mkiss kernel driver problems

mkiss kernel driver problems

[Matti Aarnio]

Mostly the mkiss line discipline module seems to work, but..

  mkiss: ax0: crc mode is auto.
  mkiss: ax0: Switchting to crc-smack

that typo (extra 't') is obvious, but I am wondering,
why it has growing error counter:

ax0       Link encap:AMPR AX.25  HWaddr OH2MQK-1
          UP BROADCAST RUNNING NOARP  MTU:512  Metric:1
          RX packets:9325 errors:152 dropped:0 overruns:0 frame:0
          TX packets:0 errors:0 dropped:0 overruns:0 carrier:0
          collisions:0 txqueuelen:10
          RX bytes:622232 (607.6 KiB)  TX bytes:0 (0.0 b)

I am feeding this interface from my own software via pty interface,
where the client side is turned on to KISS line discipline.

My own software is not complaining about incomplete writes,
thus I wonder what the kernel is doing ?  Also my software
is sending on only AX.25 frames that it considers to be valid.

Does anybody remember how the  mkiss.c  is supposed to work?

  73 de Matti, OH2MQK

Re: mkiss kernel driver problems

[Matti Aarnio]

On Thu, Oct 01, 2009 at 02:58:09PM +0300, Matti Aarnio wrote:
> Mostly the mkiss line discipline module seems to work, but..
> 
>   mkiss: ax0: crc mode is auto.
>   mkiss: ax0: Switchting to crc-smack
> 
> that typo (extra 't') is obvious, but I am wondering,
> why it has growing error counter:

Aargh..  error rate is near 4/256 - meaning I forgot to KISS encode
the SMACK CRC bytes...  

But this Brown Paper Bag  type "user error" is no excuse for
that kernel typo.


Why do I need to feed AX.25 frames from user program to kernel
with SMACK code variant?  My  aprx  software reads async serial
ports itself, and does all manner of port re-init in case of
for example an USB serial dongle needing it.

To live with kernel AX.25 network, serial ports that are not
attached to kernel with kissattach/mkiss do need to be fed
to kernel.  However the feed method may sometimes write incomplete
frame to kernel, and having 1-(2^-16) change of spotting that
is better than without SMACK -- plain KISS has 0 change of spotting
incomplete frame.

What I really would like is to use socket to feed to kernel
AX.25 packets that I got from somewhere, but somehow it does
look like it is not possible at all.  I would do AX.25 socket
creation,  bind() it with callsign I want, and then to send()
using that socket _without_ destination address having ever
been defined.  But I do not think that is really possible, all
sockets are "application" side, never "network devices".
Or can somebody come up with a fancy way to use for example
the raw socket for this ?

That is, a socket which says "I am device N0NNN-5", and sends
raw AX.25 frame to kernel AX.25 "network."

Heck..  I would like to bridge AX.25 network using SCTP's
reliable datagram protocol (or IP/UDP).  The Bridge software
would reside in userspace, but how could I pass a packet
most easily to kernel without going via pty+kiss mess ?


> ax0       Link encap:AMPR AX.25  HWaddr OH2MQK-1
>           UP BROADCAST RUNNING NOARP  MTU:512  Metric:1
>           RX packets:9325 errors:152 dropped:0 overruns:0 frame:0
>           TX packets:0 errors:0 dropped:0 overruns:0 carrier:0
>           collisions:0 txqueuelen:10
>           RX bytes:622232 (607.6 KiB)  TX bytes:0 (0.0 b)
> 
> I am feeding this interface from my own software via pty interface,
> where the client side is turned on to KISS line discipline.
> 
> My own software is not complaining about incomplete writes,
> thus I wonder what the kernel is doing ?  Also my software
> is sending on only AX.25 frames that it considers to be valid.
> 
> Does anybody remember how the  mkiss.c  is supposed to work?

  73 de Matti, OH2MQK