Date: Fri, 28 Nov 2025 08:51:32 -0800 From: Adrian Chadd <adrian@freebsd.org> To: Florian Smeets <flo@freebsd.org> Cc: FreeBSD Net <freebsd-net@freebsd.org>, freebsd-current <freebsd-current@freebsd.org> Subject: Re: looking for testers for if_rge - RTL8125/8126/8127 ethernet driver Message-ID: <CAJ-Vmo=nj8TvjMg_XFkNZ6MC28DcOzdKKqVOqboCr0T7eTQO-g@mail.gmail.com> In-Reply-To: <cfd43166-a287-4ea0-97f5-dbbf08856bdc@FreeBSD.org> References: <CAJ-VmonP4%2BQmpSUfM6Xatta83e6zkjMShzL7ob1nJVOH=Tjwhg@mail.gmail.com> <cfd43166-a287-4ea0-97f5-dbbf08856bdc@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Thu, 27 Nov 2025 at 10:13, Florian Smeets <flo@freebsd.org> wrote:
>
> On 23.11.25 03:16, Adrian Chadd wrote:
> > hi!
> >
> > i've ported Kevin Lo's openbsd driver for these realtek chipsets to FreeBSD.
> > It works well enough for me to use on my laptop w/ RTL8125B / Killer E3000.
> > I'm now opening it up to others who are willing to build/run a kernel
> > module to test the driver out and report back.
> >
> This is great. Finally, an in tree driver for these very common NICs.
> The 1100.00 version of the net/realtek-re-kmod was just unreliable for
> me (constant hangs, no matter which options I turned off and on). I've
> only done light testing with the official 1101.00 driver. I was able to
> wedge it with less than a minute of iperf3, and the ifconfig down/up
> dance that was able to revive the interface with 1100.00 was not able to
> recover the interface.
>
> I ran if_rge on my NAS and did some testing. I haven't had one hang with
> this driver, even after pounding the network for hours. That's a big
> plus for me. Thanks.
>
> I was able to achieve close to 2.5Gb/s TX and close to 1Gb/s RX with
> iperf3 --bidir.
>
> CPU usage appears to be substantially higher than with the official
> Realtek driver.
That's a good data point.
>
> [intr{irq59: rge0}] goes to around 50% of one core, and [kernel{rge0
> taskq thread}] hovers between 20-25% when running the above iperf3 tests.
>
> With the official 1101.00 driver, the only process using > 1% CPU is
> this one [kernel{re0 taskq}] and it is around 10% with the test
> mentioned above.
I'll go dig into that a bit. It shouldn't be taking very much CPU to process
this number of packets; the bulk of the CPU should be used by the IP stack.
I'll go run some profiling over the next few days and see if I can nail down
what I'm doing poorly. Hopefully it's something stupid on my end. ;-)
-adrian
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CAJ-Vmo=nj8TvjMg_XFkNZ6MC28DcOzdKKqVOqboCr0T7eTQO-g>