netdev
[Top] [All Lists]

Re: [PATCH] NETIF_F_LLTX for devices 2

To: Andi Kleen <ak@xxxxxxx>
Subject: Re: [PATCH] NETIF_F_LLTX for devices 2
From: Jeff Garzik <jgarzik@xxxxxxxxx>
Date: Mon, 13 Sep 2004 12:10:03 -0400
Cc: "David S. Miller" <davem@xxxxxxxxxxxxx>, hadi@xxxxxxxxxx, herbert@xxxxxxxxxxxxxxxxxxx, netdev@xxxxxxxxxxx
In-reply-to: <20040913065958.GC12185@wotan.suse.de>
References: <20040908072408.GI27886@wotan.suse.de> <1094629677.1089.155.camel@jzny.localdomain> <20040908134713.1bcd46d3.davem@davemloft.net> <1094823215.1121.129.camel@jzny.localdomain> <20040911142116.GL4431@wotan.suse.de> <1094933731.2343.109.camel@jzny.localdomain> <20040911174535.2acbb957.davem@davemloft.net> <20040912100114.GB11484@havoc.gtf.org> <20040912102529.GA27096@wotan.suse.de> <20040912161604.GA23366@havoc.gtf.org> <20040913065958.GC12185@wotan.suse.de>
Sender: netdev-bounce@xxxxxxxxxxx
User-agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.7.2) Gecko/20040803
Andi Kleen wrote:
On Sun, Sep 12, 2004 at 12:16:05PM -0400, Jeff Garzik wrote:

Incorrect, you are changing the callsites, which -does- affect every
driver.


Please read the code before making such claims.

The new return code is _only_ checked when NETIF_F_LLTX is set. A driver that doesn't set this new flag won't ever recognize any difference.


I read the code :)

The basic premise is that one should be _really_ conservative when touching the core RX and TX paths. Regardless of how safe _you_ feel the code is, it is very new, under-analyzed, and untried.

The NAPI-related bug recently fixed in tg3 is an example of the unintended consequences of using this new feature.

        Jeff



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