10.2 firmware going out for lunch with 11a/11bg iperf tests

[Adrian Chadd]

hiya,

I've been porting ath10k to freebsd. I have it up and running, but
only in 11abg mode - once that's stable, I plan on doing 11n, then
11ac.

So, I hit a fun roadblock. If I do an iperf test (TCP or UDP TX) then
the firmware/NIC goes out to lunch. It's not a target assert with a
firmware dump - the whole NIC detaches from the PCI bus. I have to
completely reprobe everything from scratch to get it back alive.

I've also verified this on linux-4.4 w/ firmware 10.2.4.70.9-2 and the
latest 10.2.4 drop in Kalle's branch. If I configure it up in STA mode
and have it associate to an 11a AP, the firmware/NIC goes to lunch
pretty quickly and reproducably.

Amusingly, Ben's CT firmware doesn't have the bug - it looks like he
fixed it in a commit. I tested stock 10.1 from kalle's tree and an
early 10.1 CT firmware and both died the same way.

Are any QCA devs out there? I believe Ben and I have a fix for you,
but I can't build/test 10.2 firmware. :(

Thanks!



-adrian