Bump github.com/NVIDIA/go-nvlib from 0.5.0 to 0.6.0

Bumps [github.com/NVIDIA/go-nvlib](https://github.com/NVIDIA/go-nvlib) from 0.5.0 to 0.6.0.
- [Release notes](https://github.com/NVIDIA/go-nvlib/releases)
- [Commits](https://github.com/NVIDIA/go-nvlib/compare/v0.5.0...v0.6.0)

---
updated-dependencies:
- dependency-name: github.com/NVIDIA/go-nvlib
  dependency-type: direct:production
  update-type: version-update:semver-minor
...

Signed-off-by: dependabot[bot] <support@github.com>

go.mod

@@ -3,7 +3,7 @@ module github.com/NVIDIA/nvidia-container-toolkit
 go 1.20
 
 require (
-	github.com/NVIDIA/go-nvlib v0.5.0
+	github.com/NVIDIA/go-nvlib v0.6.0
 	github.com/NVIDIA/go-nvml v0.12.4-0
 	github.com/fsnotify/fsnotify v1.7.0
 	github.com/opencontainers/runtime-spec v1.2.0

go.sum

@@ -1,5 +1,5 @@
-github.com/NVIDIA/go-nvlib v0.5.0 h1:951KGrfr+p3cs89alO9z/ZxPPWKxwht9tx9rxiADoLI=
-github.com/NVIDIA/go-nvlib v0.5.0/go.mod h1:87z49ULPr4GWPSGfSIp3taU4XENRYN/enIg88MzcL4k=
+github.com/NVIDIA/go-nvlib v0.6.0 h1:zAMBzCYT9xeyRQo0tb7HJbStkzajD6e5joyaQqJ2OGU=
+github.com/NVIDIA/go-nvlib v0.6.0/go.mod h1:9UrsLGx/q1OrENygXjOuM5Ey5KCtiZhbvBlbUIxtGWY=
 github.com/NVIDIA/go-nvml v0.12.4-0 h1:4tkbB3pT1O77JGr0gQ6uD8FrsUPqP1A/EOEm2wI1TUg=
 github.com/NVIDIA/go-nvml v0.12.4-0/go.mod h1:8Llmj+1Rr+9VGGwZuRer5N/aCjxGuR5nPb/9ebBiIEQ=
 github.com/blang/semver/v4 v4.0.0 h1:1PFHFE6yCCTv8C1TeyNNarDzntLi7wMI5i/pzqYIsAM=

vendor/github.com/NVIDIA/go-nvlib/pkg/nvlib/device/device.go

@@ -18,6 +18,7 @@ package device
 
 import (
 	"fmt"
+	"strings"
 
 	"github.com/NVIDIA/go-nvml/pkg/nvml"
 )
@@ -30,6 +31,7 @@ type Device interface {
 	GetCudaComputeCapabilityAsString() (string, error)
 	GetMigDevices() ([]MigDevice, error)
 	GetMigProfiles() ([]MigProfile, error)
+	GetPCIBusID() (string, error)
 	IsMigCapable() (bool, error)
 	IsMigEnabled() (bool, error)
 	VisitMigDevices(func(j int, m MigDevice) error) error
@@ -140,6 +142,29 @@ func (d *device) GetBrandAsString() (string, error) {
 	return "", fmt.Errorf("error interpreting device brand as string: %v", brand)
 }
 
+// GetPCIBusID returns the string representation of the bus ID.
+func (d *device) GetPCIBusID() (string, error) {
+	info, ret := d.GetPciInfo()
+	if ret != nvml.SUCCESS {
+		return "", fmt.Errorf("error getting PCI info: %w", ret)
+	}
+
+	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")
+	}
+
+	return id, nil
+}
+
 // GetCudaComputeCapabilityAsString returns the Device's CUDA compute capability as a version string.
 func (d *device) GetCudaComputeCapabilityAsString() (string, error) {
 	major, minor, ret := d.GetCudaComputeCapability()

vendor/github.com/NVIDIA/go-nvlib/pkg/nvpci/mock.go

@@ -20,6 +20,8 @@ import (
 	"fmt"
 	"os"
 	"path/filepath"
+	"regexp"
+	"strconv"
 
 	"github.com/NVIDIA/go-nvlib/pkg/nvpci/bytes"
 )
@@ -55,14 +57,82 @@ func (m *MockNvpci) Cleanup() {
 	os.RemoveAll(m.pciDevicesRoot)
 }
 
+func validatePCIAddress(addr string) error {
+	r := regexp.MustCompile(`0{4}:[0-9a-f]{2}:[0-9a-f]{2}\.[0-9]`)
+	if !r.Match([]byte(addr)) {
+		return fmt.Errorf(`invalid PCI address should match 0{4}:[0-9a-f]{2}:[0-9a-f]{2}\.[0-9]: %s`, addr)
+	}
+
+	return nil
+}
+
 // AddMockA100 Create an A100 like GPU mock device.
-func (m *MockNvpci) AddMockA100(address string, numaNode int) error {
-	deviceDir := filepath.Join(m.pciDevicesRoot, address)
-	err := os.MkdirAll(deviceDir, 0755)
+func (m *MockNvpci) AddMockA100(address string, numaNode int, sriov *SriovInfo) error {
+	err := validatePCIAddress(address)
 	if err != nil {
 		return err
 	}
 
+	deviceDir := filepath.Join(m.pciDevicesRoot, address)
+	err = os.MkdirAll(deviceDir, 0755)
+	if err != nil {
+		return err
+	}
+
+	err = createNVIDIAgpuFiles(deviceDir)
+	if err != nil {
+		return err
+	}
+
+	iommuGroup := 20
+	_, err = os.Create(filepath.Join(deviceDir, strconv.Itoa(iommuGroup)))
+	if err != nil {
+		return err
+	}
+	err = os.Symlink(filepath.Join(deviceDir, strconv.Itoa(iommuGroup)), filepath.Join(deviceDir, "iommu_group"))
+	if err != nil {
+		return err
+	}
+
+	numa, err := os.Create(filepath.Join(deviceDir, "numa_node"))
+	if err != nil {
+		return err
+	}
+	_, err = numa.WriteString(fmt.Sprintf("%v", numaNode))
+	if err != nil {
+		return err
+	}
+
+	if sriov != nil && sriov.PhysicalFunction != nil {
+		totalVFs, err := os.Create(filepath.Join(deviceDir, "sriov_totalvfs"))
+		if err != nil {
+			return err
+		}
+		_, err = fmt.Fprintf(totalVFs, "%d", sriov.PhysicalFunction.TotalVFs)
+		if err != nil {
+			return err
+		}
+
+		numVFs, err := os.Create(filepath.Join(deviceDir, "sriov_numvfs"))
+		if err != nil {
+			return err
+		}
+		_, err = fmt.Fprintf(numVFs, "%d", sriov.PhysicalFunction.NumVFs)
+		if err != nil {
+			return err
+		}
+		for i := 1; i <= int(sriov.PhysicalFunction.NumVFs); i++ {
+			err = m.createVf(address, i, iommuGroup, numaNode)
+			if err != nil {
+				return err
+			}
+		}
+	}
+
+	return nil
+}
+
+func createNVIDIAgpuFiles(deviceDir string) error {
 	vendor, err := os.Create(filepath.Join(deviceDir, "vendor"))
 	if err != nil {
 		return err
@@ -99,24 +169,6 @@ func (m *MockNvpci) AddMockA100(address string, numaNode int) error {
 		return err
 	}
 
-	_, err = os.Create(filepath.Join(deviceDir, "20"))
-	if err != nil {
-		return err
-	}
-	err = os.Symlink(filepath.Join(deviceDir, "20"), filepath.Join(deviceDir, "iommu_group"))
-	if err != nil {
-		return err
-	}
-
-	numa, err := os.Create(filepath.Join(deviceDir, "numa_node"))
-	if err != nil {
-		return err
-	}
-	_, err = numa.WriteString(fmt.Sprintf("%v", numaNode))
-	if err != nil {
-		return err
-	}
-
 	config, err := os.Create(filepath.Join(deviceDir, "config"))
 	if err != nil {
 		return err
@@ -156,3 +208,53 @@ func (m *MockNvpci) AddMockA100(address string, numaNode int) error {
 
 	return nil
 }
+
+func (m *MockNvpci) createVf(pfAddress string, id, iommu_group, numaNode int) error {
+	functionID := pfAddress[len(pfAddress)-1]
+	// we are verifying the last character of pfAddress is integer.
+	functionNumber, err := strconv.Atoi(string(functionID))
+	if err != nil {
+		return fmt.Errorf("can't conver physical function pci address function number %s to integer: %v", string(functionID), err)
+	}
+
+	vfFunctionNumber := functionNumber + id
+	vfAddress := pfAddress[:len(pfAddress)-1] + strconv.Itoa(vfFunctionNumber)
+
+	deviceDir := filepath.Join(m.pciDevicesRoot, vfAddress)
+	err = os.MkdirAll(deviceDir, 0755)
+	if err != nil {
+		return err
+	}
+
+	err = createNVIDIAgpuFiles(deviceDir)
+	if err != nil {
+		return err
+	}
+
+	vfIommuGroup := strconv.Itoa(iommu_group + id)
+
+	_, err = os.Create(filepath.Join(deviceDir, vfIommuGroup))
+	if err != nil {
+		return err
+	}
+	err = os.Symlink(filepath.Join(deviceDir, vfIommuGroup), filepath.Join(deviceDir, "iommu_group"))
+	if err != nil {
+		return err
+	}
+
+	numa, err := os.Create(filepath.Join(deviceDir, "numa_node"))
+	if err != nil {
+		return err
+	}
+	_, err = numa.WriteString(fmt.Sprintf("%v", numaNode))
+	if err != nil {
+		return err
+	}
+
+	err = os.Symlink(filepath.Join(m.pciDevicesRoot, pfAddress), filepath.Join(deviceDir, "physfn"))
+	if err != nil {
+		return err
+	}
+
+	return nil
+}

vendor/github.com/NVIDIA/go-nvlib/pkg/nvpci/nvpci.go

@@ -76,6 +76,32 @@ type nvpci struct {
 var _ Interface = (*nvpci)(nil)
 var _ ResourceInterface = (*MemoryResources)(nil)
 
+// SriovInfo indicates whether device is VF/PF for SRIOV capable devices.
+// Only one should be set at any given time.
+type SriovInfo struct {
+	PhysicalFunction *SriovPhysicalFunction
+	VirtualFunction  *SriovVirtualFunction
+}
+
+// SriovPhysicalFunction stores info about SRIOV physical function.
+type SriovPhysicalFunction struct {
+	TotalVFs uint64
+	NumVFs   uint64
+}
+
+// SriovVirtualFunction keeps data about SRIOV virtual function.
+type SriovVirtualFunction struct {
+	PhysicalFunction *NvidiaPCIDevice
+}
+
+func (s *SriovInfo) IsPF() bool {
+	return s != nil && s.PhysicalFunction != nil
+}
+
+func (s *SriovInfo) IsVF() bool {
+	return s != nil && s.VirtualFunction != nil
+}
+
 // NvidiaPCIDevice represents a PCI device for an NVIDIA product.
 type NvidiaPCIDevice struct {
 	Path       string
@@ -90,7 +116,7 @@ type NvidiaPCIDevice struct {
 	NumaNode   int
 	Config     *ConfigSpace
 	Resources  MemoryResources
-	IsVF       bool
+	SriovInfo  SriovInfo
 }
 
 // IsVGAController if class == 0x300.
@@ -178,9 +204,11 @@ func (p *nvpci) GetAllDevices() ([]*NvidiaPCIDevice, error) {
 	}
 
 	var nvdevices []*NvidiaPCIDevice
+	// Cache devices for each GetAllDevices invocation to speed things up.
+	cache := make(map[string]*NvidiaPCIDevice)
 	for _, deviceDir := range deviceDirs {
 		deviceAddress := deviceDir.Name()
-		nvdevice, err := p.GetGPUByPciBusID(deviceAddress)
+		nvdevice, err := p.getGPUByPciBusID(deviceAddress, cache)
 		if err != nil {
 			return nil, fmt.Errorf("error constructing NVIDIA PCI device %s: %v", deviceAddress, err)
 		}
@@ -206,6 +234,16 @@ func (p *nvpci) GetAllDevices() ([]*NvidiaPCIDevice, error) {
 
 // GetGPUByPciBusID constructs an NvidiaPCIDevice for the specified address (PCI Bus ID).
 func (p *nvpci) GetGPUByPciBusID(address string) (*NvidiaPCIDevice, error) {
+	// Pass nil as to force reading device information from sysfs.
+	return p.getGPUByPciBusID(address, nil)
+}
+
+func (p *nvpci) getGPUByPciBusID(address string, cache map[string]*NvidiaPCIDevice) (*NvidiaPCIDevice, error) {
+	if cache != nil {
+		if pciDevice, exists := cache[address]; exists {
+			return pciDevice, nil
+		}
+	}
 	devicePath := filepath.Join(p.pciDevicesRoot, address)
 
 	vendor, err := os.ReadFile(path.Join(devicePath, "vendor"))
@@ -265,16 +303,6 @@ func (p *nvpci) GetGPUByPciBusID(address string) (*NvidiaPCIDevice, error) {
 		return nil, fmt.Errorf("unable to detect iommu_group for %s: %v", address, err)
 	}
 
-	// device is a virtual function (VF) if "physfn" symlink exists.
-	var isVF bool
-	_, err = filepath.EvalSymlinks(path.Join(devicePath, "physfn"))
-	if err == nil {
-		isVF = true
-	}
-	if err != nil && !os.IsNotExist(err) {
-		return nil, fmt.Errorf("unable to resolve %s: %v", path.Join(devicePath, "physfn"), err)
-	}
-
 	numa, err := os.ReadFile(path.Join(devicePath, "numa_node"))
 	if err != nil {
 		return nil, fmt.Errorf("unable to read PCI NUMA node for %s: %v", address, err)
@@ -328,6 +356,28 @@ func (p *nvpci) GetGPUByPciBusID(address string) (*NvidiaPCIDevice, error) {
 		className = UnknownClassString
 	}
 
+	var sriovInfo SriovInfo
+	// Device is a virtual function (VF) if "physfn" symlink exists.
+	physFnAddress, err := filepath.EvalSymlinks(path.Join(devicePath, "physfn"))
+	if err == nil {
+		physFn, err := p.getGPUByPciBusID(filepath.Base(physFnAddress), cache)
+		if err != nil {
+			return nil, fmt.Errorf("unable to detect physfn for %s: %v", address, err)
+		}
+		sriovInfo = SriovInfo{
+			VirtualFunction: &SriovVirtualFunction{
+				PhysicalFunction: physFn,
+			},
+		}
+	} else if os.IsNotExist(err) {
+		sriovInfo, err = p.getSriovInfoForPhysicalFunction(devicePath)
+		if err != nil {
+			return nil, fmt.Errorf("unable to read SRIOV physical function details for %s: %v", devicePath, err)
+		}
+	} else {
+		return nil, fmt.Errorf("unable to read %s: %v", path.Join(devicePath, "physfn"), err)
+	}
+
 	nvdevice := &NvidiaPCIDevice{
 		Path:       devicePath,
 		Address:    address,
@@ -339,9 +389,14 @@ func (p *nvpci) GetGPUByPciBusID(address string) (*NvidiaPCIDevice, error) {
 		NumaNode:   int(numaNode),
 		Config:     config,
 		Resources:  resources,
-		IsVF:       isVF,
 		DeviceName: deviceName,
 		ClassName:  className,
+		SriovInfo:  sriovInfo,
+	}
+
+	// Cache physical functions only as VF can't be a root device.
+	if cache != nil && sriovInfo.IsPF() {
+		cache[address] = nvdevice
 	}
 
 	return nvdevice, nil
@@ -407,7 +462,7 @@ func (p *nvpci) GetGPUs() ([]*NvidiaPCIDevice, error) {
 
 	var filtered []*NvidiaPCIDevice
 	for _, d := range devices {
-		if d.IsGPU() && !d.IsVF {
+		if d.IsGPU() && !d.SriovInfo.IsVF() {
 			filtered = append(filtered, d)
 		}
 	}
@@ -428,3 +483,41 @@ func (p *nvpci) GetGPUByIndex(i int) (*NvidiaPCIDevice, error) {
 
 	return gpus[i], nil
 }
+
+func (p *nvpci) getSriovInfoForPhysicalFunction(devicePath string) (sriovInfo SriovInfo, err error) {
+	totalVfsPath := filepath.Join(devicePath, "sriov_totalvfs")
+	numVfsPath := filepath.Join(devicePath, "sriov_numvfs")
+
+	// No file for sriov_totalvfs exists? Not an SRIOV device, return nil
+	_, err = os.Stat(totalVfsPath)
+	if err != nil && os.IsNotExist(err) {
+		return sriovInfo, nil
+	}
+	sriovTotalVfs, err := os.ReadFile(totalVfsPath)
+	if err != nil {
+		return sriovInfo, fmt.Errorf("unable to read sriov_totalvfs: %v", err)
+	}
+	totalVfsStr := strings.TrimSpace(string(sriovTotalVfs))
+	totalVfsInt, err := strconv.ParseUint(totalVfsStr, 10, 16)
+	if err != nil {
+		return sriovInfo, fmt.Errorf("unable to convert sriov_totalvfs to uint64: %v", err)
+	}
+
+	sriovNumVfs, err := os.ReadFile(numVfsPath)
+	if err != nil {
+		return sriovInfo, fmt.Errorf("unable to read sriov_numvfs for: %v", err)
+	}
+	numVfsStr := strings.TrimSpace(string(sriovNumVfs))
+	numVfsInt, err := strconv.ParseUint(numVfsStr, 10, 16)
+	if err != nil {
+		return sriovInfo, fmt.Errorf("unable to convert sriov_numvfs to uint64: %v", err)
+	}
+
+	sriovInfo = SriovInfo{
+		PhysicalFunction: &SriovPhysicalFunction{
+			TotalVFs: totalVfsInt,
+			NumVFs:   numVfsInt,
+		},
+	}
+	return sriovInfo, nil
+}

vendor/modules.txt

@@ -1,4 +1,4 @@
-# github.com/NVIDIA/go-nvlib v0.5.0
+# github.com/NVIDIA/go-nvlib v0.6.0
 ## explicit; go 1.20
 github.com/NVIDIA/go-nvlib/pkg/nvlib/device
 github.com/NVIDIA/go-nvlib/pkg/nvlib/info