[ECOS] Interfacing directly to the low level ethernet driver, how??
Mon Jul 2 12:19:00 GMT 2007
-----BEGIN PGP SIGNED MESSAGE-----
Michele Paselli wrote:
> Thanks guys, since in my specific application I don't need any other
> networking stacks I think I'll start implementing the I/O ethernet
> driver without any synchronization. My only concern is about Redboot,
> which also has a small networking layer. May I have problems with it
> if I don't synchronize packets? Of course I guess that then I'll not
> be able anymore to debug my system with ethernet but I can always do
> it with serial. Also, in my case I need to be extra fancy, because I
> have to receive ethernet packets in promiscouos mode, so even if the
> destination address in the packet is different from the one of the
> receiver one.
> Grant, I guess your driver will be built on top of the device specific
> one, so it will not be so different from mine. If your employer allows
> you, I would be grateful if you could contribute it, otherwise thanks
> anyway for your help.
I don't understand what all the hoopla is about this. The BSD network
stack provides for SOCK_RAW - why isn't this good enough? (Note: I
know it works, at least at some level, because the 'ping' tests all
work and they use RAW sockets!!)
> On 2007-06-28, Gary Thomas <email@example.com> wrote:
>>> It's a pretty thin layer -- it just allows you to queue up
>>> outbout packets with cyg_io_write() and read from a queue of
>>> inboung packets (with a specified protocol type) using
>>> Using RAW sockets would be nice, but adding a little code to
>>> an in-house driver is logistically easier than adding raw
>>> socket support to an "off-the-shelf" network stack and then
>>> turning around and doing it all again a couple years later
>>> when the network stack changes.
>> Your comments, while they make sense about eCos in general,
>> aren't helping.
> Sorry. I just wanted to point out that what I described is
> actually pretty simple.
>> I want to know why Michele thinks he needs to write his own
>> stack (that's what his questions were about).
>> Do you have your cyg_io code? Can you contribute it?
> I'll check with my employer.
> All you do is register the Ethernet driver as a normal "cyg_io"
> style driver and add syncronization so that simultaneous
> "write" operations from the network stack and from
> cyg_io_write() don't trip over each other. If you want to be
> extra fancy, you can add a receive queue for the custom
> protocol packets. The code is all Ethernet device specific, so
> I'm not sure how much help it would be to contribute it.
>> As for the network stack changing - I don't see that happening
>> anytime soon. The last time was 5 years ago and there's not a
>> great impetus for change now. It makes sense to me to fix
>> things that are missing or broken, rather than inventing new
>> ways of doing things.
> I agree. If we were starting now, that's probably what I'd
> try first.
> But, 7 years ago we had no experience with either eCos or
> either of the BSD network stacks, so adding a few (OK, maybe
> 50-100) lines of code the the Ethernet driver seemed like the
> safest way to go, since it didn't require us to get up to speed
> on NetBSD stack internals, and there was no danger of having to
> maintain a forked network stack. It also allowed us to
> implement a very low overhead zero-copy mechanism for raw
> ethernet I/O in a product where network stack overhead was by
> far the most significant bottleneck (I also spent several weeks
> writing and tweaking an assembly-language IP checksum routine).
Gary Thomas | Consulting for the
MLB Associates | Embedded world
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.7 (GNU/Linux)
Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org
-----END PGP SIGNATURE-----
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss
More information about the Ecos-discuss