[Top] [All Lists]

Re: [PATCH] select appropriate skb size in tcp_sendmsg when TSO is used

To: "David S. Miller" <davem@xxxxxxxxxxxxx>
Subject: Re: [PATCH] select appropriate skb size in tcp_sendmsg when TSO is used
From: Jeff Garzik <jgarzik@xxxxxxxxx>
Date: Thu, 28 Oct 2004 00:16:33 -0400
Cc: Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx>, thomas.spatzier@xxxxxxxxxx, netdev@xxxxxxxxxxx
In-reply-to: <>
References: <> <> <> <> <> <> <> <> <> <>
Sender: netdev-bounce@xxxxxxxxxxx
User-agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.7.3) Gecko/20040922
David S. Miller wrote:
On Wed, 27 Oct 2004 10:22:09 +1000
Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx> wrote:

On Tue, Oct 26, 2004 at 08:15:31PM -0400, Jeff Garzik wrote:

IMHO it is valid to disable SG without disabling checksums, no?

It's useless: The packet header is always in a separate memory location from the packet data, when using zerocopy sendfile(2).

When not using zerocopy sendfile, you are copying the data _anyway_.

I'm fine with adding this check. However I think that belongs in another patch since we don't check that in register_netdev currently.

Dave, what do you think?

I believe that allowing TX csum support without SG _is_
useful even though it is not _effective_.

It is quite desirable for a driver author to be able to
test out his TX csum offload support first, then add
SG support next.  Similarly, if a driver author suspects
some issues with either SG or TX csum support, he can
better isolate the problem if we allow this.

Jeff do you agree?

<shrug> it's never used that way in practice AFAIK, only used by confused sysadmins :)

I won't object if you preserve the behavior, but I still don't see much value in allowing it.


<Prev in Thread] Current Thread [Next in Thread>