[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <aLfxueDGLngEb7Rw@shredder>
Date: Wed, 3 Sep 2025 10:43:53 +0300
From: Ido Schimmel <idosch@...sch.org>
To: Ricard Bejarano <ricard@...arano.io>
Cc: Andrew Lunn <andrew@...n.ch>,
Mika Westerberg <mika.westerberg@...ux.intel.com>,
netdev@...r.kernel.org, michael.jamet@...el.com,
YehezkelShB@...il.com, andrew+netdev@...n.ch, davem@...emloft.net,
edumazet@...gle.com, kuba@...nel.org, pabeni@...hat.com
Subject: Re: Poor thunderbolt-net interface performance when bridged
On Tue, Sep 02, 2025 at 12:18:59PM +0200, Ricard Bejarano wrote:
> I'm afraid we don't just see it with TCP or with bridged traffic.
I wrote that it can happen with forwarded traffic, not necessarily
bridged traffic. Section 6 from here [1] shows that you get 900+ Mb/s
between blue and purple with UDP, whereas with TCP you only get around
5Mb/s.
> That was my original observation and so the title of the thread, but
> it happens at a much lower level, at the data link layer. We observed
> CRC checksum failures in link stats. You can see those in my May 27th
> message in the thread.
Assuming you are talking about [2], it shows 16763 errors out of 6360635
received packets. That's 0.2%.
> The last thing we tried was to force linearization of SKBs with fragments, to
> see if the problem is with how the driver puts those on the line, which might
> offset everything out of its place, making CRC checksums fail. I was not able
> to do it, however, as that would simply hang all my tests.
>
> Any ideas?
I suggest removing the custom patches and re-testing with TSO disabled
(on both red and blue). If this doesn't help, you can try recording
packet drops on blue like I suggested in the previous mail.
[1] https://lore.kernel.org/netdev/29E840A2-D4DB-4A49-88FE-F97303952638@bejarano.io/
[2] https://lore.kernel.org/netdev/8672A9A1-6B32-4F81-8DFA-4122A057C9BE@bejarano.io/
Powered by blists - more mailing lists