[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <b1f64c44-0d1c-480e-a272-fb017e7d8673@gmail.com>
Date: Wed, 4 Oct 2023 13:02:17 -0700
From: Florian Fainelli <f.fainelli@...il.com>
To: Jakub Kicinski <kuba@...nel.org>,
Vishvambar Panth S <vishvambarpanth.s@...rochip.com>
Cc: netdev@...r.kernel.org, linux-kernel@...r.kernel.org,
bryan.whitehead@...rochip.com, UNGLinuxDriver@...rochip.com,
davem@...emloft.net, edumazet@...gle.com, pabeni@...hat.com, andrew@...n.ch
Subject: Re: [PATCH net] net: microchip: lan743x : bidirectional throughuput
improvement
On 10/4/23 12:20, Jakub Kicinski wrote:
> On Wed, 27 Sep 2023 16:46:23 +0530 Vishvambar Panth S wrote:
>> The LAN743x/PCI11xxx DMA descriptors are always 4 dwords long, but the
>> device supports placing the descriptors in memory back to back or
>> reserving space in between them using its DMA_DESCRIPTOR_SPACE (DSPACE)
>> configurable hardware setting. Currently DSPACE is unnecessarily set to
>> match the host's L1 cache line size, resulting in space reserved in
>> between descriptors in most platforms and causing a suboptimal behavior
>> (single PCIe Mem transaction per descriptor). By changing the setting
>> to DSPACE=16 many descriptors can be packed in a single PCIe Mem
>> transaction resulting in a massive performance improvement in
>> bidirectional tests without any negative effects.
>> Tested and verified improvements on x64 PC and several ARM platforms
>> (typical data below)
>
> 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.
--
Florian
Powered by blists - more mailing lists