netdev - Re: r8169: Performance regression and latency instability

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: <792d3a56-32aa-afee-f2b4-1f867b9cf75f@applied-asynchrony.com>
Date:   Fri, 16 Aug 2019 15:59:21 +0200
From:   Holger Hoffstätte <holger@...lied-asynchrony.com>
To:     Eric Dumazet <eric.dumazet@...il.com>,
        Juliana Rodrigueiro <juliana.rodrigueiro@...ra2net.com>,
        netdev@...r.kernel.org
Cc:     hkallweit1@...il.com
Subject: Re: r8169: Performance regression and latency instability

On 8/16/19 2:35 PM, Eric Dumazet wrote:
..snip..
> I also see this relevant commit : I have no idea why SG would have any relation with TSO.
> 
> commit a7eb6a4f2560d5ae64bfac98d79d11378ca2de6c
> Author: Holger Hoffstätte <holger@...lied-asynchrony.com>
> Date:   Fri Aug 9 00:02:40 2019 +0200
> 
>      r8169: fix performance issue on RTL8168evl
>      
>      Disabling TSO but leaving SG active results is a significant
>      performance drop. Therefore disable also SG on RTL8168evl.
>      This restores the original performance.
>      
>      Fixes: 93681cd7d94f ("r8169: enable HW csum and TSO")
>      Signed-off-by: Holger Hoffstätte <holger@...lied-asynchrony.com>
>      Signed-off-by: Heiner Kallweit <hkallweit1@...il.com>
>      Signed-off-by: David S. Miller <davem@...emloft.net>

It does not - and admittedly none of this makes sense, but stay with me here.

The commit 93681cd7d94f to net-next enabled rx/tx HW checksumming and TSO
by default, but disabled TSO for one specific chip revision - the most popular
one, of course. Enabling rx/tx checksums by default while leaving SG on turned
out to be the performance issue (~780 MBit max) that I found & fixed in the
quoted commit. SG *can* be enabled when rx/tx checkusmming is *dis*abled
(I just verified again), we just had to sanitize the new default.

An alternative strategy could still be to (again?) disable everything by default
and just let people manually enable whatever settings work for their random
chip revision + BIOS combination. I'll let Heiner chime in here.

Basically these chips are dumpster fires and should not be used for anything
ever, which of course means they are everywhere.

AFAICT none of this has anything to do with Juliana's problem..

-h