Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 12 Jun 2009 17:42:40 +0000 (UTC)
From:      "Bjoern A. Zeeb" <bz@FreeBSD.org>
To:        Andrew Gallatin <gallatin@myri.com>
Cc:        FreeBSD net mailing list <freebsd-net@FreeBSD.org>
Subject:   Re: Ethernet NIC drivers depending unconditionally on INET
Message-ID:  <20090612173855.T22887@maildrop.int.zabbadoz.net>
In-Reply-To: <4A3260BD.4040307@myri.com>
References:  <20090611184555.J22887@maildrop.int.zabbadoz.net> <20090612013406.GB72855@michelle.cdnetworks.co.kr> <20090612100900.M22887@maildrop.int.zabbadoz.net> <4A3260BD.4040307@myri.com>

next in thread | previous in thread | raw e-mail | index | archive | help
On Fri, 12 Jun 2009, Andrew Gallatin wrote:

> Bjoern A. Zeeb wrote:
>
>>> As a sort of side-note, what about feature parity for INET6 for
>>> existing IPV4 features like TSO?  Who is working on that?
>> 
>> Ok, maybe we should write down the big list now. What all can we have?
>> What do we already have? What do we need? What needs to be changed?
>> 
>> IPv4 CSUM offloading
>> ULP (TCP|UDP|SCTP) CSUM offloading v4/v6
>>     We do have IFCAP_RXCSUM,IFCAP_TXCSUM but that means a
>>     different CSUM_* subset for each card, right?
>>
>>     We do have CSUM_IP, CSUM_TCP, CSUM_UDP, CSUM_SCTP.
>>     What will that be?

    insert  "need to" before "be"

> I'm not sure what you mean by this.
>
> Right now, at least on the receive side, tcp_input() for IPv6
> is completely ignoring ULP csum values sent up by drivers.
>
>
>
>> TSO v4/v6
>>     We do have IFCAP_TSO4|IFCAP_TSO6
>>
>>     We do have CSUM_TSO, so that should become CSUM_TSO4 and we'll
>>     need to add CSUM_TSO6?
>
> Cool! I had no idea that IFCAP_TSO6 was used, but apparently it is.
> When I get a chance to work on FreeBSD, I guess I'll flip that
> bit on in mxge and see if I actually get any packets with CSUM_TSO
> set.
>
> It would be helpful to have a CSUM_TSO{4,6} to reduce packet parsing.
> But as yongari pointed out, its fairly silly to make drivers parse the
> packets that the stack is sending them, and it would be ideal if
> we could easily pull the information from somewhere.

Yes, all for that; that's why we are talking about thing.  Not sure
what will make sense; perhaps we'll need to see what you all actually
need for all the drivers and combinations?

-- 
Bjoern A. Zeeb                      The greatest risk is not taking one.



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20090612173855.T22887>