diff --git a/.github/workflows/uring-kernel-version-test.yml b/.github/workflows/uring-kernel-version-test.yml
index 14170f36c8b..bc90ea0a2f6 100644
--- a/.github/workflows/uring-kernel-version-test.yml
+++ b/.github/workflows/uring-kernel-version-test.yml
@@ -40,6 +40,8 @@ jobs:
           make defconfig
           make -j$(nproc)
 
+      # We are running tests also on a Kernel version that does not support io_uring
+      # to check if the fallback mechanism works
       - name: Generate test binaries with io_uring enabled
         run: |
           # Build both integration (tokio/tests/) and unit (e.g., tokio/src/fs/file/tests.rs) tests with io_uring enabled
diff --git a/tokio/tests/fs_uring_cancel_open.rs b/tokio/tests/fs_uring_cancel_open.rs
index 5f204020564..989f9525d42 100644
--- a/tokio/tests/fs_uring_cancel_open.rs
+++ b/tokio/tests/fs_uring_cancel_open.rs
@@ -13,10 +13,21 @@ use std::fs;
 use std::future::poll_fn;
 use std::task::Poll;
 use tempfile::NamedTempFile;
+use tokio_test::assert_pending;
+
+use crate::support::io_uring::io_uring_supported;
+
+mod support {
+    pub(crate) mod io_uring;
+}
 
 // see: https://github.com/tokio-rs/tokio/issues/7979
 #[tokio::test]
 async fn file_descriptors_are_closed_when_cancelling_open_op() {
+    if !io_uring_supported() {
+        return;
+    }
+
     let tmp = NamedTempFile::new().unwrap();
     let path = tmp.path().to_path_buf();
 
@@ -37,10 +48,8 @@ async fn file_descriptors_are_closed_when_cancelling_open_op() {
                 let fut = opt.open(&path);
 
                 // If io_uring is enabled (and not falling back to the thread pool),
-                // the first poll should return Pending. We don't check if the result
-                // is actually pending because we run some checks on old kernel that
-                // do not support uring.
-                let _pending = Box::pin(fut).poll_unpin(cx);
+                // the first poll should return Pending.
+                assert_pending!(Box::pin(fut).poll_unpin(cx));
 
                 tx.send(()).unwrap();
 
diff --git a/tokio/tests/fs_uring_completed_then_dropped_before_repoll.rs b/tokio/tests/fs_uring_completed_then_dropped_before_repoll.rs
index 359e298b53a..a3debe47beb 100644
--- a/tokio/tests/fs_uring_completed_then_dropped_before_repoll.rs
+++ b/tokio/tests/fs_uring_completed_then_dropped_before_repoll.rs
@@ -6,6 +6,11 @@
     target_os = "linux"
 ))]
 
+mod support {
+    pub(crate) mod io_uring;
+}
+
+use std::fmt::Debug;
 use std::fs;
 use std::future::Future;
 use std::path::PathBuf;
@@ -16,6 +21,9 @@ use std::time::Duration;
 use tempfile::NamedTempFile;
 use tokio::sync::mpsc::{unbounded_channel, UnboundedSender};
 use tokio::time::timeout;
+use tokio_test::assert_pending;
+
+use crate::support::io_uring::io_uring_supported;
 
 /// Count currently-open fds in this process.
 fn fd_count() -> usize {
@@ -42,14 +50,17 @@ struct PollOpenOnceThenNeverRepoll<F> {
     polled_once: bool,
 }
 
-impl<F: Future> Future for PollOpenOnceThenNeverRepoll<F> {
-    type Output = ();
+impl<F: Future> Future for PollOpenOnceThenNeverRepoll<F>
+where
+    F::Output: Debug,
+{
+    type Output = F::Output;
 
     fn poll(mut self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<Self::Output> {
         if !self.polled_once {
-            // We don't check if the result is actually pending because
-            // we run some checks on old kernel that do not support uring.
-            let _pending = self.inner.as_mut().poll(cx);
+            // If io_uring is enabled (and not falling back to the thread pool),
+            // the first poll should return Pending.
+            assert_pending!(self.inner.as_mut().poll(cx));
 
             self.polled_once = true;
             self.first_poll_tx.take().unwrap().send(()).unwrap();
@@ -80,7 +91,7 @@ async fn completed_then_dropped_before_repoll(path: PathBuf) {
             second_poll_tx: Some(second_tx),
             polled_once: false,
         }
-        .await;
+        .await
     });
 
     // Wait until the inner open has been polled once and registered with io_uring.
@@ -98,6 +109,10 @@ async fn completed_then_dropped_before_repoll(path: PathBuf) {
 
 #[test]
 fn uring_completed_then_dropped() {
+    if !io_uring_supported() {
+        return;
+    }
+
     let rt = tokio::runtime::Builder::new_multi_thread()
         .worker_threads(1)
         .enable_all()
diff --git a/tokio/tests/fs_uring_runtime_shutdown.rs b/tokio/tests/fs_uring_runtime_shutdown.rs
index 429cfce1884..62f0a17e3f1 100644
--- a/tokio/tests/fs_uring_runtime_shutdown.rs
+++ b/tokio/tests/fs_uring_runtime_shutdown.rs
@@ -14,10 +14,21 @@ use std::future::poll_fn;
 use std::task::Poll;
 use tempfile::NamedTempFile;
 use tokio::runtime::Builder;
+use tokio_test::assert_pending;
+
+use crate::support::io_uring::io_uring_supported;
+
+mod support {
+    pub(crate) mod io_uring;
+}
 
 // see: https://github.com/tokio-rs/tokio/issues/7979
 #[test]
 fn shutdown_runtime_while_performing_io_uring_ops() {
+    if !io_uring_supported() {
+        return;
+    }
+
     let tmp = NamedTempFile::new().unwrap();
     let path = tmp.path().to_path_buf();
 
@@ -41,10 +52,8 @@ fn shutdown_runtime_while_performing_io_uring_ops() {
                     let fut = opt.open(&path);
 
                     // If io_uring is enabled (and not falling back to the thread pool),
-                    // the first poll should return Pending. We don't check if the result
-                    // is actually a pending because we run some CI checks based on old
-                    // kernels that don't support uring.
-                    let _pending = Box::pin(fut).poll_unpin(cx);
+                    // the first poll should return Pending.
+                    assert_pending!(Box::pin(fut).poll_unpin(cx));
 
                     tx.send(()).unwrap();
 
diff --git a/tokio/tests/support/io_uring.rs b/tokio/tests/support/io_uring.rs
new file mode 100644
index 00000000000..affb2f833be
--- /dev/null
+++ b/tokio/tests/support/io_uring.rs
@@ -0,0 +1,24 @@
+#![cfg(all(
+    tokio_unstable,
+    feature = "io-uring",
+    feature = "rt",
+    feature = "fs",
+    target_os = "linux"
+))]
+
+use io_uring::IoUring;
+
+// Currently, we are running some of the tests on Kernels where io_uring is not supported
+// to check if the fallback mechanism works, this comes with the limitation that we are not
+// able to run some checks (e.g., asserting a poll returns pending). This utiliy function
+// is useful when we want to run a test only in Linux targets where io_uring is supported.
+pub fn io_uring_supported() -> bool {
+    match IoUring::new(256) {
+        Ok(_) => true,
+        // The Kernel does not support io-uring
+        Err(e) if e.raw_os_error() == Some(libc::ENOSYS) => false,
+        Err(_) => unreachable!(
+            "The target should either support io_uring or return ENOSYS if not supported"
+        ),
+    }
+}