diff --git a/python/paddle/fluid/tests/unittests/CMakeLists.txt b/python/paddle/fluid/tests/unittests/CMakeLists.txt
index d39ed01df8ca2fd447739234525091c5d97f9587..3c5d4403e8b880f0bcd3c9365e3e6b9eea331248 100644
--- a/python/paddle/fluid/tests/unittests/CMakeLists.txt
+++ b/python/paddle/fluid/tests/unittests/CMakeLists.txt
@@ -413,7 +413,6 @@ list(REMOVE_ITEM TEST_OPS test_imperative_static_runner_mnist)
 list(REMOVE_ITEM TEST_OPS test_imperative_static_runner_while)
 # disable test_cumsum_op temporaily
 # list(REMOVE_ITEM TEST_OPS test_cumsum_op)
-list(REMOVE_ITEM TEST_OPS test_dataloader_dataset)
 
 # disable this unittest temporarily
 list(REMOVE_ITEM TEST_OPS test_imperative_data_loader_exception)
diff --git a/python/paddle/fluid/tests/unittests/test_dataloader_dataset.py b/python/paddle/fluid/tests/unittests/test_dataloader_dataset.py
index 08589f0191d8c698ecd8a4017d6c14fa476610b1..d2f4eadc9c56412adeec96c58d31bec5da3ab9ac 100644
--- a/python/paddle/fluid/tests/unittests/test_dataloader_dataset.py
+++ b/python/paddle/fluid/tests/unittests/test_dataloader_dataset.py
@@ -66,7 +66,12 @@ class TestDatasetWithDiffOutputPlace(unittest.TestCase):
             for image, label in loader:
                 self.assertTrue(image.place.is_gpu_place())
                 self.assertTrue(label.place.is_cuda_pinned_place())
-                break
+                # FIXME(dkp): when input tensor is in GPU place and
+                # iteration break in the median, it seems the GPU
+                # tensor put into blocking_queue cannot be safely
+                # released and may cause ABRT/SEGV, this should
+                # be fixed
+                # break
 
     def test_multi_process(self):
         # DataLoader with multi-process mode is not supported on MacOs and Windows currently