netdev
[Top] [All Lists]

Re: Submission for S2io 10GbE driver

To: Leonid Grossman <leonid.grossman@xxxxxxxx>
Subject: Re: Submission for S2io 10GbE driver
From: Jeff Garzik <jgarzik@xxxxxxxxx>
Date: Mon, 16 Feb 2004 17:12:40 -0500
Cc: netdev@xxxxxxxxxxx, "'Andi Kleen'" <ak@xxxxxxx>, "'Stephen Hemminger'" <shemminger@xxxxxxxx>, "'Francois Romieu'" <romieu@xxxxxxxxxxxxx>, "'jamal'" <hadi@xxxxxxxxxx>, "'Grant Grundler'" <iod00d@xxxxxx>, "'Anton Blanchard'" <anton@xxxxxxxxx>, "'Jes Sorensen'" <jes@xxxxxxxxxxxxxxxxxx>, raghavendra.koushik@xxxxxxxx, "'ravinandan arakali'" <ravinandan.arakali@xxxxxxxx>
In-reply-to: <000201c3f4d2$2a5ddd90$7410100a@xxxxxxxxxxxx>
References: <000201c3f4d2$2a5ddd90$7410100a@xxxxxxxxxxxx>
Sender: netdev-bounce@xxxxxxxxxxx
User-agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.4) Gecko/20030703
Leonid Grossman wrote:
1. At the moment, lspci output looks like
"02:02.0 Ethernet controller: Unknown device 17d5:5831 (rev 02)"; do we
need to submit a patch for drivers/pci/pci.ids?

http://pciids.sourceforge.net/

The file drivers/pci/pci.ids is only associated with /proc/pci strings, and I'm trying to deprecate it :)


2. The card fully supports Ethernet and TCP header separation in
hardware (so called receive 3-buffer mode).
The mode may have some performance advantages but so far we did not
implement the mode in Linux since it seems that Linux stack can't handle
the fragmented buffers in the receive path. Is this a correct
assumption, does receive buffer has to be continuous?

In theory, the skb can be fragmented. I'm not as much as an expert in the ipv4/tcp/socket levels of the net stack, but I don't recall any place that yet supports skb frags on receive?

I think that's likely an area that would need some minor adjustments/additions in the upstream kernels, but not major surgery, since the skb already supports creating, noticing, and freeing frags.

        Jeff




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