Add AI accelerator device mounting support in CreateControllerContainer

Need this to talk to device via Vulkan, ROCm, etc.

Signed-off-by: Eric Curtin <[email protected]>

Author: ericcurtin

pkg/standalone/containers.go

@@ -8,6 +8,7 @@ import (
 	"fmt"
 	"io"
 	"os"
+	"path/filepath"
 	"strconv"
 	"strings"
 	"time"
@@ -272,6 +273,32 @@ func CreateControllerContainer(ctx context.Context, dockerClient *client.Client,
 		hostConfig.DeviceRequests = []container.DeviceRequest{{Count: -1, Capabilities: [][]string{{"gpu"}}}}
 	}
 
+	// devicePaths contains glob patterns for common AI accelerator device files.
+	// Enable access to AI accelerator devices if they exist
+	devicePaths = []string{
+		"/dev/dri",       // Direct Rendering Infrastructure (used by Vulkan, Mesa, Intel/AMD GPUs)
+		"/dev/kfd",       // AMD Kernel Fusion Driver (for ROCm)
+		"/dev/accel",     // Intel accelerator devices
+		"/dev/davinci*",  // TI DaVinci video processors
+		"/dev/devmm_svm", // Huawei Ascend NPU
+		"/dev/hisi_hdc",  // Huawei Ascend NPU
+	}
+
+	for _, path := range devicePaths {
+		devices, err := filepath.Glob(path)
+		if err != nil {
+			// Skip on glob error, don't fail container creation
+			continue
+		}
+		for _, device := range devices {
+			hostConfig.Devices = append(hostConfig.Devices, container.DeviceMapping{
+				PathOnHost:        device,
+				PathInContainer:   device,
+				CgroupPermissions: "rwm",
+			})
+		}
+	}
+
 	// Create the container. If we detect that a concurrent installation is in
 	// progress (as indicated by a conflicting container name (which should have
 	// been detected just before installation)), then we'll allow the error to