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

[Tawheed Makhdoomi]

Hi,

A data point for the shared hw1.0 file. It is not the db845c/axolotl
test you asked for: this machine uses that firmware-5.bin with a newer
WLAN.HL.3.2.0 build, not 01387.

Hardware: Microsoft Surface Pro X (SC8180X, WCN3990 via ath10k_snoc),
firmware WLAN.HL.3.2.0.c8-00428 (wlanmdsp from the machine's Windows
install), kernel 7.4.0-rc2-next-20261109. The firmware-5.bin is the stock
ath10k/WCN3990/hw1.0/firmware-5.bin from linux-firmware (60 bytes, feature
IE 40 00 0c), so the driver logged

  features wowlan,mgmt-tx-by-reference,non-bmi

Access point: 5 GHz, 80 MHz, HE, WPA3-SAE with PMF required. A second
node of the same network without PMF served as the control.

Before, stock file:

  PMF AP      1.3-2.1 MB/s down, 1.7 MB/s up; ping to the gateway: about
              55% at 2-20 ms, the rest at ~105 and ~205 ms
  non-PMF AP  8.8 MB/s down (limited by that node's wireless backhaul);
              ping 2-27 ms, no 100 ms steps
  Windows on an identical machine, PMF AP: 30-33 MB/s, ping 1-2 ms

(MB/s is megabytes per second, HTTP download of a 20-50 MB file.)

With HTT debug the delay showed on both legs in ~102 ms quanta: tx
completion came either ~1 ms or ~100 ms after the tx, with ACK status and
no TCP retransmits. I had not found the reason; the unprotected ADDBA
described in this thread fits what I saw.

Things that made no difference on the PMF AP: power save on/off (the
driver does send the psmode disable), the ath10k thermal throttle state,
all nine board files from the Windows package, Bluetooth off, removing
the P2P device, IOMMU strict/lazy, CPU idle states.

After setting only the mfp bit in that file (feature IE 40 00 0c ->
40 10 0c), nothing else changed:

  features wowlan,mfp,mgmt-tx-by-reference,non-bmi
  station dump: MFP: yes

  PMF AP      10-29 MB/s down over ten runs after a reboot (Windows on
              the identical machine in the same minutes: 13-25 MB/s),
              4.6 MB/s up; 60 of 60 pings under 20 ms

No regression seen on the non-PMF node, and no firmware crash in a few
hours of use. I have not tested AP mode, WPA2 with optional PMF, or
2.4 GHz, and I have not yet run the iperf3 and reconnect steps of your
test procedure.

So on WLAN.HL.3.2.0.c8 the shared hw1.0 file works with mfp-support added.
I have two of these machines and can test a proposed update of that file,
and give a Tested-by for it, if that helps it along.

Thanks,
Tawheed