cli: feat amdgpu install

cli/cmd/ctl/amdgpu/install.go

@@ -0,0 +1,21 @@
+package amdgpu
+
+import (
+	"log"
+
+	"github.com/beclab/Olares/cli/pkg/pipelines"
+	"github.com/spf13/cobra"
+)
+
+func NewCmdAmdGpuInstall() *cobra.Command {
+	cmd := &cobra.Command{
+		Use:   "install",
+		Short: "Install AMD ROCm stack via amdgpu-install",
+		Run: func(cmd *cobra.Command, args []string) {
+			if err := pipelines.AmdGpuInstall(); err != nil {
+				log.Fatalf("error: %v", err)
+			}
+		},
+	}
+	return cmd
+}

cli/cmd/ctl/amdgpu/root.go

@@ -0,0 +1,16 @@
+package amdgpu
+
+import "github.com/spf13/cobra"
+
+func NewCmdAmdGpu() *cobra.Command {
+	cmd := &cobra.Command{
+		Use:   "amdgpu",
+		Short: "Manage AMD GPU ROCm stack",
+	}
+	cmd.AddCommand(NewCmdAmdGpuInstall())
+	cmd.AddCommand(NewCmdAmdGpuUninstall())
+	cmd.AddCommand(NewCmdAmdGpuStatus())
+	return cmd
+}
+
+

cli/cmd/ctl/amdgpu/status.go

@@ -0,0 +1,21 @@
+package amdgpu
+
+import (
+	"log"
+
+	"github.com/beclab/Olares/cli/pkg/pipelines"
+	"github.com/spf13/cobra"
+)
+
+func NewCmdAmdGpuStatus() *cobra.Command {
+	cmd := &cobra.Command{
+		Use:   "status",
+		Short: "Show AMD GPU driver and ROCm status",
+		Run: func(cmd *cobra.Command, args []string) {
+			if err := pipelines.AmdGpuStatus(); err != nil {
+				log.Fatalf("error: %v", err)
+			}
+		},
+	}
+	return cmd
+}

cli/cmd/ctl/amdgpu/uninstall.go

@@ -0,0 +1,21 @@
+package amdgpu
+
+import (
+	"log"
+
+	"github.com/beclab/Olares/cli/pkg/pipelines"
+	"github.com/spf13/cobra"
+)
+
+func NewCmdAmdGpuUninstall() *cobra.Command {
+	cmd := &cobra.Command{
+		Use:   "uninstall",
+		Short: "Uninstall AMD ROCm stack via amdgpu-install",
+		Run: func(cmd *cobra.Command, args []string) {
+			if err := pipelines.AmdGpuUninstall(); err != nil {
+				log.Fatalf("error: %v", err)
+			}
+		},
+	}
+	return cmd
+}

cli/cmd/ctl/root.go

@@ -1,6 +1,7 @@
 package ctl
 
 import (
+	"github.com/beclab/Olares/cli/cmd/ctl/amdgpu"
 	"github.com/beclab/Olares/cli/cmd/ctl/disk"
 	"github.com/beclab/Olares/cli/cmd/ctl/gpu"
 	"github.com/beclab/Olares/cli/cmd/ctl/node"
@@ -33,6 +34,7 @@ func NewDefaultCommand() *cobra.Command {
 	cmds.AddCommand(os.NewOSCommands()...)
 	cmds.AddCommand(node.NewNodeCommand())
 	cmds.AddCommand(gpu.NewCmdGpu())
+	cmds.AddCommand(amdgpu.NewCmdAmdGpu())
 	cmds.AddCommand(user.NewUserCommand())
 	cmds.AddCommand(disk.NewDiskCommand())
 

cli/pkg/amdgpu/tasks.go

@@ -0,0 +1,133 @@
+package amdgpu
+
+import (
+	"fmt"
+	"os/exec"
+	"path"
+	"path/filepath"
+
+	"github.com/beclab/Olares/cli/pkg/common"
+	cc "github.com/beclab/Olares/cli/pkg/core/common"
+	"github.com/beclab/Olares/cli/pkg/core/connector"
+	"github.com/beclab/Olares/cli/pkg/core/logger"
+	"github.com/beclab/Olares/cli/pkg/core/task"
+	"github.com/beclab/Olares/cli/pkg/utils"
+
+	"github.com/Masterminds/semver/v3"
+	"github.com/pkg/errors"
+)
+
+// InstallAmdRocmModule installs AMD ROCm stack on supported Ubuntu if AMD GPU is present.
+type InstallAmdRocmModule struct {
+	common.KubeModule
+}
+
+func (m *InstallAmdRocmModule) Init() {
+	m.Name = "InstallAMDGPU"
+
+	installAmd := &task.RemoteTask{
+		Name:  "InstallAmdRocm",
+		Hosts: m.Runtime.GetHostsByRole(common.Master),
+		Action: &InstallAmdRocm{
+			// no manifest needed
+		},
+		Parallel: false,
+		Retry:    1,
+	}
+
+	m.Tasks = []task.Interface{
+		installAmd,
+	}
+}
+
+// InstallAmdRocm installs ROCm using amdgpu-install on Ubuntu 22.04/24.04 for AMD GPUs.
+type InstallAmdRocm struct {
+	common.KubeAction
+}
+
+func (t *InstallAmdRocm) Execute(runtime connector.Runtime) error {
+	si := runtime.GetSystemInfo()
+	if !si.IsLinux() || !si.IsUbuntu() || !(si.IsUbuntuVersionEqual(connector.Ubuntu2204) || si.IsUbuntuVersionEqual(connector.Ubuntu2404)) {
+		return nil
+	}
+
+	amdGPUExists, err := utils.HasAmdIGPU(runtime)
+	if err != nil {
+		return err
+	}
+	// skip rocm install
+	if !amdGPUExists {
+		return nil
+	}
+	rocmV, _ := utils.RocmVersion()
+	min := semver.MustParse("7.1.1")
+	if rocmV != nil && rocmV.LessThan(min) {
+		return fmt.Errorf("detected ROCm version %s, which is lower than required %s; please uninstall existing ROCm/AMDGPU components before installation with command: olares-cli amdgpu uninstall", rocmV.Original(), min.Original())
+	}
+	if rocmV != nil && rocmV.GreaterThan(min) {
+		logger.Warnf("Warning: detected ROCm version %s great than maximum tested version %s")
+		return nil
+	}
+	if rocmV != nil && rocmV.Equal(min) {
+		logger.Infof("detected ROCm version %s, skip rocm install...", min.Original())
+		return nil
+	}
+
+	// ensure python3-setuptools and python3-wheel
+	_, _ = runtime.GetRunner().SudoCmd("apt-get update", false, true)
+	checkPkgs := "dpkg -s python3-setuptools python3-wheel >/dev/null 2>&1 || DEBIAN_FRONTEND=noninteractive apt-get install -y python3-setuptools python3-wheel"
+	if _, err := runtime.GetRunner().SudoCmd(checkPkgs, false, true); err != nil {
+		return errors.Wrap(errors.WithStack(err), "failed to install python3-setuptools and python3-wheel")
+	}
+	// ensure amdgpu-install exists
+	if _, err := exec.LookPath("amdgpu-install"); err != nil {
+		var debURL string
+		if si.IsUbuntuVersionEqual(connector.Ubuntu2404) {
+			debURL = "https://repo.radeon.com/amdgpu-install/7.1.1/ubuntu/noble/amdgpu-install_7.1.1.70101-1_all.deb"
+		} else {
+			debURL = "https://repo.radeon.com/amdgpu-install/7.1.1/ubuntu/jammy/amdgpu-install_7.1.1.70101-1_all.deb"
+		}
+		tmpDeb := path.Join(runtime.GetBaseDir(), cc.PackageCacheDir, "gpu", "amdgpu-install_7.1.1.70101-1_all.deb")
+		if _, err := runtime.GetRunner().SudoCmd(fmt.Sprintf("install -d -m 0755 %s", filepath.Dir(tmpDeb)), false, true); err != nil {
+			return err
+		}
+		cmd := fmt.Sprintf("sh -c 'wget -O %s %s'", tmpDeb, debURL)
+		if _, err := runtime.GetRunner().SudoCmd(cmd, false, true); err != nil {
+			return errors.Wrap(errors.WithStack(err), "failed to download amdgpu-install deb")
+		}
+		if _, err := runtime.GetRunner().SudoCmd(fmt.Sprintf("DEBIAN_FRONTEND=noninteractive apt-get install -y %s", tmpDeb), false, true); err != nil {
+			return errors.Wrap(errors.WithStack(err), "failed to install amdgpu-install deb")
+		}
+	}
+	// run installer for rocm usecase
+	if _, err := runtime.GetRunner().SudoCmd("amdgpu-install -y --usecase=rocm", false, true); err != nil {
+		return errors.Wrap(errors.WithStack(err), "failed to install AMD ROCm via amdgpu-install")
+	}
+	fmt.Println()
+	logger.Warn("Warning: To enable ROCm, please reboot your machine after installation.")
+	return nil
+}
+
+type AmdgpuInstallAction struct {
+	common.KubeAction
+}
+
+func (t *AmdgpuInstallAction) Execute(runtime connector.Runtime) error {
+	if _, err := runtime.GetRunner().SudoCmd("amdgpu-install -y --usecase=rocm", false, true); err != nil {
+		return errors.Wrap(errors.WithStack(err), "failed to install AMD ROCm via amdgpu-install")
+	}
+	return nil
+}
+
+type AmdgpuUninstallAction struct {
+	common.KubeAction
+}
+
+func (t *AmdgpuUninstallAction) Execute(runtime connector.Runtime) error {
+	if _, err := runtime.GetRunner().SudoCmd("amdgpu-install --uninstall -y", false, true); err != nil {
+		return errors.Wrap(errors.WithStack(err), "failed to uninstall AMD ROCm via amdgpu-install")
+	}
+	fmt.Println()
+	logger.Warn("Warning: Please reboot your machine after uninstall to fully remove ROCm components.")
+	return nil
+}

cli/pkg/bootstrap/precheck/module.go

@@ -81,6 +81,7 @@ func (m *RunPrechecksModule) Init() {
 		new(NvidiaCardArchChecker),
 		new(NouveauChecker),
 		new(CudaChecker),
+		new(RocmChecker),
 	}
 	runPreChecks := &task.LocalTask{
 		Name: "RunPrechecks",

cli/pkg/bootstrap/precheck/tasks.go

@@ -372,6 +372,48 @@ func (c *CudaChecker) Check(runtime connector.Runtime) error {
 	return nil
 }
 
+// RocmChecker checks AMD ROCm version for AMD GPU on Ubuntu 22.04/24.04 only.
+type RocmChecker struct{}
+
+func (r *RocmChecker) Name() string {
+	return "ROCm"
+}
+
+func (r *RocmChecker) Check(runtime connector.Runtime) error {
+	if !runtime.GetSystemInfo().IsLinux() {
+		return nil
+	}
+	si := runtime.GetSystemInfo()
+	if !si.IsUbuntu() || !(si.IsUbuntuVersionEqual(connector.Ubuntu2204) || si.IsUbuntuVersionEqual(connector.Ubuntu2404)) {
+		return nil
+	}
+
+	// detect AMD GPU presence
+	amdGPUExists, err := utils.HasAmdIGPU(runtime)
+	if err != nil {
+		return err
+	}
+	// no AMD GPU found, no need to check rocm
+	if !amdGPUExists {
+		return nil
+	}
+
+	curV, err := utils.RocmVersion()
+	if err != nil && !os.IsNotExist(err) {
+		return err
+	}
+
+	if os.IsNotExist(err) {
+		return nil
+	}
+
+	min := semver.MustParse("7.1.1")
+	if curV.LessThan(min) {
+		return fmt.Errorf("detected ROCm version %s, which is lower than required %s; please uninstall existing ROCm/AMDGPU components before installation with command: olares-cli amdgpu uninstall", curV.Original(), min.Original())
+	}
+	return nil
+}
+
 //////////////////////////////////////////////
 // precheck - task
 

cli/pkg/core/connector/system.go

@@ -51,10 +51,12 @@ func (d DebianVersion) String() string {
 }
 
 const (
-	Ubuntu20 UbuntuVersion = "20."
-	Ubuntu22 UbuntuVersion = "22."
-	Ubuntu24 UbuntuVersion = "24."
-	Ubuntu25 UbuntuVersion = "25."
+	Ubuntu20   UbuntuVersion = "20."
+	Ubuntu22   UbuntuVersion = "22."
+	Ubuntu24   UbuntuVersion = "24."
+	Ubuntu25   UbuntuVersion = "25."
+	Ubuntu2204 UbuntuVersion = "22.04"
+	Ubuntu2404 UbuntuVersion = "24.04"
 
 	Debian9  DebianVersion = "9"
 	Debian10 DebianVersion = "10"

cli/pkg/phase/system/linux.go

@@ -3,8 +3,7 @@ package system
 import (
 	"strings"
 
-	"github.com/beclab/Olares/cli/pkg/gpu"
-
+	"github.com/beclab/Olares/cli/pkg/amdgpu"
 	"github.com/beclab/Olares/cli/pkg/bootstrap/os"
 	"github.com/beclab/Olares/cli/pkg/bootstrap/patch"
 	"github.com/beclab/Olares/cli/pkg/bootstrap/precheck"
@@ -12,6 +11,7 @@ import (
 	"github.com/beclab/Olares/cli/pkg/container"
 	"github.com/beclab/Olares/cli/pkg/core/module"
 	"github.com/beclab/Olares/cli/pkg/daemon"
+	"github.com/beclab/Olares/cli/pkg/gpu"
 	"github.com/beclab/Olares/cli/pkg/images"
 	"github.com/beclab/Olares/cli/pkg/k3s"
 	"github.com/beclab/Olares/cli/pkg/manifest"
@@ -82,6 +82,7 @@ func (l *linuxPhaseBuilder) build() []module.Module {
 		addModule(&terminus.WriteReleaseFileModule{}).
 		addModule(gpuModuleBuilder(func() []module.Module {
 			return []module.Module{
+				&amdgpu.InstallAmdRocmModule{},
 				&gpu.InstallDriversModule{
 					ManifestModule: manifest.ManifestModule{
 						Manifest: l.manifestMap,

cli/pkg/pipelines/amdgpu.go

@@ -0,0 +1,101 @@
+package pipelines
+
+import (
+	"strings"
+
+	"github.com/beclab/Olares/cli/pkg/amdgpu"
+	"github.com/beclab/Olares/cli/pkg/common"
+	"github.com/beclab/Olares/cli/pkg/core/action"
+	"github.com/beclab/Olares/cli/pkg/core/connector"
+	"github.com/beclab/Olares/cli/pkg/core/logger"
+	"github.com/beclab/Olares/cli/pkg/core/module"
+	"github.com/beclab/Olares/cli/pkg/core/pipeline"
+	"github.com/beclab/Olares/cli/pkg/core/task"
+)
+
+type singleTaskModule struct {
+	common.KubeModule
+	name string
+	act  action.Action
+}
+
+func (m *singleTaskModule) Init() {
+	m.Name = m.name
+	m.Tasks = []task.Interface{
+		&task.LocalTask{
+			Name:   m.name,
+			Action: m.act,
+		},
+	}
+}
+
+func AmdGpuInstall() error {
+	arg := common.NewArgument()
+	arg.SetConsoleLog("amdgpuinstall.log", true)
+	runtime, err := common.NewKubeRuntime(common.AllInOne, *arg)
+	if err != nil {
+		return err
+	}
+	p := &pipeline.Pipeline{
+		Name:    "InstallAMDGPUDrivers",
+		Runtime: runtime,
+		Modules: []module.Module{
+			&amdgpu.InstallAmdRocmModule{},
+		},
+	}
+	return p.Start()
+}
+
+func AmdGpuUninstall() error {
+	arg := common.NewArgument()
+	arg.SetConsoleLog("amdgpuuninstall.log", true)
+	runtime, err := common.NewKubeRuntime(common.AllInOne, *arg)
+	if err != nil {
+		return err
+	}
+	p := &pipeline.Pipeline{
+		Name:    "UninstallAMDGPUDrivers",
+		Runtime: runtime,
+		Modules: []module.Module{
+			&singleTaskModule{name: "AmdgpuUninstall", act: new(amdgpu.AmdgpuUninstallAction)},
+		},
+	}
+	return p.Start()
+}
+
+func AmdGpuStatus() error {
+	arg := common.NewArgument()
+	runtime, err := common.NewKubeRuntime(common.AllInOne, *arg)
+	if err != nil {
+		return err
+	}
+	runtime.SetRunner(
+		&connector.Runner{
+			Host: &connector.BaseHost{
+				Name: common.LocalHost,
+				Arch: runtime.GetSystemInfo().GetOsArch(),
+				Os:   runtime.GetSystemInfo().GetOsType(),
+			},
+		},
+	)
+	amdModel, _ := runtime.GetRunner().SudoCmd("lspci | grep -iE 'VGA|3D|Display' | grep -iE 'AMD|ATI' | head -1 || true", false, false)
+	drvVer, _ := runtime.GetRunner().SudoCmd("modinfo amdgpu 2>/dev/null | awk -F': ' '/^version:/{print $2}' || true", false, false)
+	rocmVer, _ := runtime.GetRunner().SudoCmd("cat /opt/rocm/.info/version 2>/dev/null || true", false, false)
+
+	if strings.TrimSpace(amdModel) != "" {
+		logger.Infof("AMD GPU: %s", strings.TrimSpace(amdModel))
+	} else {
+		logger.Info("AMD GPU: not detected")
+	}
+	if strings.TrimSpace(drvVer) != "" {
+		logger.Infof("AMDGPU driver %s", strings.TrimSpace(drvVer))
+	} else {
+		logger.Info("AMDGPU driver version: unknown")
+	}
+	if strings.TrimSpace(rocmVer) != "" {
+		logger.Infof("ROCm version: %s", strings.TrimSpace(rocmVer))
+	} else {
+		logger.Info("ROCm version: not installed")
+	}
+	return nil
+}

cli/pkg/terminus/welcome.go

@@ -9,6 +9,7 @@ import (
 	"github.com/beclab/Olares/cli/pkg/core/connector"
 	"github.com/beclab/Olares/cli/pkg/core/logger"
 	"github.com/beclab/Olares/cli/pkg/core/task"
+	"github.com/beclab/Olares/cli/pkg/utils"
 )
 
 type WelcomeMessage struct {
@@ -68,6 +69,15 @@ func (t *WelcomeMessage) Execute(runtime connector.Runtime) error {
 	logger.Infof("Username: %s", t.KubeConf.Arg.User.UserName)
 	logger.Infof("Password: %s", t.KubeConf.Arg.User.Password)
 	fmt.Printf("\n------------------------------------------------\n\n\n\n\n")
+	fmt.Println()
+
+	// If AMD GPU on Ubuntu 22.04/24.04, print warning about reboot for ROCm
+	if si := runtime.GetSystemInfo(); si.IsUbuntu() && (si.IsUbuntuVersionEqual(connector.Ubuntu2204) || si.IsUbuntuVersionEqual(connector.Ubuntu2404)) {
+		if hasAmd, _ := utils.HasAmdIGPU(runtime); hasAmd {
+			logger.Warnf("\x1b[31mWarning: To enable ROCm, please reboot your machine after activation.\x1b[0m")
+			fmt.Println()
+		}
+	}
 
 	return nil
 }

cli/pkg/utils/amdgpu.go

@@ -0,0 +1,67 @@
+package utils
+
+import (
+	"fmt"
+	"os"
+	"strings"
+
+	"github.com/Masterminds/semver/v3"
+	"github.com/beclab/Olares/cli/pkg/core/connector"
+)
+
+func HasAmdIGPU(execRuntime connector.Runtime) (bool, error) {
+	// Detect by CPU model names that bundle AMD AI NPU/graphics
+	targets := []string{
+		"AMD Ryzen AI Max+ 395",
+		"AMD Ryzen AI Max 390",
+		"AMD Ryzen AI Max 385",
+		"AMD Ryzen AI 9 HX 375",
+		"AMD Ryzen AI 9 HX 370",
+		"AMD Ryzen AI 9 365",
+	}
+	// try lscpu first: extract 'Model name' field
+	out, err := execRuntime.GetRunner().SudoCmd("lscpu 2>/dev/null | awk -F': *' '/^Model name/{print $2; exit}' || true", false, false)
+	if err != nil {
+		return false, err
+	}
+	if out != "" {
+		lo := strings.ToLower(strings.TrimSpace(out))
+		for _, t := range targets {
+			if strings.Contains(lo, strings.ToLower(t)) {
+				return true, nil
+			}
+		}
+	}
+	// fallback to /proc/cpuinfo
+	out, err = execRuntime.GetRunner().SudoCmd("awk -F': *' '/^model name/{print $2; exit}' /proc/cpuinfo 2>/dev/null || true", false, false)
+	if err != nil {
+		return false, err
+	}
+	if out != "" {
+		lo := strings.ToLower(strings.TrimSpace(out))
+		for _, t := range targets {
+			if strings.Contains(lo, strings.ToLower(t)) {
+				return true, nil
+			}
+		}
+	}
+	return false, nil
+}
+
+func RocmVersion() (*semver.Version, error) {
+	const rocmVersionFile = "/opt/rocm/.info/version"
+	data, err := os.ReadFile(rocmVersionFile)
+	if err != nil {
+		// no ROCm installed, nothing to check
+		if os.IsNotExist(err) {
+			return nil, err
+		}
+		return nil, err
+	}
+	curStr := strings.TrimSpace(string(data))
+	cur, err := semver.NewVersion(curStr)
+	if err != nil {
+		return nil, fmt.Errorf("invalid rocm version: %s", curStr)
+	}
+	return cur, nil
+}