On Mon, Jun 12, 2023 at 03:30:23PM +0200, petter@xxxxxxxxxx wrote: >> Some time ago https://bugzilla.kernel.org/show_bug.cgi?id=217034 was >> created. From the beginning it was just about some error printouts. >> Then Andreas (who created the bug report) mentioned that it seems to >> work worse after bumping the firmware to > 9.9.10. After some fixes >> from Sascha the error printouts dissappeared. But when I also started >> to run this using firmware > 9.9.10 I also got problems. On my i.MX8 >> and RPi4 board it works fine, but on some of my less powerful boards >> such as and older RPi and my i.MX6 SoloX board, it always fails using >> 9.9.10 firmware. After some digging in the git log, I discovered <> that HW scan offload was introduced in a later firmware. So when I >> disable HW offload scan it seems to work again on all my boards. But >> still I want to understand why the HW offload scan don't work for >> me. >> >> Like described in the bug report I get below when running on latest >> 6.4 mainline with all relevant patches around rtw88 applied. >I can't reproduce this here. I am currently running v6.4-rc3 plus: > >wifi: rtw88: usb: silence log flooding error message > >I tested on a i.MX6S (not SoloX) board with Firmware 9.9.14. > >A "nmcli dev wifi rescan" works just fine and the link also continues to >work. > >I verified that FW_FEATURE_SCAN_OFFLOAD is set and used in the driver, >also that it's not set in Firmware 9.9.9. I also tried to put some >load on the link by running iperf3, still no difference. > >Sascha > Thanks for your valuable feedback, this finding made think in another direction about this. First I tried to reproduced the issue using the same setup 6.4-rc3 + 'wifi: rtw88: usb: silence log flooding error message' but instead of using NM, I used wpa_cli to perform the scan. This time it works fine. After some digging I realised that I had a business application still running, that uses libnm and among other things reacts on some callback during scan. So I agree with your finding that this doesn't seems to be related to the actual HW offload scanning after all, instead it seems like using nmcli + HW offload scan + our internal application using libnm might trigger the same behavior that I reported in: https://lore.kernel.org/linux-wireless/20230526055551.1823094-1-petter@xxxxxxxxxx/T/#t I will try to do some investigations if I can use this info to find a better way to reproduce that issue. However it's still the case that I cannot reproduce above problem on for example an i.MX8 using maxcpus=1. Thanks, Petter
