netdev - Re: [PATCH v3 net-next 2/4] net: Allow to use SMP threads for backlog NAPI.

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: <ae886b7975751a2c148fa4addce26c456678c735.camel@redhat.com>
Date: Tue, 05 Mar 2024 11:44:06 +0100
From: Paolo Abeni <pabeni@...hat.com>
To: Sebastian Andrzej Siewior <bigeasy@...utronix.de>
Cc: netdev@...r.kernel.org, "David S. Miller" <davem@...emloft.net>, Eric
 Dumazet <edumazet@...gle.com>, Jakub Kicinski <kuba@...nel.org>, Jesper
 Dangaard Brouer <hawk@...nel.org>, Thomas Gleixner <tglx@...utronix.de>,
 Wander Lairson Costa <wander@...hat.com>, Yan Zhai <yan@...udflare.com>
Subject: Re: [PATCH v3 net-next 2/4] net: Allow to use SMP threads for
 backlog NAPI.

On Tue, 2024-03-05 at 11:35 +0100, Sebastian Andrzej Siewior wrote:
> On 2024-03-05 11:08:35 [+0100], Paolo Abeni wrote:
> > 
> > Does not apply cleanly after commit 1200097fa8f0d, please rebase and
> > repost. Note that we are pretty close to the net-next PR, this is at
> > risk for this cycle.
> 
> will do.
> 
> > Side note: is not 110% clear to me why the admin should want to enable
> > the threaded backlog for the non RT case. I read that the main
> > difference would be some small perf regression, could you clarify?
> 
> I am not aware of a perf regression.

I probably inferred too much from the udp lookback case.

> Jakub was worried about a possible regression with this and while asking
> nobody came up with an actual use case where this is used. So it is as
> he suggested, optional for everyone but forced-enabled for RT where it
> is required.
> I had RH benchmarking this and based on their 25Gbe and 50Gbe NICs and
> the results look good. If anything it looked a bit better with this on
> the 50Gbe NICs but since those NICs have RSS…
> 
> I have this default off so that nobody complains and yet has to
> possibility to test and see if it leads to a problem. If not, we could
> enable it by default and after a few cycles and then remove the IPI code
> a few cycles later with absent complains.

I think this late in the cycle is better to keep backlog threads off by
default.

Thanks,

Paolo