On Wed, 2020-04-08 at 10:17 +0200, Sebastian Reitenbach wrote:
> Hi,
> 
>  
> Am Dienstag, Februar 04, 2020 23:10 CET, schrieb Kurt Miller 
> <li...@intricatesoftware.com>: 
>  
> > 
> > On Mon, 2020-02-03 at 00:28 +0100, Sebastian Reitenbach wrote:
> > > 
> > > Hi,
> > > 
> > > I've two rock64 1 with 2GB memory, the other with 1GB memory. 
> > > Both with NetBSD-evbarm-aarch64-202001141930Z-rock64.img
> > > as well as the one with 2GB on OpenBSD -current, have super
> > > lousy network performance. Both are Rock64_V2.0 from 2017-0713.
> > > It's connected to Gigabit switch.
> > > 
> > > Related to it, I found this threat regarding similar bad
> > > networking on Linux:
> > > https://forum.pine64.org/showthread.php?tid=5043&page=4
> > > 
> > > so tried to set MTU to 1492 and also experimented with something
> > > even much smaller, but to no avail, I can't successfully download
> > > base66.tgz for example, it gets stuck after a few kb, until it eventually
> > > gives up.
> > > 
> > > ICMP pinging kind of works, about 5% packet loss.
> > > 
> > > I also tried different media options, forcing to 100MBit full and half 
> > > duplex,
> > > as well as 10MBit full and half duplex. Also tried to force the switch 
> > > down to
> > > 100 MBIT or 10 MBit. The situation improves, for example, I get much more 
> > > of a base66.tgz downloaded, but it still gives up in the end.
> > > 
> > > ifconfig dwge0 debug doesn't bring nothing new in dmesg.
> > > 
> > > I don't think so but, can I disable rx/tx like they say in the Linux 
> > > thread using ethtool?
> > > 
> > > Also this threat here talks about network problems with V2.0 board:
> > > https://forum.pine64.org/showthread.php?tid=7545
> > > 
> > > In this threat here:
> > > https://forum.pine64.org/showthread.php?tid=5712
> > > they play with values of allwinner,tx-delay-ps, maybe playing with
> > > tx-delay in ./linux-4.20/arch/arm64/boot/dts/rockchip/rk3328-rock64.dts 
> > > might help?
> > > 
> > > Can I set this somewhere before booting, without the need to rebuild the 
> > > dtb package,
> > > and recreate an image for testing?
> > > 
> > > I even found this post:
> > > https://sanisimov.com/2019/08/fixing-rock64-v2-gigabit-ethernet/
> > > where someone is fixing it by soldering on the board.
> > > Before I go down that rabbit hole, anyone with working (gigabit) network
> > > on the Rock64, which board version?
> > I took a look at the links you posted. A few of them indicate that
> > v2.0 board has an electrical issue as the root cause that was later
> > fixed in the v3.0 board. You may want to consider adding the resistor
> > as described by Sergey Anisimov in the last link.
> just to report back: I finally got around to do the soldering,
> afterward downloading a install67.iso without issues.
> 
> so, soldering definitely helps.

That's great. Thanks for following up.

The current options for Rock64 gen2 networking are to:

1) Downgrade to 100baseTX
  # ifconfig dwge0 media 100baseTX mediaopt full-duplex
2) Modifying the board as described here:
  https://sanisimov.com/2019/08/fixing-rock64-v2-gigabit-ethernet/

For Rock64 gen3 networking is solid without downgrading
or modifying the board, however you will need to use
u-boot from -current package v2020.04p0 or higher.

Regards,
-Kurt

> blubb# ifconfig
> dwge0                                                                         
>                                         
>               
> dwge0: flags=808843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST,AUTOCONF4> mtu 1500
>         lladdr 4e:a3:7d:d8:88:6b
>         index 1 priority 0 llprio 3
>         groups: egress
>         media: Ethernet autoselect (1000baseT full-duplex)
>         status: active
>         inet 10.0.0.220 netmask 0xffffff00 broadcast 10.0.0.255
> blubb# netstat -I dwge0  
> Name    Mtu   Network     Address              Ipkts Ifail    Opkts Ofail 
> Colls
> dwge0   1500  <Link>      4e:a3:7d:d8:88:6b   325652     0   202049     0     > 0
> dwge0   1500  10.0.0/24   10.0.0.220          325652     0   202049     0     > 0
> 
> cheers,
> Sebastian
> 

Reply via email to