Re: [PATCH net] net: microchip: lan743x : bidirectional throughuput improvement

[VishvambarPanth.S]

On Wed, 2023-11-01 at 12:52 +0530, Vishvambar Panth S wrote:
> On Wed, 2023-10-04 at 13:09 -0700, Jakub Kicinski wrote:
> > EXTERNAL EMAIL: Do not click links or open attachments unless you
> > know the content is safe
> > 
> > On Wed, 4 Oct 2023 13:02:17 -0700 Florian Fainelli wrote:
> > > > Nobody complained for 5 years, and it's not a regression.
> > > > Let's not treat this as a fix, please repost without the Fixes
> > > > tag for
> > > > net-next.
> > > 
> > > As a driver maintainer, you may want to provide some guarantees
> > > to
> > > your
> > > end users/customers that from stable version X.Y.Z the
> > > performance
> > > issues have been fixed. Performance improvements are definitively
> > > border
> > > line in terms of being considered as bug fixes though.
> > 
> > I understand that, but too often people just "feel like a device
> > which
> > advertises X Mbps / Gbps should reach line rate" while no end user
> > cares.
> > 
> > Luckily stable rules are pretty clear about this (search for
> > "performance"):
> > https://docs.kernel.org/process/stable-kernel-rules.html
> > 
> > As posted it doesn't fulfill the requirements 
> 
> Thanks for your feedback. I apologize for the delayed response.
>  
> The data presented in the patch description was aimed to convince a
> reviewer with the visible impact of the performance boosts in both
> x64
> and ARM platforms. However, the main motivation behind the patch was
> not merely a "good-to-have" improvement but a solution to the
> throughput issues reported by multiple customers in several
> platforms.
> We received lots of customer requests through our ticket site system
> urging us to address the performance issues on multiple kernel
> versions
> including LTS. While it's acknowledged that stable branch rules
> typically do not consider performance fixes that are not documented
> in
> public Bugzilla, this performance enhancement is essential to many of
> our customers and their end users and we believe should therefore be
> considered for stable branch on the basis of it’s visible user
> impact.
>  
> Few issues reported by our customers are mentioned below, even though
> these issues have existed for a long time, the data presented below
> is
> collected from the customer within last 3 months. 
> 
> Customer-A using lan743x with Hisilicon- Kirin 990 processor in 5.10
> kernel, reported a mere ~300Mbps in Rx UDP. The fix significantly
> improved the performance to ~900Mbps Rx  in their platform.
> 
> Customer-B using lan743x with v5.10 has an issue with Tx UDP being
> only
> 157Mbps in their platform. Including the fix in the patch boosts the
> performance to ~600Mbps in Tx UDP.
> 
> Customer-C using lan743x with ADAS Ref Design in v5.10 reported UDP
> Tx/Rx to be 126/723 Mbps and the fix improved the performance to
> 828/956 Mbps.
> 
> Customer-D using lan743x with Qcom 6490 with v5.4 wanted improvements
> for their platform from UDP Rx 200Mbps. The fix along with few other
> changes helped us to bring Rx perf to 800Mbps in customer’s platform
>  
> This is a kind request for considering the acceptance of this patch
> into the net branch, as it has a significant positive impact on users
> and does not have any adverse effects.
>  
> Thanks,
> Vishvambar Panth S
>  
> 
It has come to my attention that some people may not have received my
whole reply dated Nov 1st (as per
https://patchwork.kernel.org/project/netdevbpf/patch/20230927111623.9966-1-vishvambarpanth.s@xxxxxxxxxxxxx/#25577895
), possibly due to a non-ASCII character at the cut-off point.
Therefore, I am resending the part that was cut short below.
 
Jakub, would it be possible for you to apply the patch to the net
branch given the additional justification now posted below?
 
-----
 
Thanks for your feedback. I apologize for the delayed response.

The data presented in the patch description was aimed to convince a
reviewer with the visible impact of the performance boosts in both x64
and ARM platforms. However, the main motivation behind the patch was
not merely a "good-to-have" improvement but a solution to the
throughput issues reported by multiple customers in several platforms.
We received lots of customer requests through our ticket site system
urging us to address the performance issues on multiple kernel versions
including LTS. While it's acknowledged that stable branch rules
typically do not consider performance fixes that are not documented in
public Bugzilla, this performance enhancement is essential to many of
our customers and their end users and we believe should therefore be
considered for stable branch on the basis of it’s visible user impact.
Few issues reported by our customers are mentioned below, even though
these issues have existed for a long time, the data presented below is
collected from the customer within last 3 months.
 
Customer-A using lan743x with Hisilicon- Kirin 990 processor in 5.10
kernel, reported a mere ~300Mbps in Rx UDP. The fix significantly
improved the performance to ~900Mbps Rx  in their platform.
 
Customer-B using lan743x with v5.10 has an issue with Tx UDP being only
157Mbps in their platform. Including the fix in the patch boosts the
performance to ~600Mbps in Tx UDP.
 
Customer-C using lan743x with ADAS Ref Design in v5.10 reported UDP
Tx/Rx to be 126/723 Mbps and the fix improved the performance to
828/956 Mbps.
 
Customer-D using lan743x with Qcom 6490 with v5.4 wanted improvements
for their platform from UDP Rx 200Mbps. The fix along with few other
changes helped us to bring Rx perf to 800Mbps in customer’s platform

This is a kind request for considering the acceptance of this patch
into the net branch, as it has a significant positive impact on users
and does not have any adverse effects.

Thanks,
Vishvambar Panth S