netdev - Re: [PATCH net-next v5 0/4] Add support to do threaded napi busy poll

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: <CAAywjhSV93FBE+3oqQ4LyQ9GD8F8YMaiPnp4uTbcK9a2uGRStQ@mail.gmail.com>
Date: Wed, 30 Apr 2025 09:47:53 -0700
From: Samiullah Khawaja <skhawaja@...gle.com>
To: David Laight <david.laight.linux@...il.com>
Cc: Martin Karsten <mkarsten@...terloo.ca>, 
	Willem de Bruijn <willemdebruijn.kernel@...il.com>, Joe Damato <jdamato@...tly.com>, 
	Jakub Kicinski <kuba@...nel.org>, "David S . Miller" <davem@...emloft.net>, 
	Eric Dumazet <edumazet@...gle.com>, Paolo Abeni <pabeni@...hat.com>, almasrymina@...gle.com, 
	willemb@...gle.com, netdev@...r.kernel.org
Subject: Re: [PATCH net-next v5 0/4] Add support to do threaded napi busy poll

On Wed, Apr 30, 2025 at 5:37 AM David Laight
<david.laight.linux@...il.com> wrote:
>
> On Mon, 28 Apr 2025 14:05:17 -0400
> Martin Karsten <mkarsten@...terloo.ca> wrote:
>
> ...
> > That's not the same. The existing busy-poll mechanism still needs an
> > application thread. This thing will just go off and spin, whether
> > there's an application or not.
>
> I think this (and elsewhere) should be 'busy spin' not 'busy poll'.
> That would make it much more obvious that it really is a cpu intensive
> spin loop.
"Busy Poll" is synonymous to polling a socket/napi/queue while
spinning. For example it is used for socket busy poll and epoll busy
poll. Calling it "busy spin" would be confusing I think. I will add a
note in documentation, as suggested, to explain that it will spin a
core and depending on requirements one might want to share the core by
using affinity and priority configurations.
>
> Note that on some cpu all the 'cores' run at the same speed.
> So that putting one into a 'spin' will cause the frequency of all of
> them to increase - thus speeding up a benchmark.
>
> Rather the opposite of tests where a cpu busy thread (doing work)
> gets bounced around physical cpu - so keeps being run on ones
> running a low clock speed.
>
>         David