Merge pull request #4271 from unsuman/fix/krunkit-gpu

vmdriver(krunkit): Use Vulkan GPU inside container to run AI models

Author: AkihiroSuda

pkg/driver/krunkit/boot/00-add-user-to-video-render-group.sh

@@ -0,0 +1,12 @@
+#!/bin/bash
+
+# SPDX-FileCopyrightText: Copyright The Lima Authors
+# SPDX-License-Identifier: Apache-2.0
+
+set -eux -o pipefail
+
+u="${LIMA_CIDATA_USER:-$USER}"
+getent group render >/dev/null 2>&1 || groupadd -f render
+getent group video >/dev/null 2>&1 || groupadd -f video
+sudo usermod -aG render "$u" || true
+sudo usermod -aG video "$u" || true

pkg/driver/krunkit/boot/01-gpu-device-perms.sh

@@ -0,0 +1,22 @@
+#!/bin/bash
+
+# SPDX-FileCopyrightText: Copyright The Lima Authors
+# SPDX-License-Identifier: Apache-2.0
+
+set -eux -o pipefail
+
+# Make DRM render/card nodes world-accessible
+install -d -m 0755 /etc/udev/rules.d
+cat >/etc/udev/rules.d/70-lima-drm.rules <<'EOF'
+KERNEL=="render[D]*", SUBSYSTEM=="drm", MODE="0666"
+KERNEL=="card*", SUBSYSTEM=="drm", MODE="0666"
+EOF
+
+# Apply to existing nodes now and future ones via udev
+udevadm control --reload || true
+udevadm trigger --subsystem-match=drm || true
+
+if [ -d /dev/dri ]; then
+	chmod 0666 /dev/dri/render[D]* 2>/dev/null || true
+	chmod 0666 /dev/dri/card* 2>/dev/null || true
+fi

pkg/driver/krunkit/hack/install-vulkan-gpu.sh

@@ -5,7 +5,7 @@ package krunkit
 
 import (
 	"context"
-	_ "embed"
+	"embed"
 	"errors"
 	"fmt"
 	"net"
@@ -39,25 +39,9 @@ type LimaKrunkitDriver struct {
 	krunkitWaitCh chan error
 }
 
-type KrunkitOpts struct {
-	GPUAccel *bool `yaml:"gpuAccel,omitempty"`
-}
-
-func NewKrunkitOpts(cfg *limatype.LimaYAML) (*KrunkitOpts, error) {
-	var krunkitOpts KrunkitOpts
-	if err := limayaml.Convert(cfg.VMOpts[vmType], &krunkitOpts, "vmOpts.krunkit"); err != nil {
-		return nil, err
-	}
-
-	return &krunkitOpts, nil
-}
-
 var (
 	_      driver.Driver   = (*LimaKrunkitDriver)(nil)
 	vmType limatype.VMType = "krunkit"
-
-	//go:embed hack/install-vulkan-gpu.sh
-	gpuProvisionScript string
 )
 
 func New() *LimaKrunkitDriver {
@@ -188,17 +172,6 @@ func validateConfig(cfg *limatype.LimaYAML) error {
 		return fmt.Errorf("field `mountType` must be %q or %q for krunkit driver, got %q", limatype.VIRTIOFS, limatype.REVSSHFS, *cfg.MountType)
 	}
 
-	// If GPU acceleration is requested, ensure Fedora image/template is used
-	krunkitOpts, err := NewKrunkitOpts(cfg)
-	if err != nil {
-		return err
-	}
-	if krunkitOpts.GPUAccel != nil && *krunkitOpts.GPUAccel {
-		if !isFedoraConfigured(cfg) {
-			logrus.Warn("gpuAccel: true requires a Fedora image (use a Fedora base template or image)")
-		}
-	}
-
 	return nil
 }
 
@@ -231,51 +204,39 @@ func (l *LimaKrunkitDriver) FillConfig(_ context.Context, cfg *limatype.LimaYAML
 
 	cfg.VMType = ptr.Of(vmType)
 
-	krunkitOpts, err := NewKrunkitOpts(cfg)
-	if err != nil {
-		return err
-	}
-
-	if krunkitOpts.GPUAccel == nil {
-		krunkitOpts.GPUAccel = ptr.Of(false)
-	}
-
-	if *krunkitOpts.GPUAccel {
-		gpuInstallScript := limatype.Provision{
-			Mode:   limatype.ProvisionModeData,
-			Script: ptr.Of(gpuProvisionScript),
-			ProvisionData: limatype.ProvisionData{
-				Content:     ptr.Of(gpuProvisionScript),
-				Path:        ptr.Of("/usr/local/bin/install-vulkan-gpu.sh"),
-				Permissions: ptr.Of("0755"),
-				Overwrite:   ptr.Of(false),
-				Owner:       cfg.User.Name,
-			},
-		}
-		cfg.Provision = append(cfg.Provision, gpuInstallScript)
-		cfg.Message = "To enable GPU support for krunkit, run the following command inside the VM:\n\033[32msudo install-vulkan-gpu.sh\033[0m\n"
-	}
-
 	return validateConfig(cfg)
 }
 
+//go:embed boot/*.sh
+var bootFS embed.FS
+
 func (l *LimaKrunkitDriver) BootScripts() (map[string][]byte, error) {
-	// Override default reboot-if-required with a no-op because Fedora does not support this well and
-	// takes a long time to start up.
-	krunkitOpts, err := NewKrunkitOpts(l.Instance.Config)
-	if err != nil {
-		return nil, err
-	}
-	if krunkitOpts.GPUAccel == nil || !*krunkitOpts.GPUAccel {
-		return nil, nil
+	scripts := make(map[string][]byte)
+
+	entries, err := bootFS.ReadDir("boot")
+	if err == nil && !isFedoraConfigured(l.Instance.Config) {
+		for _, entry := range entries {
+			if entry.IsDir() {
+				continue
+			}
+
+			content, err := bootFS.ReadFile("boot/" + entry.Name())
+			if err != nil {
+				return nil, err
+			}
+
+			scripts[entry.Name()] = content
+		}
 	}
-	scripts := map[string][]byte{
-		"00-reboot-if-required.sh": []byte(`#!/bin/sh
+
+	// Disabled by krunkit driver for Fedora to make boot time faster
+	if isFedoraConfigured(l.Instance.Config) {
+		scripts["00-reboot-if-required.sh"] = []byte(`#!/bin/sh
 set -eu
-# Disabled by krunkit driver
 exit 0
-`),
+`)
 	}
+
 	return scripts, nil
 }