[openbeosnetteam] Daily kernel update

Good morning again!

Well, another day, another euro!

Daniel Reinhold pointed out that we weren't linking against the correct
libnet.so in the test app, so one of the fixes on my machine is to fix that
problem. he's concerned that I've opted to use rpath's to fix the lib's
location as a temporary development fix. How do others feel about this?

I've spent a lot of today trying to find some bugs in the server and have
come to the conclusion I really don't like kernel debugging much, though
it's slowly making more sense. also the bug quota is dropping off nicely.

I've actually had to stop using the old net_server alongside the new one as
they don't cohabit very well, and to use the old one you'll need to remove
the /dev/net/socket link or it'll segfault for some reason.

I've also radically altered the way we register modules. We now simply call
an init routine which adds the necessary domain and protocol information to
the core as required. The method isn't perfect but does seem to work and is
an improvement on the old way (I think).

The test app now works without crashing the machine and the userland version
should also work once I commit my changes. This will be sometime soon but I
need to change the stuff as described above to get my network card back!

There is still a segfault that happens from time to time and needs to be
found and fixed, but otherwise the code seems to be working reasonable well.
Serial debugging has proved to slow things incredibly... with it the test
app was giving about 10-15 sockets / sec, without it gives up to 1143! Small
difference really.

While the test figures aren't really fair to compare just yet, net_server
gives about 400-450, BONE 1800 and our code about 1050-1100, so we're in the
ballpark, but then we haven't even done any optimising...

Anyway, that's not overly important just yet. Changing the locking may give
us a little bit extra, but we'll see.

Tomorrow will hopefully see all the code in CVS and this bug found and
fixed. Also, ipv4 and icmp modules need to be converted and some more work
done on getting the driver to talk to the stack.

On a related note, are people reading these and should I carry on writing
them? Are we happy with the direction things are going? Is anyone alive out
there?

david

Other related posts: