netdev - Re: [PATCH] net: fix saving TX flow hash in sock for outgoing connections

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] [day] [month] [year] [list]

Message-ID: <1414805411.16849.104.camel@decadent.org.uk>
Date:	Sat, 01 Nov 2014 01:30:11 +0000
From:	Ben Hutchings <ben@...adent.org.uk>
To:	David Miller <davem@...emloft.net>
Cc:	eric.dumazet@...il.com, Sathya.Perla@...lex.Com,
	netdev@...r.kernel.org, therbert@...gle.com
Subject: Re: [PATCH] net: fix saving TX flow hash in sock for outgoing
 connections

On Wed, 2014-10-22 at 16:14 -0400, David Miller wrote:
> From: Eric Dumazet <eric.dumazet@...il.com>
> Date: Wed, 22 Oct 2014 12:09:56 -0700
> 
> > On Wed, 2014-10-22 at 18:35 +0000, Sathya Perla wrote:
> >> Agree. Are you suggesting that drivers should automatically
> >> register an XPS configuration? I thought it was upto the user
> >> to enable it...
> > 
> > Yep, search for netif_set_xps_queue()
> > 
> > (commit 537c00de1c9ba9876b9)
> > 
> > Look at commit d03a68f8217ea0349 for an example of how it can be done,
> > if user do not override this later.
> 
> Very few people know about this :-/
> 
> I only see 4 drivers adjusted to do this, it would be nice if this
> was more widespread.

It seems to require that the driver also sets IRQ affinity hints, which
I've never been very comfortable with.  Drivers either set queue n = CPU
n without regard for topology, or optimise for whichever
micro-architecture the vendor most cares about.  irqbalance then
slavishly follows these hints, so we have each driver (rather than the
administrator) setting policy.

Alternately the driver could create an irq_cpu_rmap for its TX interupts
and that could be used to select queues based on current affinity
(<http://thread.gmane.org/gmane.linux.network/186698>).  But we already
seem to have too many ways to do TX queue selection.

Ben.

-- 
Ben Hutchings
The world is coming to an end.	Please log off.

Download attachment "signature.asc" of type "application/pgp-signature" (812 bytes)