feat(roles/common): added power optimization

power optimization via unbinding unsed gpu cards

roles/common/defaults/main.yaml

@@ -1,3 +1,10 @@
 ---
 common_journal_retention_days: 21
 common_journal_max_disk_limit: "5%"
+
+gpu_unbind_devices: []
+# Example:
+# gpu_unbind_devices:
+#   - pci_id: "0000:06:00.0"
+#     driver: "amdgpu"
+#     label: "amd_igpu_cezanne"

roles/common/handlers/main.yaml

@@ -3,3 +3,7 @@
   ansible.builtin.service:
     name: ssh
     state: restarted
+
+- name: Reload systemd
+  ansible.builtin.systemd:
+    daemon_reload: true

roles/common/tasks/optimization.yaml

@@ -23,3 +23,22 @@
     - { user: "*", type: "soft", limit: "65536" }
     - { user: "*", type: "hard", limit: "65536" }
   tags: limits
+
+- name: Deploy GPU unbind systemd units
+  become: true
+  ansible.builtin.template:
+    src: gpu-unbind.service.j2
+    dest: /etc/systemd/system/gpu-unbind-{{ item.label }}.service
+    mode: "0644"
+  loop: "{{ gpu_unbind_devices }}"
+  notify: Reload systemd
+  tags: optimization
+
+- name: Enable and start GPU unbind servie
+  become: true
+  ansible.builtin.systemd:
+    name: gpu-unbind-{{ item.label }}
+    enabled: true
+    state: started
+  loop: "{{ gpu_unbind_devices }}"
+  tags: optimization

roles/common/templates/gpu-unbind.service.j2

@@ -0,0 +1,18 @@
+[Unit]
+Description=Unbind {{ item.label }} ({{ item.pci_id }}) and enable runtime PM
+After=multi-user.target
+
+[Service]
+Type=oneshot
+RemainAfterExit=yes
+ExecStart=/bin/sh -c '\
+  if [ -e /sys/bus/pci/drivers/{{ item.driver }}/{{ item.pci_id }} ]; then \
+    echo "{{ item.pci_id }}" > /sys/bus/pci/drivers/{{ item.driver }}/unbind; \
+  fi && \
+  echo auto > /sys/bus/pci/devices/{{ item.pci_id }}/power/control'
+ExecStop=/bin/sh -c '\
+  echo on > /sys/bus/pci/devices/{{ item.pci_id }}/power/control && \
+  echo "{{ item.pci_id }}" > /sys/bus/pci/drivers/{{ item.driver }}/bind'
+
+[Install]
+WantedBy=multi-user.target