netdev
[Top] [All Lists]

Re: [PATCH 7/7] netpoll: avoid kfree_skb on packets with destructo

To: "David S. Miller" <davem@xxxxxxxxxxxxx>
Subject: Re: [PATCH 7/7] netpoll: avoid kfree_skb on packets with destructo
From: Matt Mackall <mpm@xxxxxxxxxxx>
Date: Thu, 3 Mar 2005 13:39:11 -0800
Cc: Jeff Garzik <jgarzik@xxxxxxxxx>, netdev@xxxxxxxxxxx, jmoyer@xxxxxxxxxx
In-reply-to: <20050303132906.2b5d597f.davem@xxxxxxxxxxxxx>
References: <7.454130102@xxxxxxxxxxx> <8.454130102@xxxxxxxxxxx> <20050303130031.066f0862.davem@xxxxxxxxxxxxx> <42277ED6.4020707@xxxxxxxxx> <20050303132906.2b5d597f.davem@xxxxxxxxxxxxx>
Sender: netdev-bounce@xxxxxxxxxxx
User-agent: Mutt/1.5.6+20040907i
On Thu, Mar 03, 2005 at 01:29:06PM -0800, David S. Miller wrote:
> On Thu, 03 Mar 2005 16:17:10 -0500
> Jeff Garzik <jgarzik@xxxxxxxxx> wrote:
> 
> > Heh, I was just writing this same message.
> > 
> > On a related note...  David, I would prefer if you merged up the netpoll 
> > stuff, since it touches mainly net/*
> > 
> > Is that cool w/ you?
> 
> No problem.  I still don't like this code in that it adds a locking
> penalty to everyone just by virtue of enabling netpoll.  We've worked
> so hard with things like NETIF_F_LLTX to eliminate locking, so this
> would be a huge step backwards.

The lock only happens if CONFIG_NETPOLL=y _and_ a netpoll client (eg
netconsole) is registered on the device in question.

I'm certainly open to ideas that improve upon that, but everything
I've come up with is equivalent in cost to a lock.

-- 
Mathematics is the supreme nostalgia of our time.

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