[PATCH net] net: airoha: Fix possible TX queue stall in airoha_qdma_tx_napi_poll()

[PATCH net] net: airoha: Fix possible TX queue stall in airoha_qdma_tx_napi_poll()

[Lorenzo Bianconi]

Since multiple net_device TX queues can share the same hw QDMA TX queue,
there is no guarantee we have inflight packets queued in hw belonging to a
net_device TX queue stopped in the xmit path because hw QDMA TX queue
can be full. In this corner case the net_device TX queue will never be
re-activated. In order to avoid any potential net_device TX queue stall,
we need to wake all the net_device TX queues feeding the same hw QDMA TX
queue in airoha_qdma_tx_napi_poll routine.

Fixes: 23020f0493270 ("net: airoha: Introduce ethernet support for EN7581 SoC")
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
---
 drivers/net/ethernet/airoha/airoha_eth.c | 30 ++++++++++++++++++++++++++----
 1 file changed, 26 insertions(+), 4 deletions(-)

diff --git a/drivers/net/ethernet/airoha/airoha_eth.c b/drivers/net/ethernet/airoha/airoha_eth.c
index 9e995094c32a..e7610f36b8e4 100644
--- a/drivers/net/ethernet/airoha/airoha_eth.c
+++ b/drivers/net/ethernet/airoha/airoha_eth.c
@@ -855,6 +855,19 @@ static int airoha_qdma_init_rx(struct airoha_qdma *qdma)
 	return 0;
 }
 
+static void airoha_qdma_wake_tx_queues(struct airoha_qdma *qdma)
+{
+	struct airoha_eth *eth = qdma->eth;
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(eth->ports); i++) {
+		struct airoha_gdm_port *port = eth->ports[i];
+
+		if (port && port->qdma == qdma)
+			netif_tx_wake_all_queues(port->dev);
+	}
+}
+
 static int airoha_qdma_tx_napi_poll(struct napi_struct *napi, int budget)
 {
 	struct airoha_tx_irq_queue *irq_q;
@@ -931,12 +944,21 @@ static int airoha_qdma_tx_napi_poll(struct napi_struct *napi, int budget)
 
 			txq = netdev_get_tx_queue(skb->dev, queue);
 			netdev_tx_completed_queue(txq, 1, skb->len);
-			if (netif_tx_queue_stopped(txq) &&
-			    q->ndesc - q->queued >= q->free_thr)
-				netif_tx_wake_queue(txq);
-
 			dev_kfree_skb_any(skb);
 		}
+
+		if (q->ndesc - q->queued == q->free_thr) {
+			/* Since multiple net_device TX queues can share the
+			 * same hw QDMA TX queue, there is no guarantee we have
+			 * inflight packets queued in hw belonging to a
+			 * net_device TX queue stopped in the xmit path.
+			 * In order to avoid any potential net_device TX queue
+			 * stall, we need to wake all the net_device TX queues
+			 * feeding the same hw QDMA TX queue.
+			 */
+			airoha_qdma_wake_tx_queues(qdma);
+		}
+
 unlock:
 		spin_unlock_bh(&q->lock);
 	}

---
base-commit: 2dddb34dd0d07b01fa770eca89480a4da4f13153
change-id: 20260407-airoha-txq-potential-stall-ad52c53094e8

Best regards,
-- 
Lorenzo Bianconi <lorenzo@kernel.org>

Re: [PATCH net] net: airoha: Fix possible TX queue stall in airoha_qdma_tx_napi_poll()

[Paolo Abeni]

On 4/13/26 10:29 AM, Lorenzo Bianconi wrote:
> Since multiple net_device TX queues can share the same hw QDMA TX queue,
> there is no guarantee we have inflight packets queued in hw belonging to a
> net_device TX queue stopped in the xmit path because hw QDMA TX queue
> can be full. In this corner case the net_device TX queue will never be
> re-activated. In order to avoid any potential net_device TX queue stall,
> we need to wake all the net_device TX queues feeding the same hw QDMA TX
> queue in airoha_qdma_tx_napi_poll routine.
> 
> Fixes: 23020f0493270 ("net: airoha: Introduce ethernet support for EN7581 SoC")
> Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
> ---
>  drivers/net/ethernet/airoha/airoha_eth.c | 30 ++++++++++++++++++++++++++----
>  1 file changed, 26 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/net/ethernet/airoha/airoha_eth.c b/drivers/net/ethernet/airoha/airoha_eth.c
> index 9e995094c32a..e7610f36b8e4 100644
> --- a/drivers/net/ethernet/airoha/airoha_eth.c
> +++ b/drivers/net/ethernet/airoha/airoha_eth.c
> @@ -855,6 +855,19 @@ static int airoha_qdma_init_rx(struct airoha_qdma *qdma)
>  	return 0;
>  }
>  
> +static void airoha_qdma_wake_tx_queues(struct airoha_qdma *qdma)
> +{
> +	struct airoha_eth *eth = qdma->eth;
> +	int i;
> +
> +	for (i = 0; i < ARRAY_SIZE(eth->ports); i++) {
> +		struct airoha_gdm_port *port = eth->ports[i];
> +
> +		if (port && port->qdma == qdma)
> +			netif_tx_wake_all_queues(port->dev);
> +	}
> +}
> +
>  static int airoha_qdma_tx_napi_poll(struct napi_struct *napi, int budget)
>  {
>  	struct airoha_tx_irq_queue *irq_q;
> @@ -931,12 +944,21 @@ static int airoha_qdma_tx_napi_poll(struct napi_struct *napi, int budget)
>  
>  			txq = netdev_get_tx_queue(skb->dev, queue);
>  			netdev_tx_completed_queue(txq, 1, skb->len);
> -			if (netif_tx_queue_stopped(txq) &&
> -			    q->ndesc - q->queued >= q->free_thr)
> -				netif_tx_wake_queue(txq);
> -
>  			dev_kfree_skb_any(skb);
>  		}
> +
> +		if (q->ndesc - q->queued == q->free_thr) {

Sashiko says:

---
Can this exact equality check cause a permanent TX queue stall?
The previous logic checked if the free space was greater than or equal
to q->free_thr. If the xmit path stops the queue because the free space
drops to exactly q->free_thr, the hardware queue will have exactly
q->free_thr free slots.
When the NAPI poll routine subsequently reaps a completed descriptor,
q->queued is decremented, increasing the free space to q->free_thr + 1.
Since the free space is no longer exactly equal to the threshold, this
condition evaluates to false.
As NAPI continues to reap more descriptors, the free space strictly
increases, meaning the exact equality check will never evaluate to true
and the netdev TX queue will remain permanently stalled.
---

Please, try to triage sashiko comments proactively. Especially on NIC
drivers, validating the AI statements is extremely cumbersome for the
maintainers.

Thanks,

Paolo