net: dlink: count tx_dropped when dropping skb on link down
Subsystems (2)
PATCH 1/1 net: dlink: count tx_dropped when dropping skb on link down
Increment tx_dropped when dropping the skb due to link down.
Tested-on: D-Link DGE-550T Rev-A3
--- drivers/net/ethernet/dlink/dl2k.c | 1 + 1 file changed, 1 insertion(+)
drivers/net/ethernet/dlink/dl2k.c +1 -0
@@ -733,6 +733,7 @@ start_xmit (struct sk_buff *skb, struct net_device *dev)u64 tfc_vlan_tag =0 ;if (np->link_status ==0 ) {/* Link Down */ + dev->stats.tx_dropped++;dev_kfree_skb_any (skb);return NETDEV_TX_OK ; }
Signature · 1 line
2.52.0
Quoted 13 lines · Yeounsu Moon
Increment
tx_droppedwhen dropping the skb due to link down.Tested-on: D-Link DGE-550T Rev-A3
Signed-off-by Yeounsu Moon <[email protected]>--- drivers/net/ethernet/dlink/dl2k.c | 1 + 1 file changed, 1 insertion(+)drivers/net/ethernet/dlink/dl2k.c +0
-0
drivers/net/ethernet/dlink/dl2k.c start_xmit (struct sk_buff *skb, struct net_device *dev) · line 736
Quoted 4 lines
drivers/net/ethernet/dlink/dl2k.c +1
-0@@ -733,6 +733,7 @@ start_xmit (struct sk_buff *skb, struct net_device *dev)u64 tfc_vlan_tag =0 ;if (np->link_status ==0 ) {/* Link Down */ + dev->stats.tx_dropped++;
Do you see this being hit very often? It should be that as soon as you know the link is down, you tell the core, and it will stop calling start_xmit. If you see this counter being incremented a lot, it indicates there is a problem somewhere else.
You might want to consider converting this driver to phylink.
Andrew
Quoted 24 lines · Andrew Lunn
Quoted 13 lines · Yeounsu Moon
Increment
tx_droppedwhen dropping the skb due to link down.Tested-on: D-Link DGE-550T Rev-A3
Signed-off-by Yeounsu Moon <[email protected]>--- drivers/net/ethernet/dlink/dl2k.c | 1 + 1 file changed, 1 insertion(+)drivers/net/ethernet/dlink/dl2k.c +0
-0drivers/net/ethernet/dlink/dl2k.c start_xmit (struct sk_buff *skb, struct net_device *dev) · line 736
Quoted 4 lines
drivers/net/ethernet/dlink/dl2k.c +1
-0@@ -733,6 +733,7 @@ start_xmit (struct sk_buff *skb, struct net_device *dev)u64 tfc_vlan_tag =0 ;if (np->link_status ==0 ) {/* Link Down */ + dev->stats.tx_dropped++;Do you see this being hit very often? It should be that as soon as you know the link is down, you tell the core, and it will stop calling
start_xmit. If you see this counter being incremented a lot, it indicates there is a problem somewhere else.
Quoted 3 lines
You might want to consider converting this driver to phylink.
Andrew
Sorry for the late reply. I recently started my first job and have been a bit busy settling in.
To answer your question: this path is hit extremely rarely. In practice, I only observed it in rather extreme cases, such as forcibly disconnecting the physical link (e.g. unplugging the Ethernet cable). I have not seen it occurring during normal operation.
Given that, dropping the packet in this situation seems reasonable, so I added a counter to make such cases visible. However, I am not entirely sure whether this is the right direction, or whether this is just masking a problem that should be handled elsewhere.
Do you think this should instead be addressed by improving how link-down is propagted to the core, or would converting this driver to phylink help avoid or improve this situation?
Yeounsu
Andrew Lunn
Reviewed-byQuoted 29 lines · Yeounsu Moon
Quoted 24 lines · Andrew Lunn
Quoted 13 lines · Yeounsu Moon
Increment
tx_droppedwhen dropping the skb due to link down.Tested-on: D-Link DGE-550T Rev-A3
Signed-off-by Yeounsu Moon <[email protected]>--- drivers/net/ethernet/dlink/dl2k.c | 1 + 1 file changed, 1 insertion(+)drivers/net/ethernet/dlink/dl2k.c +0
-0drivers/net/ethernet/dlink/dl2k.c start_xmit (struct sk_buff *skb, struct net_device *dev) · line 736
Quoted 4 lines
drivers/net/ethernet/dlink/dl2k.c +1
-0@@ -733,6 +733,7 @@ start_xmit (struct sk_buff *skb, struct net_device *dev)u64 tfc_vlan_tag =0 ;if (np->link_status ==0 ) {/* Link Down */ + dev->stats.tx_dropped++;Do you see this being hit very often? It should be that as soon as you know the link is down, you tell the core, and it will stop calling
start_xmit. If you see this counter being incremented a lot, it indicates there is a problem somewhere else.Quoted 3 lines
You might want to consider converting this driver to phylink.
Andrew
Quoted 7 lines
Sorry for the late reply. I recently started my first job and have been a bit busy settling in.
To answer your question: this path is hit extremely rarely. In practice, I only observed it in rather extreme cases, such as forcibly disconnecting the physical link (e.g. unplugging the Ethernet cable). I have not seen it occurring during normal operation.
In that case, i think this is fine. There is a race condition here between indicating the carrier is down and the network stopping passing packets, so this can happen, and incrementing the counter is fine.
Andrew
Quoted 4 lines · Andrew Lunn
You might want to consider converting this driver to phylink.
Andrew
One more note: I’ve started looking into phylink based on your suggestion, and I appreciate you pointing me to that keyword.
It seems useful not only for this issue, but also for improving the driver more generally.
Thanks for the pointer.