netdev - Re: [PATCH] net: Check skb->rxhash in gro

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Fri, 10 Jan 2014 10:54:37 -0800
From:	Eric Dumazet <eric.dumazet@...il.com>
To:	Tom Herbert <therbert@...gle.com>
Cc:	David Miller <davem@...emloft.net>,
	Linux Netdev List <netdev@...r.kernel.org>
Subject: Re: [PATCH] net: Check skb->rxhash in gro_receive

On Fri, 2014-01-10 at 10:15 -0800, Tom Herbert wrote:

> The objective is to compute the rxhash at most once per packet.

Once per GRO packet ( up to 45 MSS) or once per incoming frame ( 1 MSS
) ?

Your patch computes rxhash for every incoming frame.

> 
> > If your patch was doing this, I would have no complain.
> >
> > (No new conditional branch, and skb->rxhash, if provided by the NIC,
> > can give a hint.)
> >
> > diff --git a/net/core/dev.c b/net/core/dev.c
> > index ce01847793c0..c9f7a26d7ce7 100644
> > --- a/net/core/dev.c
> > +++ b/net/core/dev.c
> > @@ -3798,7 +3798,8 @@ static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
> >         for (p = napi->gro_list; p; p = p->next) {
> >                 unsigned long diffs;
> >
> > -               diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
> > +               diffs = p->rxhash ^ skb->rxhash;
> > +               diffs |= (unsigned long)p->dev ^ (unsigned long)skb->dev;
> >                 diffs |= p->vlan_tci ^ skb->vlan_tci;
> >                 if (maclen == ETH_HLEN)
> >                         diffs |= compare_ether_header(skb_mac_header(p),
> >
> As I said in the commit log, the skb->rxhash should be a very strong
> indicator of that flows will match (maybe >99% ?), so putting that
> first potentially short circuits a lot of work even in just this
> function.

Are you speaking of your "goto skip;" ?

compare_ether_header() is done with 10 instructions on x86_64

There is no point trying to avoid it.

Really, 99% of the time gro_list contains zero or one single slot, I
have hard data saying so.

If you want to optimize the case where list is fully populated (because
of yet another synthetic benchmark you use), you really need to build a
temporary list so that all layers do not even have to check
NAPI_GRO_CB(p)->same_flow

Each gro handler would remove non matching flow from this temp list.

-> when we finally reach tcp_gro_receive(), list would contain a single
element.



--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html