Yes, this is caused by the newer ath10k drivers, firmware-5.bin_10.4-3.10-00047 (and 46 before) were running nicely with the 4.19 based ath10k for me before.
Hello all,
Has anyone tried putting the R7800 in monitor mode and sniffing for packets?
I have been experimenting with sniffing using the R7800; however, I seem to be dropping all the aggregate frames, and picking up control frames only. I suspect there's a setting to do with max AMSDU size in monitor mode that needs to be changed, but i'm not sure where this is.
Any advice/help would be appreciated!
Lyndon
Update: changing my entire network to Tx/Rx on 20MHZ bandwidth seems to eliminate the problem; i'm wondering if its an issue with my packets source, rather than any individual setting on the R7800
I read that suggestion on the dd-wrt forum as well. So, I set my radio1 networks to 20MHz as well. Seems less unstable now.
@shelterx -- where do we find said firmware and how do we update it? What about the driver, does that need to be updated to match firmware?
# opkg list | grep ath10k | grep -v \\-ct
1 Like
Or go here:
1 Like
Any advise on resolving flashing problem with R7800? Any firmware except the Netgear's one leads to bootloop. Can telnet to the router running the original firmware and run linux commands there.
Did you TFTP install Openwrt firmware or attempt to flash it through the Web UI?
Yes, I tried installing via TFTP the original 18.06.04 (openwrt-18.06.4-ipq806x-netgear_r7800-squashfs-factory.img). Also several custom builds. But only R7800-V1.0.2.62.img seems to work.
If you do IRQ affinity by hand, I've seen the recommendation to put both eth0 and eth1 IRQs on CPU1. Do eth0 and eth1 both handle WiFi? I don't understand how it all maps from VLANs to CPUs to IRQs.
Is the proper way to add this to rc.local?
echo 2 > /proc/irq/31/smp_affinity
echo 2 > /proc/irq/32/smp_affinity
The values you can set seem to be 1,2,3. What do the bits mean? 1=CPU0, 2=CPU1, 3=??? (either?) ?
They are bits. A bitmask telling the allowed cores.
(It might be more clear if there were more cores.)
1 Like
I just tried hnyman's latest -ct build (r10829-91c84e87c2). Unfortunately the same wifi issues with 2 of my clients 
Back to the "old" builds.
dmesg shows some crashes too, but it seems to recover from them.
I reported them on GitHub if anyone is interested in the full dmesg:
Not specifically for the R7800, but one of the latest trunk builds I tried (actually, the first 4.19 kernel I tried with ath10k (not -ct) and the fix for the crash on association (https://git.openwrt.org/?p=openwrt/openwrt.git;a=commitdiff;h=2ceee0e02362519f08be2a288bd8f42ca6cef257) had terrible WiFi performance. My device is an Archer C7 v2 (QCA988x).
For instance, going from a wireless laptop to a wired pc on local lan would not reach speeds above 100Mbps, where previous builds (kernel 4.14) could easily do 300Mpbs and up. Does this have anything to do with the "SWBA overrun on vdev 0, skipped old beacon" messages/bug?
At least partially. When monitoring CPU utilisation for a longer period, the mainline ath10k with "SWBA overrun" errors seemed to cause somewhat higher idle CPU load than the -ct version. Likely the "SWBA overrun" version also causes higher CPU load (and lower throughput) during higher traffic loads, too.
I used to have a lot of reboots with 18.6.1 and 18.6.2. Now with 18.6.4 everything works stable, never a problem. I rolled my own firmware with the builder though. Unfortunately my network has changed a lot so I'm not sure if the stability is because of my network setup has changed or if it's an actual improvement in Openwrt. (I have mostly the same devices (NAS'es, VM's etc), on my network but I now use a wireless AC bridge so I can move the devices away from the living room, if anything this puts even more stress on the wireless connection)
I briefly tried DD-WRT and it was stable too, but Openwrt just "feels" more right with this router, and had better performance and WIFI speeds. Tried an earlier KONG build of Openwrt, it was a little more stable, but not quite. Also had a lot of problems when logging onto the router, I sometimes had to switch browsers. (it wasnt necessarily the "https" option's fault, which KONG's build used, because I use it myself in my own build now without problem).
It might still have problems in a technical sense, which some of you could analyze through logs and "-CT" version or whatnot. But for all intents and purposes, it serves my needs.
Except one really strange thing, Outward (from the router through ethernet) Samba performance slows way down to 2MB/s when I'm using a Kernel with MUQSS CPU scheduler, building the same kernel but with CFS scheduler fixes that. I can also put an external switch in-between, that also fixes it. This is with an old Fast Ethernet chipset though, and I haven't tried to replicate this on other PC's.
Thanks. Any mention of what might be the cause? I'm not sure why it would suddenly not work at 80HT- could it be a buffering issue which causes the R7800 to drop high throughput packets?
What's the status on kernel 4.19 for the R7800? Although I haven't tried running 4.14 for a long time, I believe the issue with fragmented UDP responses causing reboots still exists. From what I've seen the fixes in 4.19 never made it into 4.14.
The patches aren't available in master yet, but I've been running them (and the changes to get USB working) for a couple of months now (nbg6817).
I've also been running 4.19 for some months now with @chunkeey's 4.19 patch, but I haven't applied any patches to fix USB since I don't use it. Does USB need more testing?