[PATCH net] net: airoha: Do not wake all netdev TX queues in airoha_qdma_wake_netdev_txqs()

[PATCH net] net: airoha: Do not wake all netdev TX queues in airoha_qdma_wake_netdev_txqs()

[Lorenzo Bianconi]

Do not wake every netdev TX queue across all ports sharing the QDMA
running netif_tx_wake_all_queues routine in airoha_qdma_wake_netdev_txqs()
but only the ones that are mapped the specific QDMA stopped hw TX queue.
This patch can potentially avoid waking already stopped netdev TX queues
that are mapped to a different QDMA hw TX queue.
Introduce airoha_qdma_get_txq utility routine.

Fixes: b94769eb2f30 ("net: airoha: Fix possible TX queue stall in airoha_qdma_tx_napi_poll()")
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
---
 drivers/net/ethernet/airoha/airoha_eth.c | 19 +++++++++++++++----
 drivers/net/ethernet/airoha/airoha_eth.h |  5 +++++
 2 files changed, 20 insertions(+), 4 deletions(-)

diff --git a/drivers/net/ethernet/airoha/airoha_eth.c b/drivers/net/ethernet/airoha/airoha_eth.c
index 19f67c7dd8e1..2ca569501045 100644
--- a/drivers/net/ethernet/airoha/airoha_eth.c
+++ b/drivers/net/ethernet/airoha/airoha_eth.c
@@ -847,13 +847,24 @@ static void airoha_qdma_wake_netdev_txqs(struct airoha_queue *q)
 {
 	struct airoha_qdma *qdma = q->qdma;
 	struct airoha_eth *eth = qdma->eth;
-	int i;
+	int i, qid = q - &qdma->q_tx[0];
 
 	for (i = 0; i < ARRAY_SIZE(eth->ports); i++) {
 		struct airoha_gdm_port *port = eth->ports[i];
+		int j;
+
+		if (!port)
+			continue;
 
-		if (port && port->qdma == qdma)
-			netif_tx_wake_all_queues(port->dev);
+		if (port->qdma != qdma)
+			continue;
+
+		for (j = 0; j < port->dev->num_tx_queues; j++) {
+			if (airoha_qdma_get_txq(qdma, j) != qid)
+				continue;
+
+			netif_wake_subqueue(port->dev, j);
+		}
 	}
 	q->txq_stopped = false;
 }
@@ -1965,7 +1976,7 @@ static netdev_tx_t airoha_dev_xmit(struct sk_buff *skb,
 	u16 index;
 	u8 fport;
 
-	qid = skb_get_queue_mapping(skb) % ARRAY_SIZE(qdma->q_tx);
+	qid = airoha_qdma_get_txq(qdma, skb_get_queue_mapping(skb));
 	tag = airoha_get_dsa_tag(skb, dev);
 
 	msg0 = FIELD_PREP(QDMA_ETH_TXMSG_CHAN_MASK,
diff --git a/drivers/net/ethernet/airoha/airoha_eth.h b/drivers/net/ethernet/airoha/airoha_eth.h
index 87b328cfefb0..c3ea7aadbd82 100644
--- a/drivers/net/ethernet/airoha/airoha_eth.h
+++ b/drivers/net/ethernet/airoha/airoha_eth.h
@@ -631,6 +631,11 @@ u32 airoha_rmw(void __iomem *base, u32 offset, u32 mask, u32 val);
 #define airoha_qdma_clear(qdma, offset, val)			\
 	airoha_rmw((qdma)->regs, (offset), (val), 0)
 
+static inline u16 airoha_qdma_get_txq(struct airoha_qdma *qdma, u16 qid)
+{
+	return qid % ARRAY_SIZE(qdma->q_tx);
+}
+
 static inline bool airoha_is_lan_gdm_port(struct airoha_gdm_port *port)
 {
 	/* GDM1 port on EN7581 SoC is connected to the lan dsa switch.

---
base-commit: a663bac71a2f0b3ac6c373168ca57b2a6e6381aa
change-id: 20260421-airoha-wake_netdev_txqs-optmization-65171ce4ebad

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

Re: [PATCH net] net: airoha: Do not wake all netdev TX queues in airoha_qdma_wake_netdev_txqs()

[Lorenzo Bianconi]

[-- Attachment #1: Type: text/plain, Size: 3452 bytes --]

> Do not wake every netdev TX queue across all ports sharing the QDMA
> running netif_tx_wake_all_queues routine in airoha_qdma_wake_netdev_txqs()
> but only the ones that are mapped the specific QDMA stopped hw TX queue.
> This patch can potentially avoid waking already stopped netdev TX queues
> that are mapped to a different QDMA hw TX queue.
> Introduce airoha_qdma_get_txq utility routine.
> 
> Fixes: b94769eb2f30 ("net: airoha: Fix possible TX queue stall in airoha_qdma_tx_napi_poll()")
> Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
> ---
>  drivers/net/ethernet/airoha/airoha_eth.c | 19 +++++++++++++++----
>  drivers/net/ethernet/airoha/airoha_eth.h |  5 +++++
>  2 files changed, 20 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/net/ethernet/airoha/airoha_eth.c b/drivers/net/ethernet/airoha/airoha_eth.c
> index 19f67c7dd8e1..2ca569501045 100644
> --- a/drivers/net/ethernet/airoha/airoha_eth.c
> +++ b/drivers/net/ethernet/airoha/airoha_eth.c
> @@ -847,13 +847,24 @@ static void airoha_qdma_wake_netdev_txqs(struct airoha_queue *q)
>  {
>  	struct airoha_qdma *qdma = q->qdma;
>  	struct airoha_eth *eth = qdma->eth;
> -	int i;
> +	int i, qid = q - &qdma->q_tx[0];
>  
>  	for (i = 0; i < ARRAY_SIZE(eth->ports); i++) {
>  		struct airoha_gdm_port *port = eth->ports[i];
> +		int j;
> +
> +		if (!port)
> +			continue;
>  
> -		if (port && port->qdma == qdma)
> -			netif_tx_wake_all_queues(port->dev);
> +		if (port->qdma != qdma)
> +			continue;
> +
> +		for (j = 0; j < port->dev->num_tx_queues; j++) {
> +			if (airoha_qdma_get_txq(qdma, j) != qid)
> +				continue;
> +
> +			netif_wake_subqueue(port->dev, j);
> +		}
>  	}
>  	q->txq_stopped = false;
>  }
> @@ -1965,7 +1976,7 @@ static netdev_tx_t airoha_dev_xmit(struct sk_buff *skb,
>  	u16 index;
>  	u8 fport;
>  
> -	qid = skb_get_queue_mapping(skb) % ARRAY_SIZE(qdma->q_tx);
> +	qid = airoha_qdma_get_txq(qdma, skb_get_queue_mapping(skb));
>  	tag = airoha_get_dsa_tag(skb, dev);
>  
>  	msg0 = FIELD_PREP(QDMA_ETH_TXMSG_CHAN_MASK,
> diff --git a/drivers/net/ethernet/airoha/airoha_eth.h b/drivers/net/ethernet/airoha/airoha_eth.h
> index 87b328cfefb0..c3ea7aadbd82 100644
> --- a/drivers/net/ethernet/airoha/airoha_eth.h
> +++ b/drivers/net/ethernet/airoha/airoha_eth.h
> @@ -631,6 +631,11 @@ u32 airoha_rmw(void __iomem *base, u32 offset, u32 mask, u32 val);
>  #define airoha_qdma_clear(qdma, offset, val)			\
>  	airoha_rmw((qdma)->regs, (offset), (val), 0)
>  
> +static inline u16 airoha_qdma_get_txq(struct airoha_qdma *qdma, u16 qid)
> +{
> +	return qid % ARRAY_SIZE(qdma->q_tx);
> +}
> +
>  static inline bool airoha_is_lan_gdm_port(struct airoha_gdm_port *port)
>  {
>  	/* GDM1 port on EN7581 SoC is connected to the lan dsa switch.
> 
> ---
> base-commit: a663bac71a2f0b3ac6c373168ca57b2a6e6381aa
> change-id: 20260421-airoha-wake_netdev_txqs-optmization-65171ce4ebad
> 
> Best regards,
> -- 
> Lorenzo Bianconi <lorenzo@kernel.org>
> 

commenting on Sashiko retported issues:
https://sashiko.dev/#/patchset/20260421-airoha-wake_netdev_txqs-optmization-v1-1-e0be95115d53%40kernel.org

- Can this cause an infinite NETDEV_TX_BUSY livelock when a QDMA hardware queue is full?
  The issue is already fixed in the following patch:
  https://patchwork.kernel.org/project/netdevbpf/patch/20260421-airoha-fix-bql-v1-1-f135afe4275b@kernel.org/

Regards,
Lorenzo

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]