netdev - Re: [PATCH net-next] mlx4: optimize xmit path

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 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]

Date:	Sun, 28 Sep 2014 02:22:33 +0200
From:	Hannes Frederic Sowa <hannes@...essinduktion.org>
To:	Eric Dumazet <eric.dumazet@...il.com>
Cc:	Or Gerlitz <gerlitz.or@...il.com>,
	Alexei Starovoitov <ast@...mgrid.com>,
	"David S. Miller" <davem@...emloft.net>,
	Jesper Dangaard Brouer <brouer@...hat.com>,
	Eric Dumazet <edumazet@...gle.com>,
	John Fastabend <john.r.fastabend@...el.com>,
	Linux Netdev List <netdev@...r.kernel.org>,
	Amir Vadai <amirv@...lanox.com>,
	Or Gerlitz <or.gerlitz@...il.com>
Subject: Re: [PATCH net-next] mlx4: optimize xmit path

On Sun, Sep 28, 2014, at 02:05, Eric Dumazet wrote:
> On Sun, 2014-09-28 at 01:44 +0200, Hannes Frederic Sowa wrote:
> > Hi Eric,
> > 
> > On Sun, Sep 28, 2014, at 00:56, Eric Dumazet wrote:
> > > -       ring->cons += txbbs_skipped;
> > > +
> > > +       /* we want to dirty this cache line once */
> > > +       ACCESS_ONCE(ring->last_nr_txbb) = last_nr_txbb;
> > > +       ACCESS_ONCE(ring->cons) = ring_cons + txbbs_skipped;
> > > +
> > 
> > Impressive work!
> > 
> > I wonder if another macro might be useful for those kind of
> > dereferences, because ACCESS_ONCE is associated with correctness in my
> > mind and those usages only try to optimize access patterns.
> > Does OPTIMIZER_HIDE_VAR generate the same code?
> 
> 
> If we have 
> 
> ring->cons += txbbs_skipped;
> 
> Then compiler might issue a RMW instruction.
> 
> And this is bad in this case.
> 
> I really want to _write_ into this location, and its fast because I
> already have in ring_cons the content I fetched maybe hundred of
> nanoseconds before, or even thousand of nanoseconds before.
> 
> ACCESS_ONCE(XXXX) = y
> 
> Is not only for correctness.
> 
> It exactly documents the fact that we want to perform a single write.
> 
> I believe it is time that people understand how useful is this helper
> (Less than 700 occurrences in the whole kernel today, not including
> Documentation/*)

Understood, thanks.

For me ACCESS_ONCE was something which slowed down code till today. Also
I have the feeling that instruction scheduling in the compiler could do
a better job in some places...

Now I wonder if it is worth it playing around with the restrict keyword
and strict-aliasing in networking. ;)

Bye,
Hannes
--
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