diff --git a/tests/distributed/test_multiproc_executor.py b/tests/distributed/test_multiproc_executor.py
index d0a417a788..cd94138654 100644
--- a/tests/distributed/test_multiproc_executor.py
+++ b/tests/distributed/test_multiproc_executor.py
@@ -15,6 +15,7 @@ import pytest
 from tests.utils import multi_gpu_test
 from vllm.config import VllmConfig
 from vllm.engine.arg_utils import EngineArgs
+from vllm.platforms import current_platform
 from vllm.v1.core.sched.output import SchedulerOutput
 from vllm.v1.executor import multiproc_executor
 from vllm.v1.executor.multiproc_executor import MultiprocExecutor
@@ -378,8 +379,8 @@ def test_multiproc_executor_properties():
 def test_multiproc_executor_multi_node():
     """Test MultiprocExecutor with multi-node configuration.
     This simulates 2 nodes with TP=4:
-    - Node 0 (rank 0): Uses GPUs 0,1 (CUDA_VISIBLE_DEVICES=0,1) with TP=2
-    - Node 1 (rank 1): Uses GPUs 2,3 (CUDA_VISIBLE_DEVICES=2,3) with TP=2
+    - Node 0 (rank 0): Uses GPUs 0,1 (device_control_env_var=0,1) with TP=2
+    - Node 1 (rank 1): Uses GPUs 2,3 (device_control_env_var=2,3) with TP=2
     Total world_size = 4, nnodes = 2
     """
     # Python 3.14+ changed default multiprocessing start method to 'forkserver'
@@ -392,15 +393,20 @@ def test_multiproc_executor_multi_node():
     # symm_mem does not work for simulating multi instance in single node
     os.environ["VLLM_ALLREDUCE_USE_SYMM_MEM"] = "0"
 
+    # Use the platform-specific device control env var (e.g.
+    # CUDA_VISIBLE_DEVICES on CUDA, ZE_AFFINITY_MASK on XPU) so that each
+    # node only sees its own devices on different platform.
+    device_control_env_var = current_platform.device_control_env_var
+
     def run_node(node_rank: int, result_queue: multiprocessing.Queue, port: int):
         """Run a single node's executor."""
         executor = None
         try:
-            # Set CUDA_VISIBLE_DEVICES for this node
+            # Restrict visible devices for this node
             if node_rank == 0:
-                os.environ["CUDA_VISIBLE_DEVICES"] = "0,1"
+                os.environ[device_control_env_var] = "0,1"
             else:
-                os.environ["CUDA_VISIBLE_DEVICES"] = "2,3"
+                os.environ[device_control_env_var] = "2,3"
 
             # Create config for this node
             vllm_config = create_vllm_config(