Re: [Patch v3] net: mana: Batch ringing RX queue doorbell on receiving packets

[Paolo Abeni]

On Sun, 2023-07-02 at 20:18 +0000, Long Li wrote:
> > > > > > > > Subject: Re: [Patch v3] net: mana: Batch ringing RX
> > > > > > > > queue
> > > > > > > > doorbell
> > > > > > > > on receiving
> > > > > > > > packets
> > > > > > > > 
> > > > > > > > On Fri, 30 Jun 2023 20:42:28 +0000 Long Li wrote:
> > > > > > > > > > > > > > > > > > > > 5.15 and kernel 6.1. (those
> > > > > > > > > > > > > > > > > > > > kernels are longterm)
> > > > > > > > > > > > > > > > > > > > They need
> > > > > > > > > > > > > > > > > > > > this
> > > > > > > > > > > > > > > > > > > > fix to achieve the performance
> > > > > > > > > > > > > > > > > > > > target.
> > > > > > > > > > > > > > > > 
> > > > > > > > > > > > > > > > Why can't they be upgraded to get that
> > > > > > > > > > > > > > > > performance
> > > > > > > > > > > > > > > > target, and
> > > > > > > > > > > > > > > > all
> > > > > > > > > > > > > > > > the other goodness that those kernels
> > > > > > > > > > > > > > > > have? We don't
> > > > > > > > > > > > > > > > normally
> > > > > > > > > > > > > > > > backport new features, right?
> > > > > > > > > > > > 
> > > > > > > > > > > > I think this should be considered as a fix, not
> > > > > > > > > > > > a new
> > > > > > > > > > > > feature.
> > > > > > > > > > > > 
> > > > > > > > > > > > MANA is designed to be 200GB full duplex at the
> > > > > > > > > > > > start. Due
> > > > > > > > > > > > to
> > > > > > > > > > > > lack of
> > > > > > > > > > > > hardware testing capability at early stage of
> > > > > > > > > > > > the project,
> > > > > > > > > > > > we
> > > > > > > > > > > > could
> > > > > > > > > > > > only test 100GB for the Linux driver. When
> > > > > > > > > > > > hardware is
> > > > > > > > > > > > fully
> > > > > > > > > > > > capable
> > > > > > > > > > > > of reaching designed spec, this bug in the
> > > > > > > > > > > > Linux driver
> > > > > > > > > > > > shows up.
> > > > > > > > 
> > > > > > > > That part we understand.
> > > > > > > > 
> > > > > > > > If I were you I'd try to convince Greg and Paolo that
> > > > > > > > the
> > > > > > > > change is
> > > > > > > > small and
> > > > > > > > significant for user experience. And answer Greg's
> > > > > > > > question why
> > > > > > > > upgrading the
> > > > > > > > kernel past 6.1 is a challenge in your environment.
> > > > 
> > > > I was under the impression that this patch was considered to be
> > > > a
> > > > feature, 
> > > > not a bug fix. I was trying to justify that the "Fixes:" tag
> > > > was
> > > > needed. 
> > > > 
> > > > I apologize for misunderstanding this.
> > > > 
> > > > Without this fix, it's not possible to run a typical workload
> > > > designed for 200Gb
> > > > physical link speed.
> > > > 
> > > > We see a large number of customers and Linux distributions
> > > > committed
> > > > on 5.15 
> > > > and 6.1 kernels. They planned the product cycles and
> > > > certification
> > > > processes 
> > > > around these longterm kernel versions. It's difficult for them
> > > > to
> > > > upgrade to newer
> > > > kernel versions.

I think there are some misunderstanding WRT distros and stable kernels.
(Commercial) distros will backport the patch as needed, regardless such
patch landing in the 5.15 upstream tree or not. Individual users
running their own vanilla 5.15 kernel can't expect performance
improvement landing there.

All in all I feel undecided. I would endorse this change going trough
net-next (without the stable tag). I would feel less torn with this
change targeting -net without the stable tag. Targeting -net with the
stable tag sounds a bit too much to me.

Cheers,
Paolo