env-ansible/tasks/miner/drivers/amdgpu.yml

---
# Install OpenCL drivers.

## Step 0/3: Dependencies ##
- name: AMDGPU | Variables
  set_fact:
    amdgpu_cron_job: AMDGPU - Need To Finish Installation
    amdgpu_cron_cmd: /root/bin/scm.sh

- name: AMDGPU | Install Dependencies
  package: 
    name: 
      - dkms 
      - mesa-common-dev
      - clinfo
    state: present

- name: AMDGPU | Set GRUB Timer
  lineinfile:
    path: /etc/default/grub
    regexp: '^GRUB_TIMEOUT='
    line: 'GRUB_TIMEOUT=3'
    state: present
    backup: yes

- name: AMDGPU | Remove Previous Run's Cron Job
  cron:
    user: root
    name: "{{ amdgpu_cron_job }}"
    state: absent

# Tests
- name: AMDGPU | Check If GA Kernel Installed
  shell: 'uname --kernel-release | grep 5.4.0-65-generic'
  register: ga_kernel
  ignore_errors: yes

- name: AMDGPU | Check If HWE Kernel Installed
  shell: 'apt search linux-image-5.8.0-* | grep -c installed'
  register: hwe_kernel
  ignore_errors: yes

#- name: AMDGPU | Check If All Driver Installed
#  shell: 'which amdgpu-uninstall'
#  register: amdgpu_install
#  ignore_errors: yes

- name: AMDGPU | Check If Pro Driver Installed
  shell: 'which amdgpu-pro-uninstall'
  register: amdgpu_install
  ignore_errors: yes

- name: AMDGPU | DEBUG
  debug:
    var: "{{ item }}"
  loop:
    - ga_kernel
    - hwe_kernel
    - amdgpu_install



## Step 1/3: Get Correct Kernel ##
- name: AMDGPU | Install GA Kernel
  package:
    name: linux-image-5.4.0-65-generic
    state: present
  when: amdgpu_install.failed and hwe_kernel.stdout != "0" and ga_kernel.failed

- name: AMDGPU | Prepare GRUB To Boot GA Kernel
  lineinfile:
    path: /etc/default/grub
    regexp: '^GRUB_DEFAULT='
    line: 'GRUB_DEFAULT="Advanced options for Ubuntu>Ubuntu, with Linux 5.4.0-65-generic"'
    state: present
    backup: yes
  when: amdgpu_install.failed and hwe_kernel.stdout != "0" and ga_kernel.failed

- name: AMDGPU | Update GRUB
  shell: update-grub
  when: amdgpu_install.failed and hwe_kernel.stdout != "0" and ga_kernel.failed

- name: AMDGPU | Prepare Cron Job To SCM
  cron:
    user: root
    name: "{{ amdgpu_cron_job }}"
    special_time: reboot
    job: "{{ amdgpu_cron_cmd }}"
    state: present
    disabled: no
  when: amdgpu_install.failed and hwe_kernel.stdout != "0" and ga_kernel.failed

- name: AMDGPU | Reboot To GA Kernel
  shell: reboot
  when: amdgpu_install.failed and hwe_kernel.stdout != "0" and ga_kernel.failed

- pause:
    prompt: "Rebooting to GA kernel! Please hold."
  when: amdgpu_install.failed and hwe_kernel.stdout != "0" and ga_kernel.failed



## Step 2/3: Remove Incorrect Kernels ##
- name: AMDGPU | Remove Bad Kernels
  shell: "apt remove --purge linux-generic-hwe-20.04 \
                             linux-oem-20.04 \
                             linux-hwe-* \
                             linux-oem-* \
                             linux-modules-5.1* \
                             linux-modules-5.8.0-* \
                             linux-modules-5.6.0-*"
  when: amdgpu_install.failed and hwe_kernel.stdout != "0"

- name: AMDGPU | Update + Clean System
  shell: "apt update; apt dist-upgrade -y; apt autoremove -y"
  when: amdgpu_install.failed and hwe_kernel.stdout != "0"

# This is to ensure we can test adding `apt install --install-recommends linux-generic` later 
- name: AMDGPU | Boot Default Kernel
  lineinfile:
    path: /etc/default/grub
    regexp: '^GRUB_DEFAULT='
    line: 'GRUB_DEFAULT=0'
    state: present
    backup: yes
  when: amdgpu_install.failed and hwe_kernel.stdout != "0"

- name: AMDGPU | Update GRUB
  shell: update-grub
  when: amdgpu_install.failed and hwe_kernel.stdout != "0"

- name: AMDGPU | Prepare Cron Job To SCM
  cron:
    user: root
    name: "{{ amdgpu_cron_job }}"
    special_time: reboot
    job: "{{ amdgpu_cron_cmd }}"
    state: present
    disabled: no
  when: amdgpu_install.failed and hwe_kernel.stdout != "0"

- name: AMDGPU | Reboot To Ensure GA Is 0
  shell: reboot
  when: amdgpu_install.failed and hwe_kernel.stdout != "0"

- pause:
    prompt: "Rebooting to clear kernels! Please hold."
  when: amdgpu_install.failed and hwe_kernel.stdout != "0"



## Step 3/3: Install AMD PRO Driver ##
- name: AMDGPU | Create Downloads Folder
  shell: 'mkdir -p /root/Downloads'
  when: amdgpu_install.failed

- name: AMDGPU | Download Driver
  shell: 'scp ling@leet:InstallFiles/Drivers/amdgpu-pro-20.45-1188099-ubuntu-20.04.tar.xz /root/Downloads/'
  when: amdgpu_install.failed

- name: AMDGPU | Extract Tarball
  shell: 'cd /root/Downloads; tar -xvf amdgpu-pro-20.45-1188099-ubuntu-20.04.tar.xz'
  when: amdgpu_install.failed

## Only one driver set can exist at a time! ##
#- name: AMDGPU | Install AMDGPU All Driver
#  shell: 'cd /root/Downloads/amdgpu-pro-20.45-1188099-ubuntu-20.04; ./amdgpu-install -y'
#  when: amdgpu_install.failed

- name: AMDGPU | Install AMDGPU-Pro Driver
  shell: 'cd /root/Downloads/amdgpu-pro-20.45-1188099-ubuntu-20.04; ./amdgpu-pro-install -y --opencl=pal,legacy'
  when: amdgpu_install.failed

- name: AMDGPU | Prepare Cron Job To SCM
  cron:
    user: root
    name: "{{ amdgpu_cron_job }}"
    special_time: reboot
    job: "{{ amdgpu_cron_cmd }}"
    state: present
    disabled: no
  when: amdgpu_install.failed

- name: AMDGPU | Reboot To GA Kernel
  shell: reboot
  when: amdgpu_install.failed

- pause:
    prompt: "Rebooting to use driver! Please hold."
  when: amdgpu_install.failed



## We're done! ##
- name: AMDGPU | Test Driver
  shell: clinfo
  register: clinfo

- name: AMDGPU | Test Results
  debug:
    var: clinfo
  ignore_errors: yes