Fix early return skipping critical vfio-pci, MOFED, and nouveau operations when driver reinstall is skipped

Signed-off-by: Karthik Vetrivel <[email protected]>

Author: karthikvetrivel

cmd/driver-manager/main.go

@@ -289,7 +289,7 @@ func (dm *DriverManager) uninstallDriver() error {
 
 	if skip, reason := dm.shouldSkipUninstall(); skip {
 		dm.log.Infof("Skipping driver uninstall: %s", reason)
-		return nil
+		return dm.performPostDriverOperations(false)
 	}
 
 	// Fetch current component states
@@ -363,6 +363,14 @@ func (dm *DriverManager) uninstallDriver() error {
 		dm.log.Info("Successfully uninstalled nvidia driver components")
 	}
 
+	return dm.performPostDriverOperations(true)
+}
+
+// performPostDriverOperations executes operations that run after driver operations.
+// The componentsWereEvicted parameter indicates whether GPU operator components were
+// actually evicted during this run, which determines whether uncordoning and rescheduling
+// are necessary.
+func (dm *DriverManager) performPostDriverOperations(componentsWereEvicted bool) error {
 	// Handle vfio-pci driver unbinding
 	if err := dm.unbindVfioPCI(); err != nil {
 		dm.log.Error("Unable to unbind vfio-pci driver from all devices")
@@ -380,14 +388,17 @@ func (dm *DriverManager) uninstallDriver() error {
 	}
 
 	// Cleanup and reschedule components
-	if dm.isGPUPodEvictionEnabled() || dm.isAutoDrainEnabled() {
-		if err := dm.kubeClient.UncordonNode(dm.config.nodeName); err != nil {
-			dm.log.Warn("Failed to uncordon node")
+	// Only needed if we actually evicted components during this run
+	if componentsWereEvicted {
+		if dm.isGPUPodEvictionEnabled() || dm.isAutoDrainEnabled() {
+			if err := dm.kubeClient.UncordonNode(dm.config.nodeName); err != nil {
+				dm.log.Warn("Failed to uncordon node")
+			}
 		}
-	}
 
-	if err := dm.rescheduleGPUOperatorComponents(); err != nil {
-		dm.log.Warnf("Failed to reschedule GPU operator components: %v", err)
+		if err := dm.rescheduleGPUOperatorComponents(); err != nil {
+			dm.log.Warnf("Failed to reschedule GPU operator components: %v", err)
+		}
 	}
 
 	// Handle nouveau driver