[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <CANn89iKMhCJietM70bYx+ZEjqTLQjUvQJFno=AQ9uStkhg57Sg@mail.gmail.com>
Date: Tue, 10 Sep 2024 10:56:58 +0200
From: Eric Dumazet <edumazet@...gle.com>
To: Sean Anderson <sean.anderson@...ux.dev>
Cc: Madalin Bucur <madalin.bucur@....com>, netdev@...r.kernel.org,
Paolo Abeni <pabeni@...hat.com>, Jakub Kicinski <kuba@...nel.org>, linux-kernel@...r.kernel.org,
"David S . Miller" <davem@...emloft.net>
Subject: Re: [PATCH net] net: dpaa: Pad packets to ETH_ZLEN
On Mon, Sep 9, 2024 at 8:02 PM Sean Anderson <sean.anderson@...ux.dev> wrote:
> OK, I tested both and was able to use
>
> ./pktgen/pktgen_sample01_simple.sh -i net5 -m 7e:de:97:38:53:b9 -d 10.0.0.2 -n 3 -s 59
>
> with a call to `pg_set $DEV "frags 2"` added manually.
>
> This results in the following result
>
> OK: 109(c13+d95) usec, 1 (59byte,0frags)
>
> The original patch causes the nonlinear path to be taken (see with the
> "tx S/G [TOTAL]" statistic) while your suggestion uses the linear path.
> Both work, since there's no problem using the nonlinear path with a
> linear skb.
Maybe it works today, but this allocates an extra page for nothing,
this is an extra burden.
Vast majority of skb_padto() users call it early in ndo_start_xmit(),
let's follow this pattern.
Powered by blists - more mailing lists