[Techteam] Weekend report - Mitch Bradley

Luke Gorrie luke at bupsys.com
Sat Sep 26 19:29:34 EDT 2009


A fun week but less exciting than Mitch :-)

I started off looking into the bug report from Quanta that
they were unable to fs-update over the network (got a hang part-way).
I couldn't reproduce this on my normal setup (cross-cable), but I had
occasionally seen "funny" network behaviour in OFW, so I went back
and investigated some "that's funny.." stuff. Fixed (mostly Mitch) some
odd packet-loss behaviour from both USB dongles. I was about to
declare victory but my pegasus dongle has suddenly started acting up.
Some improvements overall, anyway :-)

Quanta's root problem, as Mitch said, looks like it was really CPU
overheating and nothing network-related. But it was useful inspiration
for digging in and fixing stuff.

More specifically:

* Added tcpdump-compatible packet tracing support to openfirmware:
    capture u:\trace.cap
  (This is very handy for network debugging of all kinds.)

* (Mitch actually solved this) Fixed a problem where the first
  ethernet packet sent to the device would usually be lost, causing
  e.g. a ping timeout or a few seconds' delay to a file download

* Fixed a couple of bugs in the Pegasus driver exposed by Mitch's use
  of ethernet loopback mode.

* Fixed a tiny bug in the openfirwmare 'pingd' ping-responder command.

P.S. I quite like this ethernet Media Independent Interface idea.
     Amazes me that the register-poking code to put an interface into
     loopback-test mode is identical on each USB ethernet adapter,
     because those MII registers are standardized.

More information about the Devel mailing list