view release on metacpan or  search on metacpan

lib/Rex/GPU/NVIDIA.pm  view on Meta::CPAN

  if ($major >= 10) {
    push @packages, "kmod-nvidia-open-dkms", "nvidia-driver", "nvidia-driver-cuda";
  }
  else {
    run "dnf module enable nvidia-driver:open-dkms -y 2>/dev/null || true", auto_die => 0;
    push @packages, "nvidia-open";
  }

  Rex::Logger::info("  Installing: " . join(", ", @packages));

  # Use run() directly: Rex::Pkg::Dnf fails when dnf exits non-zero due to
  # DKMS post-install scripts (kernel module build). Verify via rpm -q instead.
  my $pkg_str = join(" ", @packages);
  run "dnf install -y $pkg_str", auto_die => 0;

  my $check = run "rpm -q nvidia-driver 2>&1", auto_die => 0;
  die "nvidia-driver not installed after dnf install — check dnf output\n"
    if $? != 0;
}

sub _rhel_major_version {
  # Rex::Commands::Gather::operating_system_version() strips dots,
  # so "10.1" becomes "101". Use operating_system_release() for the raw
  # version string and extract the major version ourselves.
  my $release = Rex::Commands::Gather::operating_system_release();
  $release =~ /^(\d+)/;
  return $1 + 0;
}

# ============================================================
#  openSUSE Leap
# ============================================================

sub _install_driver_suse {
  my ($os, $running_kernel) = @_;

  my $version = operating_system_version();
  my $major = int($version);

  # Remove any stale NVIDIA packages first — avoids kmp/userspace version mismatch
  # caused by libnvidia-ml/libnvidia-cfg from the standard OSS non-free repo lagging
  # behind the NVIDIA GFX repo packages.
  Rex::Logger::info("  Removing any existing NVIDIA packages...");
  run q{rpm -e $(rpm -qa | grep -E '^(nvidia|libnvidia)' | grep -v 'container') 2>/dev/null || true},
    auto_die => 0;

  # Add NVIDIA GFX repo (use direct baseurls — zypper cannot parse yum .repo files)
  if ($major >= 16) {
    Rex::Logger::info("  Adding NVIDIA GFX repo (suse16)...");
    run "zypper rr nvidia-gfx 2>/dev/null || true", auto_die => 0;
    run "zypper addrepo --refresh https://download.nvidia.com/opensuse/leap/16.0/ nvidia-gfx 2>/dev/null",
      auto_die => 0;
  }
  else {
    my $leap_version = sprintf("%.1f", $version / 10);  # 156 -> 15.6
    Rex::Logger::info("  Adding NVIDIA GFX repo (opensuse15, Leap $leap_version)...");
    run "zypper rr nvidia-gfx 2>/dev/null || true", auto_die => 0;
    run "zypper addrepo --refresh https://download.nvidia.com/opensuse/leap/$leap_version/ nvidia-gfx 2>/dev/null",
      auto_die => 0;
  }
  run "zypper --gpg-auto-import-keys refresh nvidia-gfx 2>/dev/null", auto_die => 0;

  # Use the meta package — it co-installs kmp-default + userspace at the same version,
  # preventing the split that causes "Driver/library version mismatch" with nvidia-smi.
  # Pre-signed kmp packages don't need kernel-devel/headers.
  my $meta_pkg = $major >= 16
    ? "nvidia-open-driver-G07-signed-kmp-meta"
    : "nvidia-open-driver-G06-signed-kmp-meta";

  Rex::Logger::info("  Installing $meta_pkg...");
  run "zypper install -y $meta_pkg", auto_die => 0;

  # Lock the OSS non-free standalone packages so future zypper updates don't
  # pull in a stale libnvidia-ml / libnvidia-cfg and cause a mismatch again.
  run "zypper addlock libnvidia-ml libnvidia-cfg 2>/dev/null || true", auto_die => 0;
}

# ============================================================
#  Nouveau blacklisting
# ============================================================

sub _blacklist_nouveau {
  file "/etc/modprobe.d/blacklist-nouveau.conf",
    content => "blacklist nouveau\noptions nouveau modeset=0\n";

  if (is_debian()) {
    run "update-initramfs -u 2>/dev/null", auto_die => 0;
  }
  elsif (is_redhat()) {
    run "dracut --force 2>/dev/null", auto_die => 0;
  }
  elsif (is_suse()) {
    run "dracut --force 2>/dev/null", auto_die => 0;
  }
}

# ============================================================
#  Container toolkit installation
# ============================================================

sub _install_toolkit_debian {
  pkg ["curl", "gnupg"], ensure => "present";

  run "curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey | gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg 2>/dev/null",
    auto_die => 0;

  file "/etc/apt/sources.list.d/nvidia-container-toolkit.list",
    content => 'deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://nvidia.github.io/libnvidia-container/stable/deb/$(ARCH) /' . "\n";

  run "apt-get -o DPkg::Lock::Timeout=120 update -q", auto_die => 0;
  # DPkg::Lock::Timeout=120: wait for apt-daily.timer lock that fires after reboot.
  run "DEBIAN_FRONTEND=noninteractive apt-get -o DPkg::Lock::Timeout=120 install -y nvidia-container-toolkit", auto_die => 0;
  my $check = run "dpkg -l nvidia-container-toolkit 2>/dev/null | grep -q '^ii'", auto_die => 0;
  die "nvidia-container-toolkit not installed\n" if $? != 0;
}

sub _install_toolkit_redhat {
  run "curl -s -L https://nvidia.github.io/libnvidia-container/stable/rpm/nvidia-container-toolkit.repo | tee /etc/yum.repos.d/nvidia-container-toolkit.repo",
    auto_die => 0;
  run "dnf clean expire-cache", auto_die => 0;
  run "dnf install -y nvidia-container-toolkit", auto_die => 0;
  my $check = run "rpm -q nvidia-container-toolkit 2>&1", auto_die => 0;
  die "nvidia-container-toolkit not installed\n" if $? != 0;
}

sub _install_toolkit_suse {
  # The .repo file URL is yum/dnf format — zypper needs the baseurl directly.
  # Remove any stale entry (possibly added with the wrong URL) before re-adding.
  my $arch = run "uname -m", auto_die => 0;
  chomp $arch;
  $arch ||= 'x86_64';

  run "zypper rr nvidia-container-toolkit 2>/dev/null || true", auto_die => 0;
  run "rpm --import https://nvidia.github.io/libnvidia-container/gpgkey 2>/dev/null",
    auto_die => 0;
  run "zypper addrepo --refresh https://nvidia.github.io/libnvidia-container/stable/rpm/$arch nvidia-container-toolkit 2>/dev/null",
    auto_die => 0;
  run "zypper --gpg-auto-import-keys refresh nvidia-container-toolkit 2>/dev/null",
    auto_die => 0;

  run "zypper install -y nvidia-container-toolkit", auto_die => 0;
}

# ============================================================
#  Containerd configuration
# ============================================================

sub _configure_containerd_rke2 {
  file "/var/lib/rancher/rke2/agent/etc/containerd", ensure => 'directory';
  file "/var/lib/rancher/rke2/agent/etc/containerd/config.toml.tmpl",
    content => "imports = [\"/etc/containerd/conf.d/*.toml\"]\nversion = 2\n";

  _write_nvidia_containerd_config();
}

sub _configure_containerd_standalone {
  run "nvidia-ctk runtime configure --runtime=containerd 2>&1", auto_die => 0;
  run "systemctl restart containerd 2>/dev/null", auto_die => 0;
}

sub _write_nvidia_containerd_config {
  file "/etc/containerd/conf.d", ensure => 'directory';
  file "/etc/containerd/conf.d/99-nvidia.toml", content => <<'TOML';
version = 2

[plugins]
  [plugins."io.containerd.grpc.v1.cri"]
    [plugins."io.containerd.grpc.v1.cri".containerd]
      [plugins."io.containerd.grpc.v1.cri".containerd.runtimes]
        [plugins."io.containerd.grpc.v1.cri".containerd.runtimes.nvidia]
          privileged_without_host_devices = false
          runtime_engine = ""
          runtime_root = ""
          runtime_type = "io.containerd.runc.v2"
          [plugins."io.containerd.grpc.v1.cri".containerd.runtimes.nvidia.options]
            BinaryName = "/usr/bin/nvidia-container-runtime"
TOML
}

# ============================================================
#  CDI spec generation
# ============================================================


sub generate_cdi_specs {
  Rex::Logger::info("Generating NVIDIA CDI specs...");
  run "mkdir -p /etc/cdi", auto_die => 0;
  run "nvidia-ctk cdi generate --output=/etc/cdi/nvidia.yaml 2>/dev/null", auto_die => 0;
  Rex::Logger::info("  [ok] CDI specs written to /etc/cdi/nvidia.yaml");
}

# ============================================================
#  Reboot
# ============================================================

sub _reboot_and_wait {
  Rex::Logger::info("Rebooting host to activate NVIDIA driver (replacing nouveau)...");