[PATCH ath-next 1/3] wifi: ath10k: set the PMF peer flag on WCN3990

[Loic Poulain]

Hi Nagashima,

On Tue, Nov 3, 2026 at 5:45 PM Nagashima, Shunsuke <shunsuke at nagashi.ma> wrote:
>
> Hi Loic,
>
> On Tue, Nov 3, 2026 at 2:46 PM Loic Poulain wrote:
> > If Linux-firmware ships 01387, someone will need to confirm that the
> > MFP capability addition works with that specific firmware version. For
> > QCM2290, I can probably test it myself and provide a Tested-by tag if
> > you can describe the expected testing procedure.
> >
> > Yes, to Jeff.
>
> Thanks. Then I'll start with the qcm2290 file only, and leave the
> shared 01387 file until someone can test it on a db845c or an axolotl.
> Note that qrb4210/firmware-5.bin is a link to the qcm2290 file, so the
> change covers both.
>
> Test procedure:
>
> 1. With the current file, dmesg shows these features (crc32 a79c5b24):
>      wowlan,mgmt-tx-by-reference,non-bmi,single-chan-info-per-channel
>    Connect to a 5 GHz WPA3-SAE AP (PMF on) and run iperf3 to a wired
>    host, up and down (-R). Do the same with WPA2 on the same AP.
>
> 2. Make the new file (qca-swiss-army-knife ath10k-fwencoder):
>   F=wowlan,mgmt-tx-by-ref,non-bmi,single-chan-info-per-channel,mfp-support
>   cp firmware-5.bin firmware-5.mfp.bin
>   ath10k-fwencoder --modify --timestamp=1706616060 \
>       --features=$F firmware-5.mfp.bin
>    md5sum: 58982403c9c56a4b888fe0ab3104d7de. Only one byte changes
>    (offset 33, 0x00 -> 0x10). Without --timestamp the tool writes the
>    current time and the md5sum changes.
>
> 3. Put it in place of ath10k/WCN3990/hw1.0/qcm2290/firmware-5.bin and
>    reboot (ath10k reads the features only at probe). dmesg should show
>    these features (crc32 f9e8f071):
>      wowlan,mfp,mgmt-tx-by-reference,non-bmi,single-chan-info-per-channel
>
> 4. Repeat the iperf3 runs. Also disconnect and reconnect to the WPA3 AP
>    about 10 times, note how long each reconnect takes, and check dmesg
>    for "firmware crashed".
>
> On my comma four (WLAN.HL.2.0.c10-00459), adding only this bit to
> its firmware-5.bin, with the kernel unchanged, took WPA3 from 22.6-31.6
> Mbit/s up and 21.6-21.7 down to 273.8-287.7 up and 242.5-243.7 down,
> close to WPA2. Without the bit, ADDBA Requests went out unprotected and
> no Block Ack was set up, and with one AP a reconnect took about 24 s;
> with the bit it took 1-5 s.
>
> A failure would be an association failure or a firmware crash. If WPA3
> is already close to WPA2 with the current file, please tell me too:
> then this firmware may not need the bit, and the test only shows that
> it does no harm. A monitor-mode capture is optional; with the new file
> the ADDBA Request frames should be protected.
>
> If it works, I'll send the file to the ath10k list, to Jeff with you
> in CC, with your Tested-by.

It works for me on qcm2290 (Arduino Unoq), and greatly improves
throughput on my WPA3 AP.

Regards,
Loic