Re: [PATCH v4 net-next v4 6/6] net: add heuristic for enabling TCP fraglist GRO

[Felix Fietkau]

On 30.04.24 12:12, Paolo Abeni wrote:
> On Sat, 2024-04-27 at 20:23 +0200, Felix Fietkau wrote:
> > When forwarding TCP after GRO, software segmentation is very expensive,
> > especially when the checksum needs to be recalculated.
> > One case where that's currently unavoidable is when routing packets over
> > PPPoE. Performance improves significantly when using fraglist GRO
> > implemented in the same way as for UDP.
> >
> > When NETIF_F_GRO_FRAGLIST is enabled, perform a lookup for an established
> > socket in the same netns as the receiving device. While this may not
> > cover all relevant use cases in multi-netns configurations, it should be
> > good enough for most configurations that need this.
> >
> > Here's a measurement of running 2 TCP streams through a MediaTek MT7622
> > device (2-core Cortex-A53), which runs NAT with flow offload enabled from
> > one ethernet port to PPPoE on another ethernet port + cake qdisc set to
> > 1Gbps.
> >
> > rx-gro-list off: 630 Mbit/s, CPU 35% idle
> > rx-gro-list on:  770 Mbit/s, CPU 40% idle
> >
> > Signe-off-by: Felix Fietkau <nbd@xxxxxxxx>
> > ---
> >  net/ipv4/tcp_offload.c   | 32 ++++++++++++++++++++++++++++++++
> >  net/ipv6/tcpv6_offload.c | 35 +++++++++++++++++++++++++++++++++++
> >  2 files changed, 67 insertions(+)
> >
> > diff --git a/net/ipv4/tcp_offload.c b/net/ipv4/tcp_offload.c
> > index 87ae9808e260..3e9b8c6f9c8c 100644
> > --- a/net/ipv4/tcp_offload.c
> > +++ b/net/ipv4/tcp_offload.c
> > @@ -407,6 +407,36 @@ void tcp_gro_complete(struct sk_buff *skb)
> >  }
> >  EXPORT_SYMBOL(tcp_gro_complete);
> >  
> > +static void tcp4_check_fraglist_gro(struct list_head *head, struct sk_buff *skb,
> > +				    struct tcphdr *th)
> > +{
> > +	const struct iphdr *iph;
> > +	struct sk_buff *p;
> > +	struct sock *sk;
> > +	struct net *net;
> > +	int iif, sdif;
> > +
> > +	if (!(skb->dev->features & NETIF_F_GRO_FRAGLIST))
>
> Should we add an 'unlikely()' here to pair with unlikely(is_flist) in
> *gro_receive / *gro_complete?
Not sure if unlikely() will make any difference here. I think it makes 
more sense in the other places than here.

> Should this test be moved into the caller, to avoid an unconditional
> function call in the ipv6 code?

The function is already called from tcp4_gro_receive, which is only 
called by IPv4 code. Also, since it's a static function called in only 
one place, it gets inlined by the compiler (at least in my builds).
Not sure what unconditional function call you're referring to.

- Felix