From owner-freebsd-hackers  Fri Aug 16 16:40:31 1996
Return-Path: owner-hackers
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.5/8.7.3) id QAA29255
          for hackers-outgoing; Fri, 16 Aug 1996 16:40:31 -0700 (PDT)
Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.2.228.19])
          by freefall.freebsd.org (8.7.5/8.7.3) with SMTP id QAA29243
          for <freebsd-hackers@freefall.freebsd.org>; Fri, 16 Aug 1996 16:40:21 -0700 (PDT)
Received: (from bde@localhost) by godzilla.zeta.org.au (8.6.12/8.6.9) id JAA07420; Sat, 17 Aug 1996 09:34:19 +1000
Date: Sat, 17 Aug 1996 09:34:19 +1000
From: Bruce Evans <bde@zeta.org.au>
Message-Id: <199608162334.JAA07420@godzilla.zeta.org.au>
To: freebsd-hackers@freefall.freebsd.org, ponds!rivers@dg-rtp.dg.com
Subject: Re: more general info on SIO problems in 2.1.5.
Sender: owner-hackers@FreeBSD.ORG
X-Loop: FreeBSD.org
Precedence: bulk

Originally-To: ponds!freefall.cdrom.com!freebsd-hackers@freefall.freebsd.org,
ponds!rivers@dg-rtp.dg.com 

>I can reproduce this easily with kermit - the particular status
>of it right now (in reproduction) is (output of /bin/ps -gaxl):

>  UID   PID  PPID CPU PRI NI   VSZ  RSS WCHAN  STAT  TT       TIME COMMAND
>    0  1647  1459  15   3  0   744  104 ttyin  I+    p0    0:00.64 kermit

Use pstat to debug tty hangs.

>kermit is connected to /dev/cuaa0 - whose tty settings are:

>ponds# stty -f /dev/cuaa0 -a
>speed 38400 baud; 0 rows; 0 columns;
>lflags: -icanon -isig -iexten -echo -echoe -echok -echoke -echonl
>        -echoctl -echoprt -altwerase -noflsh -tostop -flusho -pendin
>        -nokerninfo -extproc
>iflags: -istrip -icrnl -inlcr -igncr ixon ixoff -ixany -imaxbel ignbrk
>        -brkint -inpck ignpar -parmrk
>oflags: -opost -onlcr -oxtabs
>cflags: cread cs8 -parenb -parodd hupcl clocal -cstopb crtscts -dsrflow
>        -dtrflow -mdmbuf
>...

The flow control settings are unusual and bad.  Why use both crtscts and
ixon/ixoff flow control?  ixoff (aka TANDEM) flow control is broken as
designed and might cause hangs if the flow control characters are not
sent or received properly.  There are a couple of races in the driver
that might cause the characters to be dropped even when the line is
perfect.  ixon flow control (ordinary ^S/^Q flow control of output, which
is what is needed to honor the other side's using ixoff flow control)
works fine (provided the ^S/^Q are received), but is very inefficient.
On a 486/33, it costs 50% overhead per line at 115200 bps, limiting you
to at most one line at this speed.  On a P133, it costs 6% overhead at 
115200 bps.

> The behaviour I get is that when large amounts of data are generated 
>at the other end (after dialing a remote system) - I get several
>screen-fulls of output, then everything stops.. *However* from the
>modem lights, I see that the modem is still receiving data... it's just
>not getting to the kermit process to be written out.

This probably isn't directly caused by the flow control problems -
everything would probably stop if flow control gets stuck off.  It might
be caused by the extra overheads swamping the machine and hitting bugs
in the boundary cases, but the overheads are probably small enough not
to matter at a low speed like 38400.

Bruce