Netgear R7800 exploration (IPQ8065, QCA9984)

[tohojo]

@castiel652

My question is that can we use the tx rate with the airtime calculation utility in mac80211 to make airtime fairness work with ath10k-ct firmware + driver?

Sure, in principle that should be doable

[tohojo]

@quarky

What I did notice this round was that my R7800 was connected with many more clients compared to normal (which is 4 clients.) What happened was that I upgraded my main router's (Linksys E8450) firmware and all the Wi-Fi clients that were connected to the E8450 all connects to the R7800. That's when the slow down happened, and I get a lot more debug logs printed (refer to my earlier post in this thread.) Altho. it did recover somewhat (I got frustrated figuring out why and when for a short 30min jog) and when I returned, the router somehow 'recovered'. Wi-Fi is hit and miss after that.

So this happened even after you patched to code to not throttle multicast (the "false -> true" bit)?

[quarky]

It happen either with or without. I don’t think the patch does anything.

[slh]

@quarky

my R7800 was connected with many more clients compared to normal (which is 4 clients.)

Just out of curiosity, how many clients in total (roughly)? Just asking, because -ct defaults to a maximum station number of 32, which may be quickly reached with devices faking MAC addresses or roaming between 2.4 GHz and 5 GHz (stressing the 2.4 GHz radio more).

[quarky]

@slh

Just out of curiosity, how many clients in total (roughly)?

Around 9 clients. But I guess it’s over time that clients connect and disconnect that it builds up.

[D43m0n]

I also have 2 R7800’s and a C2600. Mostly Apple devices. The R7800’s run the non-ct driver/firmware because that is in my situation far more stable/performant than the -ct variant. But the C2600 runs the -ct driver because on that device my Apple devices are more happy with the -ct driver…

The main router (R7800) has both 2.4GHz and the 5GHz radios active with the same SSID’s. All other routers are configured as dumb AP’s with VLAN’s and only have the 5GHz radio active at a lower tx-power. My other devices are another R7800 a C2600 and a WNDR3700v2. The R7800’s and the C2600 all run the same NSS build from @ACwifidude, except the R7800’s run the non-ct driver and the C2600 runs with the ath10k-ct driver.

Like many said; “it depends…”

[shelterx]

Just so you know, you need log console messages enabled in some cases too, you don't get information about all crashes without it.

[hnyman]

@hnyman

the second PR 5057 packages the feature into kmods that can be optionally selected by any target. And it naturally also contains the R7800 DTS change.
(select kmod-ramoops and add the DTS change to the device)

The PR 5057 got merged today. The kmod is now defined, and the needed .DTS change for R7800 and XR500 is there, but the kmod is not selected automatically for the build.

You can include the ramoops crash log feature in your R7800 build by adding kmod-ramoops to the config. (or install it via opkg)

kernel crashes 'ramoops' are logged into files into RAM in /sys/fs/pstore. The files in RAM survive a warm reboot, but not a cold reboot.

[quarky]

@quarky

I'm putting a stop in my quest to find out why my R7800's Wi-Fi is unstable after 3-4 days.

Interim report. It's been more than 5 days since I reverted my R7800 back to the round-robin airtime scheduler. The issue I faced (i.e. high latency) did not happen since I reverted. I rebooted by main router's (E8450) due to firmware upgrade, which resulted in all its connected clients latching on to the R7800 and the R7800 purrs on without complains. Contrasting to the earlier test build (which is identical to the current one, except for the airtime algo.), the R7800 Wi-Fi traffic would start to crawl and latency shoots up.

Will continue to monitor and report back maybe in another week. Hopefully my R7800 will not decide to throw a fit and kernel panic.

[shelterx]

I'm back to normal master build now, I'll let you know if my wifi starts to crawl too...
If the router stays up long enough, you never know these days.

[shelterx]

Anyone get this with ath10k and firmware-5.bin_10.4-3.9.0.2-00156? It happens with many hours apart.
ath10k_pci 0000:01:00.0: received unexpected tx_fetch_ind event: in push mode

No idea what it means and I haven't noticed any issues...

[quarky]

Yeah, I see that in my R7800 running my 21.02 builds as well,

[shelterx]

@Ansuel
Any progess on the CPU scaling issue?

[Ansuel]

I still have to test some thing but nope... looks like the uptime increased from 2-3 days to 6-7 but still it looks like something wrong is still there

[wired]

Is there a way to get netconsole running in OpenWRT?
https://www.kernel.org/doc/Documentation/networking/netconsole.txt

[anon98444528]

is there an advantage to netconsole over openwrt's built in network logging?

I've tried openwrt's network logging, but for some kernel panics it won't catch anything (but neither did the serial console so I think it was just a difficult panic to catch). I'm hopeful about the ramoops kmod.

[anon98444528]

I just loaded a build from master today and lost all connectivity (wired and wireless) except for the serial console. I was testing some ath10k mods but i don't think that would kill wired access. This build also includes the ramoops kmod (modified for my device - works by the way).

This is an AP only config, nft -s list ruleset returns nothing, firewall is disabled; however, i have not updated my .config for fw4 yet (cause i'm not using the firewall).

Any event, I need to revert to an older build so kids can do their homework.

EDIT: i got it back with an image builder build from Feb 5. I'm not sure what is wrong, but I'll start with a minimal .config next time.

EDIT: After starting with a minimal .config and only adding back a subset of packages i normally would build (packages like strongswan or sqm were left out as I don't use them in AP mode), I got a working build. I'm not sure if this related to the recent fw4 upgrade or not. It could have been related to switching between builds with dsa, swconfig, and kernel 5.15.

[cydergoth]

QQ (Different topic): Is DSA DOA on the R7800 and should I look at different hardware?

[anon98444528]

No.

I suspect DSA is just delayed as it likely will need kernel 5.15 and there is not much desire to work on that yet. Also, the testing I posted about above are for the r7500v2 which is similar to the r7800 but the wifi chipset is different. I wouldn't worry too much about that until

1. an r7800 user verifies my test results (which are only for wifi) and
2. @ansuel give's up trying to fix it (I suspect not likely for a while yet)

[cydergoth]

If there is any way I can help, I can do casual testing as all my real work stuff is hardwired to the (OpenWRT x86) firewall