Merge branch 'skip-display-devices-on-name' into 'main'

Skip display devices based on model name

See merge request nvidia/cloud-native/go-nvlib!26

Makefile

@@ -96,3 +96,15 @@ $(DOCKER_TARGETS): docker-%: .build-image
 		--user $$(id -u):$$(id -g) \
 		$(BUILDIMAGE) \
 			make $(*)
+
+# Start an interactive shell using the development image.
+PHONY: .shell
+.shell:
+	$(DOCKER) run \
+		--rm \
+		-ti \
+		-e GOCACHE=/tmp/.cache \
+		-v $(PWD):$(PWD) \
+		-w $(PWD) \
+		--user $$(id -u):$$(id -g) \
+		$(BUILDIMAGE)

docker/Dockerfile.devel

@@ -15,4 +15,4 @@ ARG GOLANG_VERSION=1.16
 FROM golang:${GOLANG_VERSION}
 
 RUN go get -u golang.org/x/lint/golint
-RUN go install github.com/matryer/moq@latest
+RUN go install github.com/matryer/moq@v0.2.7

pkg/nvlib/device/api.go

@@ -36,7 +36,7 @@ type Interface interface {
 
 type devicelib struct {
 	nvml           nvml.Interface
-	selectedDeviceClasses map[Class]struct{}
+	skippedDevices map[string]struct{}
 }
 
 var _ Interface = &devicelib{}
@@ -50,10 +50,8 @@ func New(opts ...Option) Interface {
 	if d.nvml == nil {
 		d.nvml = nvml.New()
 	}
-	if d.selectedDeviceClasses == nil {
-		d.selectedDeviceClasses = map[Class]struct{}{
-			ClassCompute: {},
-		}
+	if d.skippedDevices == nil {
+		WithSkippedDevices("NVIDIA DGX Display")(d)
 	}
 	return d
 }
@@ -65,14 +63,14 @@ func WithNvml(nvml nvml.Interface) Option {
 	}
 }
 
-// WithSelectedDeviceClasses selects the specified device classes when filtering devices
-func WithSelectedDeviceClasses(classes ...Class) Option {
+// WithSkippedDevices provides an Option to set devices to be skipped by model name
+func WithSkippedDevices(names ...string) Option {
 	return func(d *devicelib) {
-		if d.selectedDeviceClasses == nil {
-			d.selectedDeviceClasses = make(map[Class]struct{})
+		if d.skippedDevices == nil {
+			d.skippedDevices = make(map[string]struct{})
 		}
-		for _, c := range classes {
-			d.selectedDeviceClasses[c] = struct{}{}
+		for _, name := range names {
+			d.skippedDevices[name] = struct{}{}
 		}
 	}
 }

pkg/nvlib/device/device.go

@@ -18,11 +18,9 @@ package device
 
 import (
 	"fmt"
-	"strings"
 
 	"github.com/NVIDIA/go-nvml/pkg/dl"
 	"gitlab.com/nvidia/cloud-native/go-nvlib/pkg/nvml"
-	"gitlab.com/nvidia/cloud-native/go-nvlib/pkg/nvpci"
 )
 
 // Device defines the set of extended functions associated with a device.Device
@@ -41,15 +39,6 @@ type device struct {
 	lib *devicelib
 }
 
-// Class represents the PCI class for a device
-type Class uint32
-
-// Define constants for common device classes
-const (
-	ClassCompute = Class(nvpci.PCI3dControllerClass)
-	ClassDisplay = Class(nvpci.PCIVgaControllerClass)
-)
-
 var _ Device = &device{}
 
 // NewDevice builds a new Device from an nvml.Device
@@ -62,16 +51,6 @@ func (d *devicelib) newDevice(dev nvml.Device) (*device, error) {
 	return &device{dev, d}, nil
 }
 
-// classIsSelected checks whether the specified class has been selected when constructing the devicelib
-func (d *devicelib) classIsSelected(c Class) bool {
-	if d.selectedDeviceClasses == nil {
-		return false
-	}
-	_, exists := d.selectedDeviceClasses[c]
-
-	return exists
-}
-
 // IsMigCapable checks if a device is capable of having MIG paprtitions created on it
 func (d *device) IsMigCapable() (bool, error) {
 	err := nvmlLookupSymbol("nvmlDeviceGetMigMode")
@@ -209,33 +188,18 @@ func (d *device) GetMigProfiles() ([]MigProfile, error) {
 	return profiles, nil
 }
 
-// getClass returns the PCI device class for the device
-func (d *device) getClass() (Class, error) {
-	info, ret := d.GetPciInfo()
+// isSkipped checks whether the device should be skipped.
+func (d *device) isSkipped() (bool, error) {
+	name, ret := d.GetName()
 	if ret != nvml.SUCCESS {
-		return 0, fmt.Errorf("failed to get PCI info: %v", ret)
+		return false, fmt.Errorf("error getting device name: %v", ret)
 	}
 
-	// We convert the BusId to a string
-	var bytes []byte
-	for _, b := range info.BusId {
-		if byte(b) == '\x00' {
-			break
-		}
-		bytes = append(bytes, byte(b))
-	}
-	id := strings.ToLower(string(bytes))
-
-	if id != "0000" {
-		id = strings.TrimPrefix(id, "0000")
+	if _, exists := d.lib.skippedDevices[name]; exists {
+		return true, nil
 	}
 
-	device, err := nvpci.New().GetGPUByPciBusID(id)
-	if err != nil {
-		return 0, fmt.Errorf("failed to construct PCI device: %v", err)
-	}
-
-	return Class(device.Class), nil
+	return false, nil
 }
 
 // VisitDevices visits each top-level device and invokes a callback function for it
@@ -255,11 +219,11 @@ func (d *devicelib) VisitDevices(visit func(int, Device) error) error {
 			return fmt.Errorf("error creating new device wrapper: %v", err)
 		}
 
-		class, err := dev.getClass()
+		isSkipped, err := dev.isSkipped()
 		if err != nil {
-			return fmt.Errorf("error getting PCI device class for device: %v", err)
+			return fmt.Errorf("error checking whether device is skipped: %v", err)
 		}
-		if !d.classIsSelected(class) {
+		if isSkipped {
 			continue
 		}