[PATCH net-next] net: bgmac: stop advertising TX checksum offload

[PATCH net-next] net: bgmac: stop advertising TX checksum offload

[Rosen Penev]

The driver advertises NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM, but
bgmac_dma_tx_add() calls skb_checksum_help() on every CHECKSUM_PARTIAL
skb, so the checksum is always computed in software anyway. The return
value is ignored too, so if skb_checksum_help() fails the frame is sent
with a bad checksum.

Remove the feature flags and the fallback. The stack then computes the
checksum before handing the skb to the driver.

Tested on an ASUS RT-N16.

Signed-off-by: Rosen Penev <rosenp@gmail.com>
---
 drivers/net/ethernet/broadcom/bgmac.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)

diff --git a/drivers/net/ethernet/broadcom/bgmac.c b/drivers/net/ethernet/broadcom/bgmac.c
index 2dff43f540d3..822903a3ac6e 100644
--- a/drivers/net/ethernet/broadcom/bgmac.c
+++ b/drivers/net/ethernet/broadcom/bgmac.c
@@ -145,9 +145,6 @@ static netdev_tx_t bgmac_dma_tx_add(struct bgmac *bgmac,
 		goto err_drop;
 	}
 
-	if (skb->ip_summed == CHECKSUM_PARTIAL)
-		skb_checksum_help(skb);
-
 	nr_frags = skb_shinfo(skb)->nr_frags;
 
 	/* ring->end - ring->start will return the number of valid slots,
@@ -1574,7 +1571,7 @@ int bgmac_enet_probe(struct bgmac *bgmac)
 		goto err_dma_free;
 	}
 
-	net_dev->features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
+	net_dev->features = NETIF_F_SG;
 	net_dev->hw_features = net_dev->features;
 	net_dev->vlan_features = net_dev->features;
 
-- 
2.56.0

Re: [PATCH net-next] net: bgmac: stop advertising TX checksum offload

[Markus Elfring]

…
> Remove the feature flags and the fallback. The stack then computes the
> checksum before handing the skb to the driver.
…

How can a “stack” compute anything?

Regards,
Markus

Re: [PATCH net-next] net: bgmac: stop advertising TX checksum offload

[Simon Horman]

On Tue, Nov 03, 2026 at 10:26:21AM +0100, Markus Elfring wrote:
> …
> > Remove the feature flags and the fallback. The stack then computes the
> > checksum before handing the skb to the driver.
> …
> 
> How can a “stack” compute anything?

Perhaps your assumption is that stack refers to stack memory.
But my assumption is that stack refers to the software stack,
or slightly more specifically, core parts of the Linux
implementation of networking.

Re: [PATCH net-next] net: bgmac: stop advertising TX checksum offload

[Simon Horman]

On Mon, Nov 02, 2026 at 01:58:06PM -0800, Rosen Penev wrote:
> The driver advertises NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM, but
> bgmac_dma_tx_add() calls skb_checksum_help() on every CHECKSUM_PARTIAL
> skb, so the checksum is always computed in software anyway. The return
> value is ignored too, so if skb_checksum_help() fails the frame is sent
> with a bad checksum.
> 
> Remove the feature flags and the fallback. The stack then computes the
> checksum before handing the skb to the driver.
> 
> Tested on an ASUS RT-N16.
> 
> Signed-off-by: Rosen Penev <rosenp@gmail.com>

Reviewed-by: Simon Horman <horms@kernel.org>