Amd device plugin

Cargo.toml

@@ -5,6 +5,7 @@ members = [
   "packages/amazon-ecs-cni-plugins",
   "packages/amazon-ssm-agent",
   "packages/amazon-vpc-cni-plugins",
+  "packages/amd-k8s-device-plugin",
   "packages/aws-iam-authenticator",
   "packages/aws-otel-collector",
   "packages/aws-signer-notation-plugin",
@@ -66,6 +67,7 @@ members = [
   "packages/libcap",
   "packages/libcrypto",
   "packages/libdevmapper",
+  "packages/libdrm",
   "packages/libelf",
   "packages/libexpat",
   "packages/libffi",

kits/bottlerocket-core-kit/Cargo.toml

@@ -15,6 +15,7 @@ path = "../kit.rs"
 amazon-ecs-cni-plugins = { path = "../../packages/amazon-ecs-cni-plugins" }
 amazon-ssm-agent = { path = "../../packages/amazon-ssm-agent" }
 amazon-vpc-cni-plugins = { path = "../../packages/amazon-vpc-cni-plugins" }
+amd-k8s-device-plugin = { path = "../../packages/amd-k8s-device-plugin" }
 aws-iam-authenticator = { path = "../../packages/aws-iam-authenticator" }
 aws-otel-collector = { path = "../../packages/aws-otel-collector" }
 aws-signer-notation-plugin = { path = "../../packages/aws-signer-notation-plugin" }
@@ -76,6 +77,7 @@ libcap = { path = "../../packages/libcap" }
 libcrypto = { path = "../../packages/libcrypto" }
 libcryptsetup = { path = "../../packages/libcryptsetup" }
 libdevmapper = { path = "../../packages/libdevmapper" }
+libdrm = { path = "../../packages/libdrm" }
 libelf = { path = "../../packages/libelf" }
 libexpat = { path = "../../packages/libexpat" }
 libffi = { path = "../../packages/libffi" }

packages/amd-k8s-device-plugin/Cargo.toml

@@ -0,0 +1,22 @@
+[package]
+name = "amd-k8s-device-plugin"
+version = "0.1.0"
+edition = "2021"
+publish = false
+build = "../build.rs"
+
+[lib]
+path = "../packages.rs"
+
+[package.metadata.build-package]
+releases-url = "https://github.com/ROCm/k8s-device-plugin/releases"
+
+[[package.metadata.build-package.external-files]]
+url = "https://github.com/ROCm/k8s-device-plugin/archive/v1.31.0.8.tar.gz"
+sha512 = "23a127b46ad15cabbdd9abe18a8b75140340dcb10f41c9efdcfd30b38db7142edabaf6a97ed2c621f0414c78c0cb87fdc81a30b5e3fb016d16cbd11210143326"
+force-upstream = true
+
+[build-dependencies]
+glibc = { path = "../glibc" }
+libdrm = { path = "../libdrm" }
+hwloc = { path = "../hwloc" }

packages/amd-k8s-device-plugin/amd-k8s-device-plugin.service

@@ -0,0 +1,19 @@
+[Unit]
+Description=Start AMD kubernetes device plugin
+After=kubelet.service
+Wants=kubelet.service
+
+[Service]
+# Ensure that the kubelet device plugin socket exists before we start
+# A brief sleep is needed to avoid the `test` failing its first check
+ExecStartPre=/usr/bin/sleep 0.1
+ExecStartPre=/usr/bin/test -S /var/lib/kubelet/device-plugins/kubelet.sock
+ExecStart=/usr/bin/amd-device-plugin -logtostderr=true -stderrthreshold=INFO -v=5
+Type=simple
+TimeoutSec=0
+RestartSec=2
+Restart=always
+StandardError=journal+console
+
+[Install]
+WantedBy=multi-user.target

packages/amd-k8s-device-plugin/amd-k8s-device-plugin.spec

@@ -0,0 +1,75 @@
+%global goproject github.com/ROCm
+%global gorepo k8s-device-plugin
+%global goimport %{goproject}/%{gorepo}
+
+%global gover 1.31.0.8
+%global rpmver %{gover}
+
+Name: %{_cross_os}amd-k8s-device-plugin
+Version: %{rpmver}
+Release: 1%{?dist}
+Summary: Kubernetes device plugin for AMD GPUs
+License: Apache-2.0
+URL: https://github.com/ROCm/k8s-device-plugin
+Source0: https://github.com/ROCm/k8s-device-plugin/archive/v%{gover}.tar.gz
+Source1: amd-k8s-device-plugin.service
+
+BuildRequires: %{_cross_os}glibc-devel
+BuildRequires: %{_cross_os}libdrm-devel
+BuildRequires: %{_cross_os}hwloc-devel
+Requires: %{name}(binaries)
+Requires: %{_cross_os}libdrm
+Requires: %{_cross_os}hwloc
+
+%description
+%{summary}.
+
+%package bin
+Summary: Kubernetes device plugin for AMD GPUs binaries
+Provides: %{name}(binaries)
+Requires: (%{_cross_os}image-feature(no-fips) and %{name})
+Conflicts: (%{_cross_os}image-feature(fips) or %{name}-fips-bin)
+
+%description bin
+%{summary}.
+
+%package fips-bin
+Summary: Kubernetes device plugin for AMD GPUs binaries, FIPS edition
+Provides: %{name}(binaries)
+Requires: (%{_cross_os}image-feature(fips) and %{name})
+Conflicts: (%{_cross_os}image-feature(no-fips) or %{name}-bin)
+
+%description fips-bin
+%{summary}.
+
+%prep
+%autosetup -n %{gorepo}-%{gover} -p1
+%cross_go_setup %{gorepo}-%{gover} %{goproject} %{goimport}
+
+%build
+%cross_go_configure %{goimport}
+%set_cross_go_flags
+
+go build -ldflags="${GOLDFLAGS}" -o amd-device-plugin ./cmd/k8s-device-plugin/
+gofips build -ldflags="${GOLDFLAGS}" -o fips/amd-device-plugin ./cmd/k8s-device-plugin/
+
+%install
+install -d %{buildroot}%{_cross_bindir}
+install -p -m 0755 amd-device-plugin %{buildroot}%{_cross_bindir}
+
+install -d %{buildroot}%{_cross_fips_bindir}
+install -p -m 0755 fips/amd-device-plugin %{buildroot}%{_cross_fips_bindir}
+
+install -d %{buildroot}%{_cross_unitdir}
+install -p -m 0644 %{S:1} %{buildroot}%{_cross_unitdir}
+
+%files
+%license LICENSE
+%{_cross_attribution_file}
+%{_cross_unitdir}/amd-k8s-device-plugin.service
+
+%files bin
+%{_cross_bindir}/amd-device-plugin
+
+%files fips-bin
+%{_cross_fips_bindir}/amd-device-plugin

packages/libdrm/Cargo.toml

@@ -0,0 +1,25 @@
+[package]
+name = "libdrm"
+version = "0.1.0"
+edition = "2021"
+publish = false
+build = "../build.rs"
+
+[lib]
+path = "../packages.rs"
+
+[package.metadata.build-package]
+releases-url = "https://dri.freedesktop.org/libdrm/"
+
+[[package.metadata.build-package.external-files]]
+url = "https://dri.freedesktop.org/libdrm/libdrm-2.4.123.tar.xz"
+sha512 = "d3e6ba26bbd609fd87ca131690547eeb6a903c0a8c28b7f5cd5d0947619da09f31daf7bf4b6c38bf5e5dc173e2ccba476338ef682d8cf06d6b71ba73fc9b948d"
+force-upstream = true
+
+[[package.metadata.build-package.external-files]]
+url = "https://dri.freedesktop.org/libdrm/libdrm-2.4.123.tar.xz.sig"
+sha512 = "8eb04246661102ed93d80c1793882423d7798c5f46dfe6bc1c710de0afc5016cc99ccd1b02442587c8af5439eee3cfd008570ed671ffa28c7f02a3d9b78d4e95"
+force-upstream = true
+
+[build-dependencies]
+glibc = { path = "../glibc" }

packages/libdrm/libdrm.spec

@@ -0,0 +1,61 @@
+Name: %{_cross_os}libdrm
+Version: 2.4.123
+Release: 1%{?dist}
+Summary: Direct Rendering Manager runtime library
+License: MIT
+URL: https://dri.freedesktop.org
+Source0: https://dri.freedesktop.org/libdrm/libdrm-%{version}.tar.xz
+
+BuildRequires: %{_cross_os}glibc-devel
+Requires: %{_cross_os}glibc
+
+%description
+%{summary}.
+
+%package devel
+Summary: Files for development using the direct rendering manager library
+Requires: %{name}
+
+%description devel
+%{summary}.
+
+%prep
+%autosetup -n libdrm-%{version} -p1
+
+%build
+CONFIGURE_OPTS=(
+ --auto-features=disabled
+ -Dcairo-tests=disabled
+ -Dman-pages=disabled
+ -Dvalgrind=disabled
+ -Dfreedreno=disabled
+ -Dvc4=disabled
+ -Detnaviv=disabled
+ -Dexynos=disabled
+ -Dtegra=disabled
+ -Domap=disabled
+ -Dintel=disabled
+ -Dradeon=enabled
+ -Damdgpu=enabled
+ -Dnouveau=disabled
+ -Dtests=false
+)
+
+%cross_meson "${CONFIGURE_OPTS[@]}"
+%cross_meson_build
+
+%install
+%cross_meson_install
+
+%files
+%{_cross_attribution_file}
+%{_cross_libdir}/*.so.*
+%{_cross_includedir}/libsync.h
+%{_cross_datadir}/libdrm/amdgpu.ids
+
+%files devel
+%{_cross_libdir}/*.so
+%{_cross_includedir}/libdrm/
+%{_cross_includedir}/xf86drm.h
+%{_cross_includedir}/xf86drmMode.h
+%{_cross_pkgconfigdir}/*.pc

sources/ghostdog/src/error.rs

@@ -32,10 +32,14 @@ pub(super) enum Error {
     CheckEfaFailure { source: pciclient::PciClientError },
     #[snafu(display("Failed to check if Neuron device is attached: {}", source))]
     CheckNeuronFailure { source: pciclient::PciClientError },
+    #[snafu(display("Failed to check if AMD GPU device is attached: {}", source))]
+    CheckAmdGpuFailure { source: pciclient::PciClientError },
     #[snafu(display("Did not detect EFA"))]
     NoEfaPresent,
     #[snafu(display("Did not detect Neuron"))]
     NoNeuronPresent,
+    #[snafu(display("Did not detect AMD GPU"))]
+    NoAmdGpuPresent,
     #[snafu(display("'{}' has no parent directory", path.display()))]
     NoParentDirectory { path: std::path::PathBuf },
     #[snafu(display("Failed to open '{}': {}", path.display(), source))]

sources/ghostdog/src/main.rs

@@ -58,6 +58,7 @@ enum SubCommand {
     EbsDeviceName(EbsDeviceNameArgs),
     EfaPresent(EfaPresentArgs),
     NeuronPresent(NeuronPresentArgs),
+    AmdGpuPresent(AmdGpuPresentArgs),
     MatchDriver(MatchDriverArgs),
     MatchNvidiaDriver(MatchNvidiaDriverArgs),
     WriteInfinibandGuid(WriteInfinibandGuidArgs),
@@ -73,6 +74,11 @@ struct EfaPresentArgs {}
 /// Detect if Neuron devices are attached.
 struct NeuronPresentArgs {}
 
+#[derive(FromArgs, PartialEq, Debug)]
+#[argh(subcommand, name = "amd-gpu-present")]
+/// Detect if AMD GPU devices are attached.
+struct AmdGpuPresentArgs {}
+
 #[derive(FromArgs, PartialEq, Debug)]
 #[argh(subcommand, name = "scan")]
 /// Scan a device to see if it is an ephemeral disk.
@@ -163,6 +169,9 @@ fn main() -> Result<()> {
         SubCommand::NeuronPresent(_) => {
             is_neuron_attached()?;
         }
+        SubCommand::AmdGpuPresent(_) => {
+            is_amd_gpu_attached()?;
+        }
         SubCommand::MatchNvidiaDriver(driver) => {
             let driver_name = driver.driver_name;
             nvidia_driver_supported(&driver_name)?;
@@ -203,6 +212,14 @@ fn is_neuron_attached() -> Result<()> {
     }
 }
 
+fn is_amd_gpu_attached() -> Result<()> {
+    if pciclient::is_amd_gpu_attached().context(error::CheckAmdGpuFailureSnafu)? {
+        Ok(())
+    } else {
+        Err(error::Error::NoAmdGpuPresent)
+    }
+}
+
 // Returns true if this is an inf1 instance
 fn is_inf1_instance() -> Result<()> {
     if pciclient::is_inf1_instance().context(error::CheckInf1FailureSnafu)? {

sources/pciclient/src/lib.rs

@@ -9,7 +9,7 @@ mod private;
 
 use private::{
     call_list_devices, check_efa_attachment, check_inf1_attachment, check_inf2_attachment,
-    check_neuron_attachment, PciClient,
+    check_neuron_attachment, check_amd_gpu_attachment, PciClient,
 };
 
 use bon::Builder;
@@ -123,6 +123,11 @@ pub fn is_neuron_attached() -> Result<bool> {
     check_neuron_attachment(PciClient {})
 }
 
+/// Call `lspci` and check if there is any AMD GPU device attached.
+pub fn is_amd_gpu_attached() -> Result<bool> {
+    check_amd_gpu_attachment(PciClient {})
+}
+
 /// Call `lspci` and check if there are inf1 devices attached
 pub fn is_inf1_instance() -> Result<bool> {
     check_inf1_attachment(PciClient {})