Overflow issue on RT-AC58U (IPQ4018)


On this router with active use of the Internet, there is a problem of memory overflow. After that, the router stops responding, although the radios continues to work.

Build version: checked on all daily builds starting from October 2018 up to SNAPSHOT r9106-bbcd063(krnl:4.14.94) (and I'm sure on today will be the same).
Installed packages: clean fresh build + removed wpad-basic & installed wpad + luci.
Settings: Static IP on WAN (manually customized ipaddr, netmask, gateway, dns by ISP; bandwidth limits on 20Mbps (up/down)), DHCP on LAN (3 static leases + 3 dhcp clients), all radios is on.
Issue: regardless of the type of client connection (LAN\WiFi), when I try to download a file larger than 400-600MB, Lucy first stops working (with different messages about lack of memory), then the router simply stops responding to any requests from any interface. The same problem may can appears after 16-24 hours without rebooting.

In my opinion, there is a problem with memory leaks or incorrect work with free RAM.

I want to solve this problem. I am ready to help in debugging. (there are 5 such devices at hand)

For the initial debugging, aside from posting kernel logs (dmesg), I'd suggest to test switching to ath10k instead of ath10k-ct (in particular the kmod packages); background, kmod-ath10k might contain patches to reduce its memory footprint, which might, or may not have made it to kmod-ath10k-ct. In general, for a device with two ath10k radios, 128 MB are pretty borderline.

1 Like

@slh I understand you. Now I will reassemble build on ath10k (+kmod-ath10k) and testing it. I will post logs a bit later.

On the original ath10k-firmware and kmod-ath10k everything works fine. Tested on releases: r9248 (krnl: 4.14.96); r9270 (krnl: 4.14.97). Uptime: 60 hours, 14(7+7) wireless clients, 1 wired. Under heavy load, the amount of free memory was 40-60% of 128Mb.
Candela Technologies in its implementation of the firmware has problems with the use of memory. But I will wait for new releases of *-ct and test them.

Issue resolved. @slh thank you!


Keep in mind that 128 MB RAM will remain marginal for a device with two ath10k{,-ct} regardless of that mitigation attempt.

This topic was automatically closed 10 days after the last reply. New replies are no longer allowed.