atomic_set(&workload->shadow_ctx_active, 1);
                break;
        case INTEL_CONTEXT_SCHEDULE_OUT:
-               /* If the status is -EINPROGRESS means this workload
-                * doesn't meet any issue during dispatching so when
-                * get the SCHEDULE_OUT set the status to be zero for
-                * good. If the status is NOT -EINPROGRESS means there
-                * is something wrong happened during dispatching and
-                * the status should not be set to zero
-                */
-               if (workload->status == -EINPROGRESS)
-                       workload->status = 0;
                atomic_set(&workload->shadow_ctx_active, 0);
                break;
        default:
                wait_event(workload->shadow_ctx_status_wq,
                           !atomic_read(&workload->shadow_ctx_active));
 
+               /* If this request caused GPU hang, req->fence.error will
+                * be set to -EIO. Use -EIO to set workload status so
+                * that when this request caused GPU hang, didn't trigger
+                * context switch interrupt to guest.
+                */
+               if (likely(workload->status == -EINPROGRESS)) {
+                       if (workload->req->fence.error == -EIO)
+                               workload->status = -EIO;
+                       else
+                               workload->status = 0;
+               }
+
                i915_gem_request_put(fetch_and_zero(&workload->req));
 
                if (!workload->status && !vgpu->resetting) {