*BSD News Article 17943


Return to BSD News archive

Xref: sserve comp.os.386bsd.bugs:1046 comp.windows.x.i386unix:2318
Newsgroups: comp.os.386bsd.bugs,comp.windows.x.i386unix
Path: sserve!newshost.anu.edu.au!munnari.oz.au!network.ucsd.edu!swrinde!cs.utexas.edu!wupost!gumby!destroyer!cs.ubc.ca!unixg.ubc.ca!acs.ucalgary.ca!cpsc.ucalgary.ca!xenlink!fsa.ca!deraadt
From: deraadt@fsa.ca (Theo de Raadt)
Subject: Re: XFree86 1.3 crashes under 386BSD
In-Reply-To: veit@mururoa.gmd.de's message of Fri, 2 Jul 1993 12: 02:07 GMT
Message-ID: <DERAADT.93Jul3181840@newt.fsa.ca>
Sender: news@fsa.ca
Nntp-Posting-Host: newt.fsa.ca
Organization: little lizard city
References: <20ea6n$5v5@email.tuwien.ac.at> <2C33DD07.3337@news.service.uci.edu>
	<2114d5$lci@aggedor.rmit.OZ.AU> <1993Jul2.120207.3262@gmd.de>
Date: Sun, 4 Jul 1993 01:18:40 GMT
Lines: 40

In article <1993Jul2.120207.3262@gmd.de> veit@mururoa.gmd.de (Holger Veit) writes:
> In article <2114d5$lci@aggedor.rmit.OZ.AU>, rxxram@minyos.xx.rmit.OZ.AU (Richard A. Muirden) writes:
> |> This is interesting - I don't have those kinds of problems with XFree86 1.3 -
> |> what I did have the other day is a xdm crash of sorts. basically the mouse
> |> worked fine and shot all over and I could even highlight the text in
> |> the xconsole window but the keyboard wouldn't input at all and that made it
> |> kinda hard to login from the console :)
> |> 
> |> I didn't try a telnet in so I don't know if the whole system had died or
> |> what - I just got very impaitent and rebooted :)
> |> 
> |> -richard

> You are running 386bsd with pccons, and the effect was that over night your
> keyboard was killed? This is a known effect, happening with all console drivers
> which abuse /dev/vga as a second entry to /dev/console (i.e. have /dev/vga in
> /etc/ttys). In this case syslogd closes and reopens /dev/console at midnight, and
> thus grabs the keyboard away from the xserver. There are two possible fixes for that,
> one that does counting the open/closes on /dev/console and "/dev/vga" and does the
> actual close when both counters are zero (a slightly different alternative uses
> the reference count for a vnode for this), or give the server an own device for
> raw keyboard access rather than /dev/console. This latter has been done in codrv.

The two possible ways to fix the problem, mentioned above, are
required because of some early patchkit patches. Chris Demetriou and I
looked into this one night, we believe the original patches are
incorrect. The standard NET2 code for /dev/console does have some
problems, but if it's clear from a look at some hp300 tty device
drivers that the patches are wrong. It is clear the i386 device
drivers are written wrong (ie. any device driver that might become a
console device driver.) I think we'd like to solve the problem
correctly before we supply some half-hacked solution that appears to
work.

BSDI may still have some bugs in this area too.

Meanwhile, both of the hacks described above are ugly.
 <tdr.
--
This space not left unintentionally unblank.		deraadt@fsa.ca