Re: Fwd: Wrong bytes for DSA To_CPU code.

[Shady Atef via tcpdump-workers <tcpdump-workers () lists tcpdump org>]

> --- Begin Message ---
>
>
> From: Shady Atef <shadyoatef () gmail com>
>
> Date: Wed, 28 Jul 2021 16:45:18 +0200
>
> Thank you for the clarifications, Andrew.
>
> I have an extra question, does the CPU ports use Priority flow control
> (PFC) using the PRI field in the DSA  tag ?
> Shall a device sending DSA frames to another one react to pause frames and
> so on ?
> I think no, it doesn't make sense for me. But I can't find anything the
> confirms or denies that.
>
> Thanks.
>
> On Wed, Jul 28, 2021 at 2:27 AM Andrew Lunn <andrew () lunn ch> wrote:
>
> > On Sun, Jul 25, 2021 at 11:11:41PM +0200, Shady Atef via tcpdump-workers
> > wrote:
> > > Date: Sun, 25 Jul 2021 23:11:41 +0200
> > > From: Shady Atef <shadyoatef () gmail com>
> > > To: tcpdump-workers () lists tcpdump org
> > > Subject: Fwd: Wrong bytes for DSA To_CPU code.
> > >
> > > I think the following page on the website has something wrong.
> > > It lists b17,b16 and b12 as the code field for the To_CPU frames. While
> > in
> > > the `print_dsa.c` or the Linux kernel implementation, b18, b17 and b12
> > are
> > > used.
> >
> > The kernel gets it correct.
> >
> > > Another question, I work in a team building verification tool for
> > Ethernet
> > > designs.
> > > Is there a method to differentiate between the normal Ethernet frames and
> > > DSA-tagged one ?
> >
> > By looking at the frame itself? You can detect EDSA frames, since they
> > have an Ethertype value which indicates EDSA. But the value is not
> > registered with IEEE as far as i know. And the switches allow the
> > value to be configured, so a vendor crap driver could use a different
> > value. DSA you cannot detect in a reliable way. Maybe using heuristics
> > you can hunt around for something that looks like an IP header, see
> > that all frames use the same offset, and imply there is some extra
> > header in there. But i think it will be hard to say what that header
> > is. If you have looked at the tag drivers in Linux, you will notice
> > DSA is not the only tagger to use 4 bytes.
> >
> > If you are dong the capture on Linux, you can determine what tagging
> > protocol is in use of an interface from a file under
> > /sys/class/net/.../tagging.
> >
> > > In other words, can a port multiplex between the two types
> > > ? or each port in a DSA device is known to utilize tagging or not ?
> >
> > Marvell SOHO switches using {E}DSA will have all frames using
> > {E}DSA. It is possible at runtime in Linux to configure which of
> > {E}DSA is used, depending on the hardware capabilities, but you won't
> > see a stream of frames mixing them.
> >
> > When you are talking about 'Linux DSA', as opposed to 'Marvell DSA',
> > the majority of switches will have all frames with tags. However it is
> > not a hard rule. The sja1105 does some funky things for example.
> >
> >     Andrew
>
>
> -- 
> *Regards*
> *Shady Atef .*
>
> --- End Message ---
_______________________________________________
tcpdump-workers mailing list
tcpdump-workers () lists tcpdump org
https://lists.sandelman.ca/mailman/listinfo/tcpdump-workers