Detect GPUs (#121)

.github/workflows/master.yml

@@ -32,6 +32,6 @@ jobs:
           mv metal-hammer-initrd.img.lz4* images/metal-hammer/
       - name: Upload image tarballs to GCS
         run: gsutil -m cp -r -p images/metal-hammer gs://$GCS_BUCKET
-      - uses: release-drafter/release-drafter@v5
+      - uses: release-drafter/release-drafter@v6
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

Dockerfile

@@ -22,7 +22,7 @@ RUN curl -fLsS https://sourceforge.net/projects/e1000/files/ice%20stable/${ICE_V
 
 # ipmitool from bookworm is broken and returns with error on most commands
 FROM golang:1.22-bullseye as initrd-builder
-ENV UROOT_GIT_SHA_OR_TAG=v0.13.0
+ENV UROOT_GIT_SHA_OR_TAG=v0.14.0
 RUN apt-get update \
  && apt-get install -y --no-install-recommends \
 	ca-certificates \

REINSTALL.md

@@ -13,8 +13,8 @@ If only the `imageID` is given it tries to guess the primary disk of the old OS.
 After wiping the primary disk the reinstall procedure continues with the usual installation process up from the `installImage` method that eventually ends with the `finalizeAllocation` call, which now includes the previous mentioned `BootInfo` parameters.
 
 **metal-core** passes-through the request to **metal-api**, sets the boot order to HD and power cycles the machine again, which in turn boots the new OS.
- 
-**metal-api** removes the `allocation.Reinstall` mark and stores the `BootInfo` details together with the newly installed `imageID` in the `allcation.MachineSetup` struct.
+
+**metal-api** removes the `allocation.Reinstall` mark and stores the `BootInfo` details together with the newly installed `imageID` in the `allocation.MachineSetup` struct.
 
 This was the happy-path. But of course, things can go wrong. If for any reason the reinstallation process fails, we are potentially in one of the following two states: Either the primary disk has been wiped already (and therewith the existing OS) or not. In both cases **metal-hammer** calls **metal-core** via the `/machine/abort-reinstall/<id>` endpoint delivering the bool value `primaryDiskWiped` that indicates the actual state.  
 If **metal-core** fails to respond or the OS has already been wiped the machine reboots. Otherwise it gets the `BootInfo` of the previous installed OS stored in the DS and reboots with these details into the existing OS, just as nothing had happened at all.  

cmd/firmware/firmware.go

@@ -61,7 +61,7 @@ func (f *Firmware) Update() {
 	}
 }
 
-// Run execute a comand with arguments, returns output and error
+// Run execute a command with arguments, returns output and error
 func run(log *slog.Logger, command string, args ...string) (string, error) {
 	path, err := exec.LookPath(command)
 	if err != nil {

cmd/firmware/intel.go

@@ -22,7 +22,7 @@ func (r intel) update() error {
 	if err != nil {
 		return fmt.Errorf("unable to update intel firmware %w", err)
 	}
-	r.log.Info("intel", "updated firware output", output)
+	r.log.Info("intel", "updated firmware output", output)
 	return nil
 }
 

cmd/metal-client.go

@@ -10,6 +10,7 @@ import (
 
 	v1 "github.com/metal-stack/metal-api/pkg/api/v1"
 	metalgo "github.com/metal-stack/metal-go"
+	"github.com/metal-stack/metal-go/api/client/machine"
 	"google.golang.org/grpc"
 	"google.golang.org/grpc/credentials"
 	"google.golang.org/grpc/keepalive"
@@ -18,7 +19,7 @@ import (
 type MetalAPIClient struct {
 	log    *slog.Logger
 	conn   grpc.ClientConnInterface
-	Driver metalgo.Client
+	driver metalgo.Client
 }
 
 // NewMetalAPIClient fetches the address,hmac and certificates from pixie needed to communicate with metal-api,
@@ -71,10 +72,12 @@ func NewMetalAPIClient(log *slog.Logger, spec *Specification) (*MetalAPIClient,
 	return &MetalAPIClient{
 		log:    log,
 		conn:   conn,
-		Driver: driver,
+		driver: driver,
 	}, nil
 }
-
+func (c *MetalAPIClient) Machine() machine.ClientService {
+	return c.driver.Machine()
+}
 func (c *MetalAPIClient) Event() v1.EventServiceClient {
 	return v1.NewEventServiceClient(c.conn)
 }

cmd/network/ethtool.go

@@ -15,7 +15,7 @@ import (
 	"github.com/metal-stack/metal-hammer/pkg/os/command"
 )
 
-// EthtoolCommand to gather ethernet informations
+// EthtoolCommand to gather ethernet information
 const ethtoolCommand = command.Ethtool
 
 // Ethtool to query/set ethernet interfaces
@@ -89,7 +89,7 @@ func (e *Ethtool) disableFirmwareLLDP(ifi string) {
 
 var buggyIntelNicDriverNames = []string{"i40e"}
 
-// stopFirmwareLLDP stop Firmeware LLDP not persistent over reboots, only during runtime.
+// stopFirmwareLLDP stop Firmware LLDP not persistent over reboots, only during runtime.
 // mount -t debugfs none /sys/kernel/debug
 // echo lldp stop > /sys/kernel/debug/i40e/0000:01:00.2/command
 // where <0000:01:00.2> is the pci address of the ethernet nic, this can be inspected by lspci,

cmd/network/ntpdate.go

@@ -33,7 +33,7 @@ func getTime(log *slog.Logger, servers []string) (t time.Time, err error) {
 	return
 }
 
-// NtpDate set the system time to the time comming from a ntp source
+// NtpDate set the system time to the time coming from a ntp source
 func NtpDate(log *slog.Logger) {
 	t, err := getTime(log, ntpServers)
 	if err != nil {

cmd/register/register.go

@@ -20,6 +20,7 @@ import (
 	"github.com/metal-stack/metal-hammer/cmd/network"
 	"github.com/metal-stack/metal-hammer/cmd/storage"
 	"github.com/metal-stack/v"
+	"github.com/u-root/u-root/pkg/pci"
 	"github.com/vishvananda/netlink"
 )
 
@@ -81,6 +82,32 @@ func (r *Register) readHardwareDetails() (*v1.BootServiceRegisterRequest, error)
 	if err != nil {
 		return nil, fmt.Errorf("unable to get system cpu(s) %w", err)
 	}
+	r.log.Info("cpu", "processors", cpu.String())
+	var metalCPUs []*v1.MachineCPU
+	for _, cpu := range cpu.Processors {
+		metalCPUs = append(metalCPUs, &v1.MachineCPU{
+			Vendor:  cpu.Vendor,
+			Model:   cpu.Model,
+			Cores:   cpu.NumCores,
+			Threads: cpu.NumThreads,
+		})
+	}
+
+	// 0000:bd:00.0: DisplayVGA: NVIDIA Corporation AD102GL [RTX 6000 Ada Generation]
+
+	gpus, err := r.detectGPUs()
+	if err != nil {
+		return nil, fmt.Errorf("unable to get system gpu(s) %w", err)
+	}
+
+	var metalGPUs []*v1.MachineGPU
+	for _, g := range gpus {
+		r.log.Info("found gpu", "gpu", g.String())
+		metalGPUs = append(metalGPUs, &v1.MachineGPU{
+			Vendor: g.VendorName,
+			Model:  g.DeviceName,
+		})
+	}
 
 	// Nics
 	nics := []*v1.MachineNic{}
@@ -166,6 +193,8 @@ func (r *Register) readHardwareDetails() (*v1.BootServiceRegisterRequest, error)
 		CpuCores: uint32(cpu.TotalCores),
 		Nics:     nics,
 		Disks:    disks,
+		Cpus:     metalCPUs,
+		Gpus:     metalGPUs,
 	}
 
 	// IPMI
@@ -178,7 +207,7 @@ func (r *Register) readHardwareDetails() (*v1.BootServiceRegisterRequest, error)
 	board := r.inband.Board()
 	b := board.BIOS
 	if b == nil {
-		return nil, fmt.Errorf("unable to read bios informations from bmc")
+		return nil, fmt.Errorf("unable to read bios information from bmc")
 	}
 	bios := &v1.MachineBIOS{
 		Version: b.Version,
@@ -198,6 +227,36 @@ func (r *Register) readHardwareDetails() (*v1.BootServiceRegisterRequest, error)
 	return request, nil
 }
 
+func (r *Register) detectGPUs() (pci.Devices, error) {
+	pciReader, err := pci.NewBusReader("*")
+	if err != nil {
+		return nil, err
+	}
+
+	var devices pci.Devices
+	if devices, err = pciReader.Read(); err != nil {
+		return nil, err
+	}
+
+	devices.SetVendorDeviceName()
+
+	var result pci.Devices
+	for _, device := range devices {
+		// "vendor":"NVIDIA Corporation","device":"AD102GL [RTX 6000 Ada Generation]"}
+		if !strings.Contains(strings.ToLower(device.VendorName), "nvidia") {
+			continue
+		}
+
+		// TODO if new models must be supported, this code must be refactored
+		if strings.Contains(strings.ToLower(device.DeviceName), "rtx") {
+			r.log.Info("add gpu", "vendor", device.VendorName, "device", device.DeviceName)
+			result = append(result, device)
+		}
+	}
+
+	return result, nil
+}
+
 // save the content of kernel ring buffer to /var/log/syslog
 // by calling the appropriate syscall.
 // Only required if Memory is gathered by ghw.Memory()

cmd/root.go

@@ -106,7 +106,7 @@ func Run(log *slog.Logger, spec *Specification, hal hal.InBand) (*event.EventEmi
 		return eventEmitter, fmt.Errorf("register %w", err)
 	}
 
-	resp, err := metalAPIClient.Driver.Machine().FindMachine(machine.NewFindMachineParams().WithID(spec.MachineUUID), nil)
+	resp, err := metalAPIClient.Machine().FindMachine(machine.NewFindMachineParams().WithID(spec.MachineUUID), nil)
 	if err != nil {
 		return eventEmitter, fmt.Errorf("fetch %w", err)
 	}
@@ -143,7 +143,7 @@ func Run(log *slog.Logger, spec *Specification, hal hal.InBand) (*event.EventEmi
 	if err != nil {
 		return eventEmitter, fmt.Errorf("wait for installation %w", err)
 	}
-	resp, err = metalAPIClient.Driver.Machine().FindMachine(machine.NewFindMachineParams().WithID(spec.MachineUUID), nil)
+	resp, err = metalAPIClient.Machine().FindMachine(machine.NewFindMachineParams().WithID(spec.MachineUUID), nil)
 	if err != nil {
 		return eventEmitter, fmt.Errorf("wait for installation %w", err)
 	}

cmd/supwd.go

@@ -16,15 +16,21 @@ func (h *Hammer) createBmcSuperuser() error {
 		return fmt.Errorf("failed to fetch SuperUser password %w", err)
 	}
 
-	if resp.FeatureDisabled {
+	if resp.SuperUserPassword == "" {
+		h.log.Warn("creation of superuser disabled because password is empty")
 		return nil
 	}
 
 	bmcConn := h.Hal.BMCConnection()
+
 	err = bmcConn.CreateUser(bmcConn.SuperUser(), api.AdministratorPrivilege, resp.SuperUserPassword)
 	if err != nil {
-		return fmt.Errorf("failed to create bmc superuser: %s %w", bmcConn.SuperUser().Name, err)
+		// FIXME: this happens always after the first creation on X12 and newer boards
+		// return fmt.Errorf("failed to create bmc superuser: %s %w", bmcConn.SuperUser().Name, err)
+		h.log.Error("failed to create bmc superuser", "user", bmcConn.SuperUser().Name, "error", err)
+		return nil
 	}
 
+	h.log.Info("created superuser", "user", bmcConn.SuperUser().Name)
 	return nil
 }