[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20250902080957.47265-4-simon.schippers@tu-dortmund.de>
Date: Tue, 2 Sep 2025 10:09:56 +0200
From: Simon Schippers <simon.schippers@...dortmund.de>
To: willemdebruijn.kernel@...il.com, jasowang@...hat.com, mst@...hat.com,
eperezma@...hat.com, stephen@...workplumber.org,
netdev@...r.kernel.org, linux-kernel@...r.kernel.org,
virtualization@...ts.linux.dev, kvm@...r.kernel.org
Cc: Simon Schippers <simon.schippers@...dortmund.de>,
Tim Gebauer <tim.gebauer@...dortmund.de>
Subject: [PATCH 3/4] netdev queue flow control for TAP
Stopping the netdev queue is done in tun_net_xmit, as TAP uses this method
as its ndo_start_xmit.
To wake the queue, the new helper wake_netdev_queue is called in
tap_do_read. This is done in the blocking wait queue and after consuming
an SKB from the ptr_ring. This helper first checks if the netdev queue has
stopped. Then with the smp_rmb(), which is paired with the smp_wmb() of
tun_net_xmit, it is known that tun_net_xmit will not produce SKBs anymore.
With that knowledge, the helper can then wake the netdev queue if there is
at least a single spare slot. This check is done by calling the method
ptr_ring_spare.
Co-developed-by: Tim Gebauer <tim.gebauer@...dortmund.de>
Signed-off-by: Tim Gebauer <tim.gebauer@...dortmund.de>
Signed-off-by: Simon Schippers <simon.schippers@...dortmund.de>
---
drivers/net/tap.c | 22 ++++++++++++++++++++++
1 file changed, 22 insertions(+)
diff --git a/drivers/net/tap.c b/drivers/net/tap.c
index 1197f245e873..4d874672bcd7 100644
--- a/drivers/net/tap.c
+++ b/drivers/net/tap.c
@@ -753,6 +753,24 @@ static ssize_t tap_put_user(struct tap_queue *q,
return ret ? ret : total;
}
+static inline void wake_netdev_queue(struct tap_queue *q)
+{
+ struct netdev_queue *txq;
+ struct net_device *dev;
+
+ rcu_read_lock();
+ dev = rcu_dereference(q->tap)->dev;
+ txq = netdev_get_tx_queue(dev, q->queue_index);
+
+ if (netif_tx_queue_stopped(txq)) {
+ /* Paired with smp_wmb() in tun_net_xmit. */
+ smp_rmb();
+ if (ptr_ring_spare(&q->ring, 1))
+ netif_tx_wake_queue(txq);
+ }
+ rcu_read_unlock();
+}
+
static ssize_t tap_do_read(struct tap_queue *q,
struct iov_iter *to,
int noblock, struct sk_buff *skb)
@@ -785,12 +803,16 @@ static ssize_t tap_do_read(struct tap_queue *q,
ret = -ERESTARTSYS;
break;
}
+ wake_netdev_queue(q);
+
/* Nothing to read, let's sleep */
schedule();
}
if (!noblock)
finish_wait(sk_sleep(&q->sk), &wait);
+ wake_netdev_queue(q);
+
put:
if (skb) {
ret = tap_put_user(q, skb, to);
--
2.43.0
Powered by blists - more mailing lists