DEBUG: add nvidia_quiesce

Author: yf225

.github/actions/setup-nvidia-helion/action.yml

@@ -72,6 +72,60 @@ runs:
               )
           }
 
+          nvidia_quiesce() {
+              (
+                  local GRACE=5
+                  local -a MODULES=("nvidia_uvm" "nvidia_peermem" "nvidia_drm" "nvidia_modeset" "nvidia")
+
+                  echo "[nvidia-quiesce] Stopping NVIDIA daemons (if any)..."
+                  sudo systemctl stop --quiet nvidia-persistenced.service || true
+                  sudo systemctl stop --quiet nvidia-fabricmanager.service || true
+                  sudo pkill -9 nvidia-persistenced 2>/dev/null || true
+                  sudo pkill -9 nvidia-fabricmanager 2>/dev/null || true
+
+                  echo "[nvidia-quiesce] Enumerating processes with an open GPU handle..."
+                  local -a pids=()
+                  local -a extra=()
+                  mapfile -t pids < <(
+                      { nvidia-smi --query-compute-apps=pid --format=csv,noheader,nounits 2>/dev/null || true; } | sort -u
+                  )
+                  mapfile -t extra < <(
+                      { sudo lsof -t /dev/nvidia* 2>/dev/null || true; } | sort -u
+                  )
+                  if ((${#extra[@]})); then
+                      pids+=("${extra[@]}")
+                  fi
+                  if ((${#pids[@]})); then
+                      mapfile -t pids < <(
+                          printf "%s\n" "${pids[@]}" | grep -E '^[0-9]+$' | sort -u
+                      )
+                  fi
+
+                  if ((${#pids[@]})); then
+                      echo "[nvidia-quiesce] Sending SIGTERM to: ${pids[*]}"
+                      sudo kill -TERM "${pids[@]}" 2>/dev/null || true
+                      sleep "${GRACE}"
+                      echo "[nvidia-quiesce] Forcing SIGKILL (if still alive)..."
+                      sudo kill -KILL "${pids[@]}" 2>/dev/null || true
+                  else
+                      echo "[nvidia-quiesce] No GPU users found."
+                  fi
+
+                  echo "[nvidia-quiesce] Unloading kernel modules..."
+                  for module in "${MODULES[@]}"; do
+                      if lsmod | awk '{print $1}' | grep -qx "${module}"; then
+                          echo "  -> ${module}"
+                          if ! sudo modprobe -r "${module}"; then
+                              echo "    WARNING: ${module} still busy; aborting."
+                              return 1
+                          fi
+                      fi
+                  done
+
+                  echo "[nvidia-quiesce] All NVIDIA modules removed. Safe to run the new driver installer."
+              )
+          }
+
           unload_nvidia_modules() {
               MODULES=("nvidia-uvm" "nvidia-drm" "nvidia-modeset" "nvidia")
               for module in "${MODULES[@]}"; do
@@ -116,6 +170,7 @@ runs:
 
                           # Turn off persistent mode so that the installation script can unload the kernel module
                           sudo killall nvidia-persistenced || true
+                          nvidia_quiesce || true
                           unload_nvidia_modules || true
                       else
                           HAS_NVIDIA_DRIVER=1