[PATCH net-next] bareudp: Use pcpu stats to update rx_dropped counter.

[PATCH net-next] bareudp: Use pcpu stats to update rx_dropped counter.

[Guillaume Nault]

Use the core_stats rx_dropped counter to avoid the cost of atomic
increments.

Signed-off-by: Guillaume Nault <gnault@redhat.com>
---
I'm using core_stats for consistency with the vxlan implementation.
If we really want to avoid using core_stats in tunnel drivers, just
let me know and I'll convert bareudp to NETDEV_PCPU_STAT_DSTATS. But
for the moment, I still prefer to favor code consistency across UDP
tunnel implementations.

 drivers/net/bareudp.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/drivers/net/bareudp.c b/drivers/net/bareudp.c
index fa2dd76ba3d9..a2abfade82dd 100644
--- a/drivers/net/bareudp.c
+++ b/drivers/net/bareudp.c
@@ -84,7 +84,7 @@ static int bareudp_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 
 		if (skb_copy_bits(skb, BAREUDP_BASE_HLEN, &ipversion,
 				  sizeof(ipversion))) {
-			DEV_STATS_INC(bareudp->dev, rx_dropped);
+			dev_core_stats_rx_dropped_inc(bareudp->dev);
 			goto drop;
 		}
 		ipversion >>= 4;
@@ -94,7 +94,7 @@ static int bareudp_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 		} else if (ipversion == 6 && bareudp->multi_proto_mode) {
 			proto = htons(ETH_P_IPV6);
 		} else {
-			DEV_STATS_INC(bareudp->dev, rx_dropped);
+			dev_core_stats_rx_dropped_inc(bareudp->dev);
 			goto drop;
 		}
 	} else if (bareudp->ethertype == htons(ETH_P_MPLS_UC)) {
@@ -108,7 +108,7 @@ static int bareudp_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 				   ipv4_is_multicast(tunnel_hdr->daddr)) {
 				proto = htons(ETH_P_MPLS_MC);
 			} else {
-				DEV_STATS_INC(bareudp->dev, rx_dropped);
+				dev_core_stats_rx_dropped_inc(bareudp->dev);
 				goto drop;
 			}
 		} else {
@@ -124,7 +124,7 @@ static int bareudp_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 				   (addr_type & IPV6_ADDR_MULTICAST)) {
 				proto = htons(ETH_P_MPLS_MC);
 			} else {
-				DEV_STATS_INC(bareudp->dev, rx_dropped);
+				dev_core_stats_rx_dropped_inc(bareudp->dev);
 				goto drop;
 			}
 		}
@@ -136,7 +136,7 @@ static int bareudp_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 				 proto,
 				 !net_eq(bareudp->net,
 				 dev_net(bareudp->dev)))) {
-		DEV_STATS_INC(bareudp->dev, rx_dropped);
+		dev_core_stats_rx_dropped_inc(bareudp->dev);
 		goto drop;
 	}
 
@@ -144,7 +144,7 @@ static int bareudp_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 
 	tun_dst = udp_tun_rx_dst(skb, family, key, 0, 0);
 	if (!tun_dst) {
-		DEV_STATS_INC(bareudp->dev, rx_dropped);
+		dev_core_stats_rx_dropped_inc(bareudp->dev);
 		goto drop;
 	}
 	skb_dst_set(skb, &tun_dst->dst);
-- 
2.39.2

Re: [PATCH net-next] bareudp: Use pcpu stats to update rx_dropped counter.

[Simon Horman]

On Fri, Oct 18, 2024 at 03:35:28PM +0200, Guillaume Nault wrote:
> Use the core_stats rx_dropped counter to avoid the cost of atomic
> increments.
> 
> Signed-off-by: Guillaume Nault <gnault@redhat.com>
> ---
> I'm using core_stats for consistency with the vxlan implementation.
> If we really want to avoid using core_stats in tunnel drivers, just
> let me know and I'll convert bareudp to NETDEV_PCPU_STAT_DSTATS. But
> for the moment, I still prefer to favor code consistency across UDP
> tunnel implementations.

Hi Guillaume, all,

I think that this patch improves things and the suggestion above
can be treated as a possible follow-up.

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

Re: [PATCH net-next] bareudp: Use pcpu stats to update rx_dropped counter.

[patchwork-bot+netdevbpf]

Hello:

This patch was applied to netdev/net-next.git (main)
by David S. Miller <davem@davemloft.net>:

On Fri, 18 Oct 2024 15:35:28 +0200 you wrote:
> Use the core_stats rx_dropped counter to avoid the cost of atomic
> increments.
> 
> Signed-off-by: Guillaume Nault <gnault@redhat.com>
> ---
> I'm using core_stats for consistency with the vxlan implementation.
> If we really want to avoid using core_stats in tunnel drivers, just
> let me know and I'll convert bareudp to NETDEV_PCPU_STAT_DSTATS. But
> for the moment, I still prefer to favor code consistency across UDP
> tunnel implementations.
> 
> [...]

Here is the summary with links:
  - [net-next] bareudp: Use pcpu stats to update rx_dropped counter.
    https://git.kernel.org/netdev/net-next/c/788d5d655bc9

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html