedError: "could not unmarshal JSON data",
+		},
+		{
+			description: "orin-13.2.1",
+			filename:    "libcuda.orin.13.2.1.so.1.1",
+			expected: &compatElfHeader{
+				Format:      1,
+				CUDAVersion: "13.2",
+			},
+		},
 	}
 
 	for _, tc := range testCases {
 		t.Run(tc.description, func(t *testing.T) {
 			libpath := filepath.Join(dataRoot, tc.filename)
 
 			h, err := GetCUDACompatElfHeader(libpath)
+			if tc.expectedError != "" {
+				require.ErrorContains(t, err, tc.expectedError)
+				require.Nil(t, h)
+				return
+			}
 			require.NoError(t, err)
 
 			require.EqualValues(t, tc.expected, h)