Garrett D'Amore wrote:
> On transmit, the best thing to do is pad with zeros.   (This is
> important to prevent a security leak where "short" packets can
> unintentionally contain data from another packet, possibly sent to a
> different host.)

Good point - I'll update.

I'm still trying to narrow down the cause for frame corruption on
transmit. Compared with similar drivers on Linux, *BSD, and even Masa's
ni driver I am not doing anything differently.

> On receive, you should never need to strip.  If the packet is less than
> the minimum size, you should probably just drop it on the floor, if the
> hardware doesn't already do that for you.

Unfortunately the RTL8029 doesn't do much in terms of hardware, however
I do have a check in place which increments the runt_errors kstat and
drops the packet. There is no logic in place to deal with padded packets.

> On another note, mgmt has asked me about your DNET gldv3 work -- can you
> provide a status update on it?

Certainly.

The GLDv3 conversion has been complete for quite some time now - I have
not encountered any regressions in nicdrv testing so far. Jason King had
a friend interested in the conversion and has been testing the changes
(64-bit variety) for the last few months with no issues as well.

To be honest, I had been holding off on submitting until I could make a
couple more passes with nicdrv to present a comprehensive set of test
results, however life, and re have interfered somewhat.

I will finalize all changes and make a last run of nicdrv testing by
next week to get ready for an RTI.

Steve
_______________________________________________
driver-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/driver-discuss

Reply via email to