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;
Subject: Re: recvfrom/recvmsg performance and CONFIG_HARDENED_USERCOPY
To:     Paolo Abeni <pabeni@redhat.com>,
        Eric Dumazet <eric.dumazet@gmail.com>,
        David Laight <David.Laight@ACULAB.COM>,
        network dev <netdev@vger.kernel.org>,
        "x86@kernel.org" <x86@kernel.org>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>
References: <23db23416d3148fa86e54dccc6152266@AcuMS.aculab.com>
 <dc10298d-4280-b9b4-9203-be4000e85c42@gmail.com>
 <8b8a3cc1c3341912e0db5c55cd0e504dd4371588.camel@redhat.com>
From:   Eric Dumazet <eric.dumazet@gmail.com>
Message-ID: <fbac8306-247f-10f3-4067-14c0610b17d6@gmail.com>
Date:   Fri, 6 Dec 2019 08:19:03 -0800
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101
 Thunderbird/68.2.2
MIME-Version: 1.0
In-Reply-To: <8b8a3cc1c3341912e0db5c55cd0e504dd4371588.camel@redhat.com>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


On 12/6/19 8:09 AM, Paolo Abeni wrote:

> Oh, nice! I though the compiler was smart enough to avoid the indirect
> call with the current code, but it looks like that least gcc 9.2.1 is
> not.
> 
> Thanks for pointing that out!
> 
> In this specific scenario I think the code you propose above is better
> than INDIRECT_CALL.
> 
> Would you submit the patch formally?

Certainly, although I am not sure this will be enough to close
the gap between recvmsg() and recvfrom() :)

Also I was wondering if a likely() or unlikely() clause would
make sense.

This could prevent an over zealous compiler optimizer
to put back the indirect call that we tried to avoid.