Re: [PATCH net-next v19 00/13] ax88179_178a: Add support for AX88179A-based chips
From: Birger Koblitz
Date: Fri Oct 30 2026 - 11:43:16 EST
Thanks so much Emir for testing!
On 29/10/2026 1:58 am, Emir Beganovic wrote:
Hi Birger,I have seen similar behaviour with iperf3 for an AX88279 and a 2.5GBit link: in the first round, some 4-10 packets get dropped by the RTL8372-switch as a link partner and then no further frame drops happen. Looking into the switch's error counts, the drops are due to Flow Control. I was not able to investigate further, but I have seen the same behaviour with RTL8156/7/9-adapters using the r8152 driver and that switch. So that seems to be "normal" and the speed of the adapter in iperf3 is very close (about 2.35Mbit/s) to the theoretical limit.
I tested v19 on a Dell DA305 USB-C adapter. It has an AX88179A
(0b95:1790, bcdDevice 0x0200, FW 1.0.7.131) behind a VIA VL822 hub,
plugged into a Dell XPS 16 (Panther Lake). The peer is an Intel I225-V
behind a switch.
I built ax88179, libphy, phylink and ax88796b as modules for a 7.2.8
kernel. v19 needed two things on that base: a4e22e9253e7 and a backport
of genphy_{read,write}_mmd_c45() from the adin1140 series.
With the current driver in configuration 1 this adapter never gets a
link ("Failed to read reg index 0x0040: -32"). With v19:
- 1000/full link with flow control
- download 940-941 Mbit/s in nine runs. Upload was 937-941 in five of
nine and lower in the other four, and I couldn't find why. Eight
later runs with tracing on all averaged about 932.
- 1-byte TCP request/response around 190-200 us, against about 1.46 msThe device can be registered only once it has been switched from CDC to the vendor mode, that is why you see the message for the unregistered device. That you see it twice can be due to the fact that it has 2 active interfaces, see the explanation given in commit c67cc43 for Anker adapters.
with cdc_ncm on the same port (its default 400 us TX timer)
- switching itself to configuration 1 when the device comes up in the
CDC configuration
- 100M and 10M links
- VLAN with rx-vlan-offload on and off, and rx-vlan-filter dropping a
VID that isn't configured
- multicast filtering and allmulti
- suspend/resume (s2idle) and wake-on-LAN with a magic packet
Two small things I noticed:
- "Switching to vendor mode" is printed twice, once per CDC interface,
and before the netdev is registered, so it shows up as
"(unnamed net_device) (uninitialized)".
The devices I have print it only once and that also disappears once the device has been switched and stays plugged even if the module is loaded/unloaded.
- After a WoL wake the link goes down and up twice within about 2.5 s,I see this, too. It is not triggered in any way by the host or the link partner, I checked. It appears that this is due to the firmware on the device trying to optimize the link settings further, the out-of tree ASIX driver displays the same behaviour, although it works very differently.
where a normal resume shows a single link up. It comes back fine
either way.
Will add in upcoming v20.
I only have this one adapter, so nothing on the other chips.
Tested-by: Emir Beganovic <beganovic.emir@xxxxxxxxx>
Birger