diff mbox

"TCP: eth0: Driver has suspect GRO implementation, TCP performance may be compromised." message with "ethtool -K eth0 gro off"

Message ID 1486042270.13103.26.camel@edumazet-glaptop3.roam.corp.google.com
State RFC, archived
Delegated to: David Miller
Headers show

Commit Message

Eric Dumazet Feb. 2, 2017, 1:31 p.m. UTC 
On Thu, 2017-02-02 at 13:34 +0100, Markus Trippelsdorf wrote:
> On 2017.02.02 at 04:32 -0800, Eric Dumazet wrote:
> > On Thu, 2017-02-02 at 12:52 +0100, Markus Trippelsdorf wrote:
> > > Hi,
> > > 
> > > from time to time I see the following warning in my kernel log:
> > > 
> > >  TCP: eth0: Driver has suspect GRO implementation, TCP performance may be compromised.
> > > 
> > > This happens although I run "/usr/sbin/ethtool -K eth0 gro off" in my
> > > local boot script. 
> > > What is the warning trying to tell me?
> > > 
> > 
> > Please report
> > 
> > ethtool -i eth0
> 
> driver: ATL1E
> version: 1.0.0.7-NAPI
> firmware-version: L1e
> expansion-rom-version:
> bus-info: 0000:02:00.0
> supports-statistics: no
> supports-test: no
> supports-eeprom-access: no
> supports-register-dump: yes
> supports-priv-flags: no
> 

Note that this driver does not implement GRO yet.

Hard to believe there is such push back on GRO in 2017.

Anyway, I suspect the test is simply buggy ;)

Comments

Eric Dumazet Feb. 2, 2017, 1:59 p.m. UTC | #1 
On Thu, 2017-02-02 at 05:31 -0800, Eric Dumazet wrote:

> Anyway, I suspect the test is simply buggy ;)
> 
> diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
> index 41dcbd568cbe2403f2a9e659669afe462a42e228..5394a39fcce964a7fe7075b1531a8a1e05550a54 100644
> --- a/net/ipv4/tcp_input.c
> +++ b/net/ipv4/tcp_input.c
> @@ -164,7 +164,7 @@ static void tcp_measure_rcv_mss(struct sock *sk, const struct sk_buff *skb)
>  	if (len >= icsk->icsk_ack.rcv_mss) {
>  		icsk->icsk_ack.rcv_mss = min_t(unsigned int, len,
>  					       tcp_sk(sk)->advmss);
> -		if (unlikely(icsk->icsk_ack.rcv_mss != len))
> +		if (unlikely(icsk->icsk_ack.rcv_mss != len && skb_is_gso(skb)))
>  			tcp_gro_dev_warn(sk, skb);
>  	} else {
>  		/* Otherwise, we make more careful check taking into account,

This wont really help.

Our tcp_sk(sk)->advmss can be lower than the MSS used by the remote
peer.

ip ro add .... advmss 512

So the test is not universal.
diff mbox

Patch

diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 41dcbd568cbe2403f2a9e659669afe462a42e228..5394a39fcce964a7fe7075b1531a8a1e05550a54 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -164,7 +164,7 @@  static void tcp_measure_rcv_mss(struct sock *sk, const struct sk_buff *skb)
 	if (len >= icsk->icsk_ack.rcv_mss) {
 		icsk->icsk_ack.rcv_mss = min_t(unsigned int, len,
 					       tcp_sk(sk)->advmss);
-		if (unlikely(icsk->icsk_ack.rcv_mss != len))
+		if (unlikely(icsk->icsk_ack.rcv_mss != len && skb_is_gso(skb)))
 			tcp_gro_dev_warn(sk, skb);
 	} else {
 		/* Otherwise, we make more careful check taking into account,