Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <20190221221941.29358-1-daniel@iogearbox.net> <20190222083131.5a141e23@carbon>
In-Reply-To: <20190222083131.5a141e23@carbon>
From:   =?UTF-8?B?QmrDtnJuIFTDtnBlbA==?= <bjorn.topel@gmail.com>
Date:   Tue, 26 Feb 2019 19:20:50 +0100
Message-ID: <CAJ+HfNhs_hBhHhNhaw9Aywh-bK7vyLAQ-t4sXada9xRpPfwnaA@mail.gmail.com>
Subject: Re: [PATCH] x86, retpolines: raise limit for generating indirect
 calls from switch-case
To:     Jesper Dangaard Brouer <brouer@redhat.com>
Cc:     Daniel Borkmann <daniel@iogearbox.net>, tglx@linutronix.de,
        LKML <linux-kernel@vger.kernel.org>,
        Netdev <netdev@vger.kernel.org>,
        "David S . Miller" <davem@davemloft.net>,
        =?UTF-8?B?QmrDtnJuIFTDtnBlbA==?= <bjorn.topel@intel.com>,
        Magnus Karlsson <magnus.karlsson@intel.com>,
        Alexei Starovoitov <ast@kernel.org>,
        Peter Zijlstra <peterz@infradead.org>,
        David Woodhouse <dwmw2@infradead.org>,
        Andy Lutomirski <luto@kernel.org>,
        Borislav Petkov <bp@alien8.de>,
        Linus Torvalds <torvalds@linux-foundation.org>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Fri, 22 Feb 2019 at 08:32, Jesper Dangaard Brouer <brouer@redhat.com> wr=
ote:
>
> On Thu, 21 Feb 2019 23:19:41 +0100
> Daniel Borkmann <daniel@iogearbox.net> wrote:
>
> > Recent work on XDP from Bj=C3=B6rn and Magnus additionally found that
> > manually transforming the XDP return code switch statement with
> > more than 5 cases into if-else combination would result in a
> > considerable speedup in XDP layer due to avoidance of indirect
> > calls in CONFIG_RETPOLINE enabled builds. On i40e driver with
> > XDP prog attached, a 20-26% speedup has been observed [0]. Aside
> > from XDP, there are many other places later in the networking
> > stack's critical path with similar switch-case processing. Rather
> > than fixing every XDP-enabled driver and locations in stack by
> > hand, it would be good to instead raise the limit where gcc would
> > emit expensive indirect calls from the switch under retpolines
>
> I'm very happy to see this.  Thanks to Bj=C3=B6rn for finding, analyzing =
and
> providing hand-coded-if-else code that demonstrated the performance
> issue for XDP.  But I do think this GCC case-values-threshold param is
> a better and more generic solution to the issue we observed and
> measured in XDP land. And hopefully other parts of the network stack
> and kernel will also benefit.
>
> Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
>
> Thanks for following up on this Daniel,

I definitely prefer a switch-statement over the if-else-messiness in
this context. Thanks for doing the deep-dive, Daniel!

FWIW,
Acked-by: Bj=C3=B6rn T=C3=B6pel <bjorn.topel@intel.com>

> --
> Best regards,
>   Jesper Dangaard Brouer
>   MSc.CS, Principal Kernel Engineer at Red Hat
>   LinkedIn: http://www.linkedin.com/in/brouer