Web to the XFree86 home page.
sudo apt-get install nvidia-glx-legacy
sudo gunzip /usr/share/doc/nvidia-glx/legacy/README.gz
sudo gunzip /usr/share/doc/nvidia-glx/legacy/nvidia-settings-user-guide.gz
sudo gunzip /usr/share/doc/nvidia-glx/legacy/examples/XF86Cenfig.sample.gz
sh NVIDIA-Linux-x86-1.0-7667-pkg1.run
The installer couldn't find an appropriate kernel module, and couldn't find one at the nVidia site, so it said it would build one.
Then it said that my kernel was compiled to include rivafb support. Rivafb is incompatible with the nVidia driver. It said it would build the driver, but things will break if the rivafb module is ever loaded.
The installer built the driver successfully, but there were some messages about "module license 'NVIDIA' taints kernel".
I referred heavily to the documentation for the nvidia driver. Under Ubuntu, a little bit of information can be found in /usr/share/doc/nvidia-glx-legacy/README.Debian. The majority of information is in the NVIDIA Accelerated Linux Driver Set README and Installation Guide which is found at the Nvidia website under the README link.
By default, xorg loads the "nv" driver. To support dual monitors in hardware, you need to load the "nvidia" driver. In xorg.conf, in the "Device" section, set "Driver" to "nvidia".
Nvidia calls using two monitors "TwinView" mode. To enable it, you have to specify all of the following lines:
Option "TwinView" Option "MetaModes"
(as root)
cd /usr/src
mkdir XFree86-4.2.0
cd XFree86-4.2.0
ftp gate.ucar.edu
anonymous@ftp.xfree86.org
siemsen@ucar.edu
cd pub/XFree86/4.2.0/binaries/Linux-ix86-glibc22
get README
get Install
(from reading the Install file, the list of other files follows)
get Xinstall.sh
get extract
get Xbin.tgz
get Xlib.tgz
get Xman.tgz
get Xdoc.tgz
get Xfnts.tgz
get Xfenc.tgz
get Xetc.tgz
get Xvar.tgz
get Xxserv.tgz
get Xmod.tgz
exit
chmod +x Xinstall.sh
./Xinstall.sh
says to back up current X, so...
^C
tar cf /tmp/usr.x11R6.tar /usr/X11R6
tar cf /tmp/etc.X11.tar /etc/X11
./Xinstall.sh
To get a decent display, you need to know what video card you have, and what monitor you have. My home system's video card is a "RAGE TURBO 8mb". To find out the chipset in the video card, run the SuperProbe program. On my home system, SuperProbe says the video card has the Mach64 chipset
alias ss='startx -- -bpp 24'
alias ss='startx -- -depth 24 -auth $HOME/.Xauthority'
Use xvidtune. Play with it to fine-tune the display placement. Use the Show button to cause xvidtune to write the current settings to sysout, in a format suitable for cut/pasting into /etc/X11/XF86Config.
After much manual reading and testing, I found that I can get 1600x1200 from the lower color depths, like 16 and below. For 24, I can get up to 1280x1024. For 32, I can get up to 1024x768.
http://support.dell.com/docs/systems/pmojav/specs.htm.
The external video port on the back of the laptop is used in at
least two scenarios: when docked and when connected to the
InFocus projector. In either case, the external video port has
to be driven. When docked, the external video port has to be
driven, but the laptop's LCD screen does not. When using the
projector, the LCD screen and the external video port have to be
driven at the same time. In either case, to drive the external
video port, you have to modify the
/etc/X11/XF86Config file to make the X
server drive the internal and video correctly.
In the file, look for
Section "Device"
Identifier "NeoMagic (laptop/notebook)"
Uncomment the lines that say
Option "intern_disp"
Option "extern_disp"
This will make using the projector or the docking station monitor possible.
If the laptop isn't configured for the correct environment, stupid problems occur. For instance, ff the wrong mouse type is selected, the external mouse won't work at all. If 3-button mouse emulation is set when it shouldn't be, HP OpenView won't handle right-button clicks correctly. If the screen resolution is wrong, windows will pop-up in the wrong places.
To fix this, when I boot I tell the laptop the environment I'm in, and it does the right thing. The environment is represented by a bash variable named SCHEME. When grub runs, it offers me a selection screen and starts the Linux kernel with the appropriate "SCHEME" variable set. Later, I source a bash script named /usr/local/bin/getscheme.sh whenever I need to know the environment that I'm in. Here's the contents of the script:
#!/bin/bash
#
# Get the "scheme", which is a value describing the machine's
# environment. It reflects the choice made when the machine was
# booted. The value is set in the /etc/grub.conf file, which sets the
# SCHEME kernel variable. This is available in the /proc/cmdline file
# and can be used by the pcmcia module initializes, though as of
# 2003-06-06 I don't make use of it. This function reads the
# /proc/cmdline file and sets a bash variable, and complains if
# anything unexpected happens. IT DOES NOT SET AN ENVIRONMENT
# VARIABLE.
function getscheme() {
local SCHEMEFILE=/proc/cmdline
if [[ -r ${SCHEMEFILE} ]]; then
SCHEME=`perl -e 'open F, "</proc/cmdline"; $_ = <F>; close F; /SCHEME=([a-z]+)/; print "$1\n";'`
if [[ ${SCHEME} != 'home' && \
${SCHEME} != 'work' && \
${SCHEME} != 'ucarnet' && \
${SCHEME} != 'foreignnet' && \
${SCHEME} != 'standalone' && \
${SCHEME} != 'ucarnetv' && \
${SCHEME} != 'foreignnetv' && \
${SCHEME} != 'standalonev' ]]; then
echo "warning: SCHEME in ${SCHEMEFILE} not recognized"
fi
else
echo "warning: ${SCHEMEFILE} not readable, setting SCHEME to \"unknown\""
SCHEME='unknown'
fi
}
By default, when a Red Hat system boots, init
reads /etc/inittab which causes init to run
/etc/X11/prefdm which runs gdm
which reads /etc/X11/gdm/gdm.conf and then
prompts for a username/password and then runs the XFree86
server, which reads /etc/X11/XF86Config.
In this sequence, there's no way to feed a command-line option to
XFree86.
To set up the screen and mouse correctly based on the environment, I need to give XFree86 a "layout" command-line option so it can select the right section of /etc/X11/XF86Config. There's no way to do this using gdm, so I disabled startup of gdm by commenting-out the prefdm line in /etc/inittab. Now, when the system boots, the X server isn't automatically started. Instead, I get a boring old text username/password prompt. I log in and start the X server manually.
To start the X server, I run aaa.sh, a script I wrote that runs startx, which starts XFree86 with a layout command-line argument. The value of of the argument comes from the SCHEME that I started the kernel with. Here's the contents of aaa.sh:
#!/bin/bash
. /usr/local/bin/getscheme.sh
getscheme
if [[ ${SCHEME} = 'home' ]]; then
startx -- -layout Home
elif [[ ${SCHEME} = 'work' ]]; then
startx -- -layout Work
elif [[ ${SCHEME} = 'ucarnet' ]]; then
startx -- -layout Ucarnet
elif [[ ${SCHEME} = 'foreignnet' ]]; then
startx -- -layout Foreignnet
elif [[ ${SCHEME} = 'standalone' ]]; then
startx -- -layout Standalone
elif [[ ${SCHEME} = 'ucarnetv' ]]; then
startx -- -layout UcarnetV
elif [[ ${SCHEME} = 'foreignnetv' ]]; then
startx -- -layout ForeignnetV
elif [[ ${SCHEME} = 'standalonev' ]]; then
startx -- -layout StandaloneV
else
echo "unknown SCHEME in aaa.sh"
fi
This works well, though it requires the extra step of running
aaa.sh every time I boot.
To allow conditional setting of resources in the
For some reason, in my home directory,
.Xresources file, add
"-D`hostname`" to the xrdb command to define the name of the
host.
The .xinitrc startup script runs
when X is started. It runs xrdb and
feeds it the .Xdefaults file.
.Xresources is a link to
.Xdefaults. The
.xinitrc script simply runs
xrdb on both files if they exist,
so it runs xrdb twice on the same file. To prevent this,
comment-out the line in .xinitrc that runs xrdb on
.Xdefaults.
Warning: Cannot convert string "MetaCtrl<Key>Insert" to type VirtualBinding
xprop -root -remove _MOTIF_DEFAULT_BINDINGS
X11 connection rejected because of wrong authentication.
error in locking authority file
/etc/X11/XF86Config, I added
Emulate3Buttons
# Emulate3Timeout 50
/etc/X11/XF86Config, one for the
touchpad and one for Logitech mice at home and work. I use
"-pointer" in ~/aaa.sh to
conditionally select the mouse based on the boot SCHEME set when
the system was booted (see
my grub page). Note that to
make HP Openview recognize the right mouse button, I had to remove
the line that said "Option "ZAxisMapping" "4 5"" from the
"Logitech" section.
~/.xinitrc file. On the home
system, the standard mouse acceleration is annoying to DeLynn. To
fix it, do
xset m 6
I checked that all font diregtories listed in the /usr/X11R6/lib/X11/fs/config actually exist. /usr/X11R6/lib/X11/fonts/cyrillic didn't, so I deleted it.
I added a font directory to the path as described in the Abiword page.
cd /usr/src
mkdir XFree86-4.1.0
cd XFree86-4.1.0
ftp gate.ucar.edu
anonymous@ftp.cs.umn.edu
siemsen@ucar.edu
cd pub/XFree86/4.1.0
get Xinstall.sh
^Z
bg
sh Xinstall.sh -check
fg
cd binaries/Linux-ix86-glibc21
get Install
^Z
bg
read the Install file
fg
binary
get Xinstall.sh
get extract
get Xbin.tgz
get Xlib.tgz
get Xman.tgz
get Xdoc.tgz
get Xfnts.tgz
get Xfenc.tgz
get Xetc.tgz
get Xvar.tgz
get Xxserv.tgz
get Xmod.tgz
get Xflat2.tgz
get Xhtml.tgz
quit
mv ../Install .
cd /usr
tar cf /usr/X11R6-3.3.6.tar X11R6
gzip X11R6-3.3.6.tar
log out, exit X, and log in as root
cd /usr/src/XFree86-4.1.0
sh Xinstall.sh
This'll install local HTML docs in
/usr/X11R6/lib/X11/doc/html/index.html,
so put a link to them into the home web page.
For the new X server to work, you have to have a new
XF86Config file, with the new
syntax. To get one, I did this as root:
XFree86 -configure
mv /root/XF86Config.new /etc/XF86Config
This allowed the new server to run, but with poor color support.
To get good color support, edit the
.bashrc file and define
alias ss='startx -- -depth 24 -auth $HOME/.Xauthority'
This fixed the color problem, and set up xauth-style security so that exmh will run (see exmh.html for details of this)..
After you start it up, the /var/log/XFree86.0.log file gives a detailed listing of what happened during startup. This can be studied for ways to tune the /etc/XF86Config file.