diff --git a/embassy-net/src/tcp.rs b/embassy-net/src/tcp.rs
index 732b6d217..7babb5293 100644
--- a/embassy-net/src/tcp.rs
+++ b/embassy-net/src/tcp.rs
@@ -95,7 +95,8 @@ impl<'a> TcpWriter<'a> {
 
     /// Flushes the written data to the socket.
     ///
-    /// This waits until all data has been sent, and ACKed by the remote host.
+    /// This waits until all data has been sent, and ACKed by the remote host. For a connection
+    /// closed with [`abort()`](TcpSocket::abort) it will wait for the TCP RST packet to be sent.
     pub async fn flush(&mut self) -> Result<(), Error> {
         self.io.flush().await
     }
@@ -198,7 +199,8 @@ impl<'a> TcpSocket<'a> {
 
     /// Flushes the written data to the socket.
     ///
-    /// This waits until all data has been sent, and ACKed by the remote host.
+    /// This waits until all data has been sent, and ACKed by the remote host. For a connection
+    /// closed with [`abort()`](TcpSocket::abort) it will wait for the TCP RST packet to be sent.
     pub async fn flush(&mut self) -> Result<(), Error> {
         self.io.flush().await
     }
@@ -262,6 +264,11 @@ impl<'a> TcpSocket<'a> {
     ///
     /// This instantly closes both the read and write halves of the socket. Any pending data
     /// that has not been sent will be lost.
+    ///
+    /// Note that the TCP RST packet is not sent immediately - if the `TcpSocket` is dropped too soon
+    /// the remote host may not know the connection has been closed.
+    /// `abort()` callers should wait for a [`flush()`](TcpSocket::flush) call to complete before
+    /// dropping or reusing the socket.
     pub fn abort(&mut self) {
         self.io.with_mut(|s, _| s.abort())
     }
@@ -347,9 +354,10 @@ impl<'d> TcpIo<'d> {
     async fn flush(&mut self) -> Result<(), Error> {
         poll_fn(move |cx| {
             self.with_mut(|s, _| {
+                let waiting_close = s.state() == tcp::State::Closed && s.remote_endpoint().is_some();
                 // If there are outstanding send operations, register for wake up and wait
                 // smoltcp issues wake-ups when octets are dequeued from the send buffer
-                if s.send_queue() > 0 {
+                if s.send_queue() > 0 || waiting_close {
                     s.register_send_waker(cx.waker());
                     Poll::Pending
                 // No outstanding sends, socket is flushed