Update vendoring for nvpci

Signed-off-by: Evan Lezar <elezar@nvidia.com>

vendor/gitlab.com/nvidia/cloud-native/go-nvlib/pkg/nvlib/device/api.go

@@ -26,7 +26,9 @@ type Interface interface {
 	GetMigDevices() ([]MigDevice, error)
 	GetMigProfiles() ([]MigProfile, error)
 	NewDevice(d nvml.Device) (Device, error)
+	NewDeviceByUUID(uuid string) (Device, error)
 	NewMigDevice(d nvml.Device) (MigDevice, error)
+	NewMigDeviceByUUID(uuid string) (MigDevice, error)
 	NewMigProfile(giProfileID, ciProfileID, ciEngProfileID int, migMemorySizeMB, deviceMemorySizeBytes uint64) (MigProfile, error)
 	ParseMigProfile(profile string) (MigProfile, error)
 	VisitDevices(func(i int, d Device) error) error
@@ -35,7 +37,8 @@ type Interface interface {
 }
 
 type devicelib struct {
-	nvml nvml.Interface
+	nvml           nvml.Interface
+	skippedDevices map[string]struct{}
 }
 
 var _ Interface = &devicelib{}
@@ -49,6 +52,12 @@ func New(opts ...Option) Interface {
 	if d.nvml == nil {
 		d.nvml = nvml.New()
 	}
+	if d.skippedDevices == nil {
+		WithSkippedDevices(
+			"DGX Display",
+			"NVIDIA DGX Display",
+		)(d)
+	}
 	return d
 }
 
@@ -59,5 +68,17 @@ func WithNvml(nvml nvml.Interface) 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.skippedDevices == nil {
+			d.skippedDevices = make(map[string]struct{})
+		}
+		for _, name := range names {
+			d.skippedDevices[name] = struct{}{}
+		}
+	}
+}
+
 // Option defines a function for passing options to the New() call
 type Option func(*devicelib)

vendor/gitlab.com/nvidia/cloud-native/go-nvlib/pkg/nvlib/device/device.go

@@ -43,6 +43,20 @@ var _ Device = &device{}
 
 // NewDevice builds a new Device from an nvml.Device
 func (d *devicelib) NewDevice(dev nvml.Device) (Device, error) {
+	return d.newDevice(dev)
+}
+
+// NewDeviceByUUID builds a new Device from a UUID
+func (d *devicelib) NewDeviceByUUID(uuid string) (Device, error) {
+	dev, ret := d.nvml.DeviceGetHandleByUUID(uuid)
+	if ret != nvml.SUCCESS {
+		return nil, fmt.Errorf("error getting device handle for uuid '%v': %v", uuid, ret)
+	}
+	return d.newDevice(dev)
+}
+
+// newDevice creates a device from an nvml.Device
+func (d *devicelib) newDevice(dev nvml.Device) (*device, error) {
 	return &device{dev, d}, nil
 }
 
@@ -130,6 +144,12 @@ func (d *device) VisitMigProfiles(visit func(MigProfile) error) error {
 
 	for i := 0; i < nvml.GPU_INSTANCE_PROFILE_COUNT; i++ {
 		giProfileInfo, ret := d.GetGpuInstanceProfileInfo(i)
+		if ret == nvml.ERROR_NOT_SUPPORTED {
+			continue
+		}
+		if ret == nvml.ERROR_INVALID_ARGUMENT {
+			continue
+		}
 		if ret != nvml.SUCCESS {
 			return fmt.Errorf("error getting GPU Instance profile info: %v", ret)
 		}
@@ -177,6 +197,20 @@ func (d *device) GetMigProfiles() ([]MigProfile, error) {
 	return profiles, nil
 }
 
+// isSkipped checks whether the device should be skipped.
+func (d *device) isSkipped() (bool, error) {
+	name, ret := d.GetName()
+	if ret != nvml.SUCCESS {
+		return false, fmt.Errorf("error getting device name: %v", ret)
+	}
+
+	if _, exists := d.lib.skippedDevices[name]; exists {
+		return true, nil
+	}
+
+	return false, nil
+}
+
 // VisitDevices visits each top-level device and invokes a callback function for it
 func (d *devicelib) VisitDevices(visit func(int, Device) error) error {
 	count, ret := d.nvml.DeviceGetCount()
@@ -189,10 +223,19 @@ func (d *devicelib) VisitDevices(visit func(int, Device) error) error {
 		if ret != nvml.SUCCESS {
 			return fmt.Errorf("error getting device handle for index '%v': %v", i, ret)
 		}
-		dev, err := d.NewDevice(device)
+		dev, err := d.newDevice(device)
 		if err != nil {
 			return fmt.Errorf("error creating new device wrapper: %v", err)
 		}
+
+		isSkipped, err := dev.isSkipped()
+		if err != nil {
+			return fmt.Errorf("error checking whether device is skipped: %v", err)
+		}
+		if isSkipped {
+			continue
+		}
+
 		err = visit(i, dev)
 		if err != nil {
 			return fmt.Errorf("error visiting device: %v", err)

vendor/gitlab.com/nvidia/cloud-native/go-nvlib/pkg/nvlib/device/mig_device.go

@@ -48,6 +48,15 @@ func (d *devicelib) NewMigDevice(handle nvml.Device) (MigDevice, error) {
 	return &migdevice{handle, d, nil}, nil
 }
 
+// NewMigDeviceByUUID builds a new MigDevice from a UUID
+func (d *devicelib) NewMigDeviceByUUID(uuid string) (MigDevice, error) {
+	dev, ret := d.nvml.DeviceGetHandleByUUID(uuid)
+	if ret != nvml.SUCCESS {
+		return nil, fmt.Errorf("error getting device handle for uuid '%v': %v", uuid, ret)
+	}
+	return d.NewMigDevice(dev)
+}
+
 // GetProfile returns the MIG profile associated with a MIG device
 func (m *migdevice) GetProfile() (MigProfile, error) {
 	if m.profile != nil {
@@ -101,6 +110,12 @@ func (m *migdevice) GetProfile() (MigProfile, error) {
 
 	for i := 0; i < nvml.GPU_INSTANCE_PROFILE_COUNT; i++ {
 		giProfileInfo, ret := parent.GetGpuInstanceProfileInfo(i)
+		if ret == nvml.ERROR_NOT_SUPPORTED {
+			continue
+		}
+		if ret == nvml.ERROR_INVALID_ARGUMENT {
+			continue
+		}
 		if ret != nvml.SUCCESS {
 			return nil, fmt.Errorf("error getting GPU Instance profile info: %v", ret)
 		}
@@ -112,6 +127,12 @@ func (m *migdevice) GetProfile() (MigProfile, error) {
 		for j := 0; j < nvml.COMPUTE_INSTANCE_PROFILE_COUNT; j++ {
 			for k := 0; k < nvml.COMPUTE_INSTANCE_ENGINE_PROFILE_COUNT; k++ {
 				ciProfileInfo, ret := gi.GetComputeInstanceProfileInfo(j, k)
+				if ret == nvml.ERROR_NOT_SUPPORTED {
+					continue
+				}
+				if ret == nvml.ERROR_INVALID_ARGUMENT {
+					continue
+				}
 				if ret != nvml.SUCCESS {
 					return nil, fmt.Errorf("error getting Compute Instance profile info: %v", ret)