Issue seems to be from ecm_nss_ported_ipv4_connection_accelerate(). It's probably good to run gdb and identify where exactly the NULL pointer is used. This function is over 1000 lines of code tho.
If I have to guess, I would think the changes would be at the netfilter code causing issues for NSS ECM.
And dump the memory @ *swapper_pg_dir, looking for null pgd entries. I don't know if kernel error 17 is a simple null ptr dereference, or one that happens in memory management code.
Or maybe a kmalloc returning null, which goes unchecked.
Mmmv.
Luci theme -> good to know!
speedtesting on router -> oki doki. I tested it on my server and pc. Both with 900-950 MB/s. So worked!
5ghz -> I rebooted the router, restarted the interface and even poweroff the router. Unfortunately no changes. It is strange overall... My IOT devices and my samsung smartphone has no problems. My win 11 and my wifes samsung smartphone cant access
Good to hear!
The default driver / firmware is ath10k-ct rather than ath10k. If other versions of openwrt worked well for those clients you probably should try the other NSS version (the one without ath10k in the title). If that doesn’t work you can post your wireless settings here (minus your wifi password of course) and we can troubleshoot if there is something causing an issue.
Hello again,
so, now the snapshot with ath10-ct is rolled out.
Still, I got 950 MB/s download and the 5Ghz is now available for my devices.
Edit:
I monitor the performance of my r7800 with prometheus-node-exporter-lua and grafana. After re-installing all exporter packages, everything works, except data for wifi. In my grafana openwrt dashboard, there are no data for wifi
First of all a big thanks to all participants.
I want to use the nss build for my EA8500.
This router is mainly used as dedicated OpenVPN and WireGuard server and Wifi is turned off.
As I am often away and need this for access from abroad it need to be stable and working.
Current Master (non nss) build works good but I could use the extra nss speed.
Any recommandations?
Can I just use @ACwifidude 's master build or would I be better off with the stable build?
I use his 22.03 nss qsdk 10 build and it has been solid for me - it doesn't reboot unless i take a power hit.
Several tips for stability (some are guesses/folklore, some i have experience...)
B) set a single cpu clock rate - some crashes seem to be coming from dynamic adjustment of the clock; yes, we'd like it as a feature but not all use cases seem stable.
iii) mmmv.
M.
Thanks useful information, I have seen some testing with a watt meter and the difference between on demand and performance governor was very little to non existent.
As the router runs cool I think to let it just use performance governor then.
It's still changing the clock rates ... i just set min, max for both cores to the same value by echoing them to their correct entries in /proc.
In /etc/rc.local:
echo 1400000 > /sys/devices/system/cpu/cpufreq/policy0/scaling_min_freq
echo 1400000 > /sys/devices/system/cpu/cpufreq/policy1/scaling_min_freq
echo 1400000 > /sys/devices/system/cpu/cpufreq/policy0/scaling_max_freq
echo 1400000 > /sys/devices/system/cpu/cpufreq/policy1/scaling_max_freq
Got it , Thanks
Now, after a couple of days, I got a problem detected.
With @ACwifidude Snapshot version, I cant get access to my bookstack.
It is running on an ubuntu vm, managed with nginx proxy manager and routed to the internet. Got a "bad gateway" url response.
All port forwarding settings in openwrt are there, so i cant see whats the problem
will there be an update to the packages, it seems more than 1 month has passed 
Sounds like the hairpin traffic issue with NSS.
Try the hack first: set the bridge in the NSS router to promiscuous mode (can be done through luci or through shell). Scroll up thread for more about it.
If that works you might want to try enabling hairpin traffic in the kernel (patch details somewhere above), although that didn’t work well for me. YMMV.
For eth0 or eth1?
Neither?
For br.lan , in best of theories.
it worked! Ty 
Updated the 22.03 build yesterday. I had my main router running the last 22.03 version for 40 days uptime. Master had some build errors yesterday when I tried to build it. I'll try to troubleshoot it today and see what the deal is.
Found some time to upgrade C2600 access points. Thought of using the non-NSS image on one so I could compare the advantage of using the NSS build on an AP, as I think I read somewhere on this thread or the NSS one that NSS cores are not really helpful on access points. Placed both beside each other, different SSID, different channel, and sat about 20 ft away. Both running 22.03.5. Identical config restored to both but for WiFi channel, SSID and IP. Ran a few speed tests flipping from one to the other. Not the best methodology ever, but tried to make it apples to apples.
There is a clear benefit on throughput. Numbers fluctuated from test to test but were relatively consistent:
Recorded CPU usage during the test and graphed it out. First dip is the download part of the test, second the upload. The NSS build consumes noticeably less CPU as expected.
With all its quirks and kinks, definitely worth using these builds on access points.
I dusted off an ebay R7800 to verify the claims I read, and here are the results.
====== WAVEFORM.COM BUFFERBLOAT TEST RESULTS======
R7800 owrt 2203.5 sqm 450/28Mbps pocake ppo22
Test Version,1.0.8
Test ID,42f8e426-d7ea-427e-8e83-9e0a4147e168
Unix Time,1683316254663
====== RESULTS SUMMARY ======
Bufferbloat Grade,A+
====== RESULTS SUMMARY ======
Mean Unloaded Latency (ms),19.24
Increase In Mean Latency During Download Test (ms),0
Increase In Mean During Upload Test (ms),0
Download speed (Mbps),130.701
Upload speed (Mbps),26.726
Disappointed, I disabled downstream shaping to see how clunky the hardware is:
====== WAVEFORM.COM BUFFERBLOAT TEST RESULTS======
R7800 owrt 2203.5 sqm 0/28Mbps pocake ppo22
Test Version,1.0.8
Test ID,f53d4645-151a-4c8a-b367-0361bf4cd1d5
Unix Time,1683316578534
====== RESULTS SUMMARY ======
Bufferbloat Grade,A
====== RESULTS SUMMARY ======
Mean Unloaded Latency (ms),19.97
Increase In Mean Latency During Download Test (ms),27.13
Increase In Mean During Upload Test (ms),0
Download speed (Mbps),231.942
Upload speed (Mbps),26.701
Blockquote
I put the R7800 back in the shipping box and black markered the box with this:
R7800: 130/26 Mbps,
and went back to my R5S as main router and C20V5 as access point.
The R7800 didn't crash with owrt 2203.5 firmware.
