Make --gpus work with nvidia gpus

Somewhat documented here: https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/cdi-support.html https://stackoverflow.com/questions/25185405/using-gpu-from-a-docker-container Fixes: #21156 Don't have access to nvidia GPUS, relying on contributor testing. Signed-off-by: Daniel J Walsh <[email protected]>
containers · Jan 12, 2024 · 46cfc98 · 46cfc98
1 parent e06abd1
commit 46cfc98
Show file tree

Hide file tree

Showing 10 changed files with 27 additions and 12 deletions.
diff --git a/cmd/podman/common/create.go b/cmd/podman/common/create.go
@@ -700,6 +700,10 @@ func DefineCreateFlags(cmd *cobra.Command, cf *entities.ContainerCreateOptions,
 		)
 		_ = cmd.RegisterFlagCompletionFunc(gidmapFlagName, completion.AutocompleteNone)
 
+		gpuFlagName := "gpus"
+		createFlags.StringSliceVar(&cf.GPUs, gpuFlagName, []string{}, "GPU devices to add to the container ('all' to pass all GPUs)")
+		_ = cmd.RegisterFlagCompletionFunc(gpuFlagName, completion.AutocompleteNone)
+
 		uidmapFlagName := "uidmap"
 		createFlags.StringSliceVar(
 			&cf.UIDMap,

diff --git a/cmd/podman/containers/run.go b/cmd/podman/containers/run.go
@@ -80,11 +80,6 @@ func runFlags(cmd *cobra.Command) {
 	flags.StringVar(&runOpts.DetachKeys, detachKeysFlagName, containerConfig.DetachKeys(), "Override the key sequence for detaching a container. Format is a single character `[a-Z]` or a comma separated sequence of `ctrl-<value>`, where `<value>` is one of: `a-cf`, `@`, `^`, `[`, `\\`, `]`, `^` or `_`")
 	_ = cmd.RegisterFlagCompletionFunc(detachKeysFlagName, common.AutocompleteDetachKeys)
 
-	gpuFlagName := "gpus"
-	flags.String(gpuFlagName, "", "This is a Docker specific option and is a NOOP")
-	_ = cmd.RegisterFlagCompletionFunc(gpuFlagName, completion.AutocompleteNone)
-	_ = flags.MarkHidden("gpus")
-
 	passwdFlagName := "passwd"
 	flags.BoolVar(&runOpts.Passwd, passwdFlagName, true, "add entries to /etc/passwd and /etc/group")
 

diff --git a/docs/source/markdown/options/gpus.md b/docs/source/markdown/options/gpus.md
@@ -0,0 +1,8 @@
+####> This option file is used in:
+####>   podman create, pod clone, pod create, run
+####> If file is edited, make sure the changes
+####> are applicable to all of those.
+#### **--gpus**=*ENTRY*
+
+GPU devices to add to the container ('all' to pass all GPUs) Currently only
+Nvidia devices are supported.
diff --git a/docs/source/markdown/podman-create.1.md.in b/docs/source/markdown/podman-create.1.md.in
@@ -159,6 +159,8 @@ See [**Environment**](#environment) note below for precedence and examples.
 
 @@option gidmap.container
 
+@@option gpus
+
 @@option group-add
 
 @@option group-entry

diff --git a/docs/source/markdown/podman-pod-clone.1.md.in b/docs/source/markdown/podman-pod-clone.1.md.in
@@ -41,6 +41,8 @@ Note: the pod implements devices by storing the initial configuration passed by
 
 @@option gidmap.pod
 
+@@option gpus
+
 #### **--help**, **-h**
 
 Print usage statement.

diff --git a/docs/source/markdown/podman-pod-create.1.md.in b/docs/source/markdown/podman-pod-create.1.md.in
@@ -80,6 +80,8 @@ Set the exit policy of the pod when the last container exits.  Supported policie
 
 @@option gidmap.pod
 
+@@option gpus
+
 #### **--help**, **-h**
 
 Print usage statement.

diff --git a/docs/source/markdown/podman-run.1.md.in b/docs/source/markdown/podman-run.1.md.in
@@ -193,6 +193,8 @@ See [**Environment**](#environment) note below for precedence and examples.
 
 @@option gidmap.container
 
+@@option gpus
+
 @@option group-add
 
 @@option group-entry

diff --git a/pkg/domain/entities/pods.go b/pkg/domain/entities/pods.go
@@ -211,6 +211,7 @@ type ContainerCreateOptions struct {
 	EnvFile            []string
 	Expose             []string
 	GIDMap             []string
+	GPUs               []string
 	GroupAdd           []string
 	HealthCmd          string
 	HealthInterval     string

diff --git a/pkg/specgenutil/specgen.go b/pkg/specgenutil/specgen.go
@@ -784,7 +784,12 @@ func FillOutSpecGen(s *specgen.SpecGenerator, c *entities.ContainerCreateOptions
 		s.ImageVolumes = imageVolumes
 	}
 
-	for _, dev := range c.Devices {
+	devices := c.Devices
+	for _, gpu := range c.GPUs {
+		devices = append(devices, "nvidia.com/gpu="+gpu)
+	}
+
+	for _, dev := range devices {
 		s.Devices = append(s.Devices, specs.LinuxDevice{Path: dev})
 	}
 

diff --git a/test/e2e/run_device_test.go b/test/e2e/run_device_test.go
@@ -119,12 +119,6 @@ var _ = Describe("Podman run device", func() {
 		Expect(session).Should(ExitCleanly())
 	})
 
-	It("podman run --gpus noop", func() {
-		session := podmanTest.Podman([]string{"run", "--gpus", "all", ALPINE, "true"})
-		session.WaitWithDefaultTimeout()
-		Expect(session).Should(ExitCleanly())
-	})
-
 	It("podman run cannot access non default devices", func() {
 		session := podmanTest.Podman([]string{"run", "-v /dev:/dev-host", ALPINE, "head", "-1", "/dev-host/kmsg"})
 		session.WaitWithDefaultTimeout()