netdev - Re: [PATCH] ifb: add multi-queue support

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <412e6f7f0911120132j5442a1ffp790277597f563953@mail.gmail.com>
Date:	Thu, 12 Nov 2009 17:32:13 +0800
From:	Changli Gao <xiaosuo@...il.com>
To:	Jarek Poplawski <jarkao2@...il.com>
Cc:	Patrick McHardy <kaber@...sh.net>,
	"David S. Miller" <davem@...emloft.net>,
	Stephen Hemminger <shemminger@...tta.com>,
	Eric Dumazet <eric.dumazet@...il.com>,
	Tom Herbert <therbert@...gle.com>, netdev@...r.kernel.org
Subject: Re: [PATCH] ifb: add multi-queue support

On Thu, Nov 12, 2009 at 4:52 PM, Jarek Poplawski <jarkao2@...il.com> wrote:
> On 12-11-2009 04:12, Changli Gao wrote:
>> On Wed, Nov 11, 2009 at 11:59 PM, Patrick McHardy <kaber@...sh.net> wrote:
> ...
>>> What protects the device from disappearing here and below during
>>> dev_queue_xmit() and netif_rx_ni()?
>>
>> For dev_queue_xmit(), dev is holded by skb->_dst, so there is no
>> problem.

Oh, I'm wrong, and not all the skbs using dev_queue_xmit() has a valid
dst, such as AF_PACKET. As skb->dev doesn't own dev, so I think it is
AF_PACKETs fault. Too bad!

>> But for netif_rx_ni(), I don't know how to prevent the device
>> disappearing, and it seems that all the NIC drivers have this problem.
>> Maybe there was the assumption about the execution context of
>> netif_rx() before. Now softirq can't be executed by softirqd, so the
>> packet receiving path maybe interleaved. I don't know how to prevent
>> it happening.
>
> Btw, maybe I miss something, but ifb with act_mirred is for many users
> on the fast path. I'm not sure you proved enough why moving it to the
> process context and additional multiqueue functionality (which seems
> "pretty cool", but probably mostly for testing purposes) don't harm
> its main use?

Because process is more flexible than other activities, and you can
specify its priority and bind it to some CPU, and so on. If you have a
SMP system, and a NIC which doesn't support MQ. When you use it as a
firewall, you'll find all the work is nearly on a CPU. As throughput
increase, that CPU will be full and no more throughput can gain,
though the other CPUs are still idle. With IFB-MQ, you can distribute
the packets among all the CPUs, and archive good throughput.

I know you worry about the latency. From my experiments, It just OK.
No much explicit latency is added.

-- 
Regards，
Changli Gao(xiaosuo@...il.com)
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html