[OpenWrt-Devel] Pending ath79 issues on ar9342 and 4.14/4.19 kernels

Petr Štetiar ynezz at true.cz
Fri Jan 4 07:11:17 EST 2019


Weedy <weedy2887 at gmail.com> [2019-01-03 19:07:19]:

Hi,

> 5. A general performance loss going from ar71xx/4.9 to ath79/4.14
> I used to be able to rsync stuff 7-10MiB/s over WiFi and now I'm at
> 4-6MiB/s (5ghz, 2.4ghz has always been crowded and shit)

Is it reported as a bug/regression somewhere already with more details, like
what device (SoC) and client is that happening on? If not, and you would like
to get it fixed, you need to create proper bug report.

You're probably comparing apples and oranges here anyway, can you please
rather compare latest snapshots, ar71xx/4.14 and ath79/4.14? Then if you find
out, that the performance is same and the regression is somewhere between
ar71xx/4.9 and ar71xx/4.14, you should probably find out (git bisect?) where
the regression was introduced and someone might be able to look into it deeper
if it could be confirmed and reproduced.

I'm using ar71xx/4.14 snapshots at home on XD3200 and don't see any issues
there, but I've just 20/2 internet uplink so I wouldn't probably even notice
such performance degradations anyway :-)

> This might be your #4

This one should be probably fixed already, so if you're able to compile your
own image, then you can try that patch[1] and see if it improves your uptime.

> I haven't been really able to catch it in the act.

It happens mainly (or I was only able to reproduce it reliably) during device
reconfiguration, using just `ifup` command. So maybe if your DMA engine got
stuck due to #3, your reset script could have performed `ifdown/ifup` which
could lead to that tx queue timeout issue.

> I had some crap in notepad that I forgot to send to the list last month

If the problems aren't reported, don't expected them fixed. It's preferred to
use bugs.openwrt.org for bug reports, use IRC/mailing list if you're unsure
that it's bug, since you're saying it "panics", it's likely a bug.

> but "NETDEV WATCHDOG" sounds familiar.

Well it could be caused by #3 as well, I can provide patch for that as well if
you want to test it.

1. https://patchwork.ozlabs.org/patch/1020158/

-- ynezz

_______________________________________________
openwrt-devel mailing list
openwrt-devel at lists.openwrt.org
https://lists.openwrt.org/mailman/listinfo/openwrt-devel



More information about the openwrt-devel mailing list