I'm still hoping that someone will respond to this.  I can see from
my ftp logs that people have downloaded the source, ...but no responses.
I presume that this is because they are all laughing too hard. ;->

Is this a conspiracy to force me to fix serial on MacPPC, so that it
becomes important? ;->

-dgl-

--- begin forwarded text


Mime-Version: 1.0
Date: Fri, 25 Jan 2002 19:44:35 -0600
To: port-macppc@netbsd.org
From: Donald Lee <donlee_ppc@icompute.com>
Subject: "soft" interrupts busted on MacPPC
Sender: port-macppc-owner@netbsd.org

About a year ago, I put into production a server running NetBSD 1.5 that
has an 8 port Cyclades serial PCI card in it.  I use the serial ports
to support dial in PPP.  I have since upgraded to 1.5.2.

I have a special kernel that I have to run to make this work.  It has several
changes.  One is to the extintr.c module, and it fixes what is
pretty clearly a bug in the "soft" interrupt handling in the kernel.
Without this fix, PPP works, but is intolerably slow, because the
protocol stack effectively "loses" interrupts, and they don't get
"noticed" until another "event" comes along.

This has not been noticed in the broader community because there is almost
no way to get a working serial port on MacPPC. :-(  However, the change
effectively removes an issue that causes inexplicable delays in network
response.  Although I have not done extensive testing, I'll bet that you
could show performance improvements on ethernet with the patch too.

I hear that NetBSD 1.5.3 is in ALPHA.  What do I have to do to get the
extintr.c changes incorporated into the kernel so I don't have to carry
the patch around?  This issue fixed in 1.6, but not in the same way because
of the SMP changes in that area of the code.

The changes can be seen on my anonymous FTP server at

	ftp://ftp.icompute.com/pub/donlee/extintr.c.nogo
and
	ftp://ftp.icompute.com/pub/donlee/extintr.c.prod

(as in... production)

Thanks in advance,

-dgl-

--- end forwarded text