search

greearb / ath10k-ct

Stand-alone ath10k driver based on Candela Technologies Linux kernel.
111 stars 40 forks source link

Wave1/qca998x - slow, unstable, DFS false positives... #144

Closed tj90241 closed 4 years ago

tj90241 commented 4 years ago

I'll gather proper data for a bug report shortly. Wanted to raise awareness that there looks to be a major regression in either Wave1/qca998x or the CT ath10k-5.7 module. Disclaimer: I've never tried CT variant of driver/firmware before... so maybe the issues have always been there - though I would imagine that Wave1 is fairly stable at this point.

Description of the problem (how to configure, how to reproduce, how often it happens): qca988x is configured as a 5ghz AP (under hostapd). Regardless of channel width (40/80), channel, or firmware used, I observed that all clients could pull no more than 5mbps down from the AP whereas the upstream ath10k driver/firmware remains rock solid stable and easily dishes >100mbps to clients in the same environment and under the same configuration. At the same time, clients could TX at about 50mbps to the AP with the CT firmware - strangely much more performant in that direction (though again... I routinely observe clients doing >100mbps in this direction with the upstream ath10k as well).

Moreover, the AP tended to crash within < 1min when running CT drivers/firmware and when iperf testing with clients. Tried several CT versions and variants with the exact same behavior. With a very limited number of clients, have not observed similar instability with upstream ath10k.

Lastly, (as also seen in the upstream firmware/driver - hence me prompting to try CT...) DFS consistently produces false positives when the airspace is dead silent and the first client turns on and attempts to auth. I tried all 80mhz channels and several 40mhz channels with the same result -- CAC consistently succeeds and the AP does not detect radar... until the moment a client powers on and attempt to auth with the AP.

Software (OS, Firmware version, kernel, driver, etc): Compiled kernel 5.7.8 (current upstream release) patched with 0001-wireless-Relax-beacon_int_min_gcd-and-ADHOC-check.patch. Some slight modifications to the ath driver so that it can emit IR on 5ghz US/FCC-approved channels. Same kernel with vanilla ath10k and 10.2.4-1.0-00047 firmware is spectacularly stable on non-DFS channels.

Hardware (NIC chipset, platform, etc): Compex WLE900VX (qca998x) in a Netgate RCC-DFF-2220 (Intel Atom C2338 SoC) running Debian Linux ("buster"). SMA pigtails off the mPCIe card to qty. (3) run of mill omni antennas off Amazon -- probably easily single digit dBi.

Logs (dmesg, maybe supplicant and/or hostap): Will gather tomorrow as I use the router as my daily driver and need to revert to ath10k upstream for the time being. Please let me know of anything useful outside of what's documented as part of a standard bug request.

greearb commented 4 years ago

I have found that my wave-1 firmware does not have as good of RF characteristics as upstream 10.2 firmware. I don't know why at this point, and I am mostly focused on wave-2 at this point. If wave-1 upstream works for you, then please use that. For what its worth, we see good performance on the WLE900VX, often near 900+Mbps in good RF conditions. In previous testing with rate-vs range on a 9882 (2x2 9880 basically), we got good numbers, but not quite as good as 10.2 upstream firmware.

tj90241 commented 4 years ago

OK, sounds good! Thankfully, it looks like Compex has some very affordable drop in replacements so I may very well do that.

greearb commented 4 years ago

We have had nice luck with the Compex 9984 single-band NICs. ath10k-ct firmware does not support the new dual-band 9984 NICs at this time....