Bug#255330: Install report: MVME162
Package: installation-reports
INSTALL REPORT
Debian-installer-version: self-built from svn head
uname -a: Linux ska.grep.be 2.2.25-mvme16x #1 Thu Mar 18 05:44:03 CET 2004 m68k GNU/Linux
Date: Sat, 19 Jun 2004 in the evening 'till around Sunday 12:00 PM.
Method: mvme16x_netboot
Machine: Motorola MVME162
Processor: Motorola 68040, 33Mhz
Memory: 64MB
Root Device: /dev/sda
Root Size/partition table:
(parted) print
Disk geometry for /dev/sda: 0.000-2044.453 megabytes
Disk label type: msdos
Minor Start End Type Filesystem Flags
1 0.030 1922.121 primary ext2 boot
2 1922.121 2044.001 extended
5 1922.151 2044.001 logical linux-swap
Output of lspci and lspci -n: N/A; this system doesn't have PCI.
Base System Installation Checklist:
[O] = OK, [E] = Error (please elaborate below), [ ] = didn't try it
Initial boot worked: [O]
Configure network HW: [O]
Config network: [E]
Detect CD: [ ]
Load installer modules: [O]
Detect hard drives: [O]
Partition hard drives: [O]
Create file systems: [O]
Mount partitions: [O]
Install base system: [O]
Install boot loader: [E]
Reboot: [O]
Comments/Problems:
The installation was successful, but there's still a lot of work to do.
Some notes I wrote down as I was performing the installation:
* Choosing Dutch as installer language will hang the installer at the point
where a hostname has to be chosen (hence the "E" at network configuration,
although it isn't the network configuration udeb at fault). The reason is
probably some non-ASCII characters being sent to the console, which modifies
terminal setup in ways it shouldn't.
As a possible way to resolve this issue, I suggested on IRC to disable
langchooser when we're working on a serial console. Both Christian Perrier and
Martin Michlmayr seemed to welcome this suggestion.
* I got a "mount: /dev/pts: no such file or directory" or similar message before
the initial appearance of the main menu. Not sure whether or not it's
harmless, but it should be looked into.
* vmelilo-installer wasn't available in the main menu. As a result, I couldn't
install a boot loader. Luckily, I had a working GNU/Linux installation on this
system and was able to boot the freshly-installed system by using a "root="
argument to the installed vmelilo, but that will, of course, not work for
everyone...
* Most VME systems, including mine, only have a serial console to work with. As
a result, I got these messages all through the base-config:
INIT: id "2" respawning too fast: disabled for 5 minutes
INIT: id "3" respawning too fast: disabled for 5 minutes
Moreover, after base-config was completed (and the "full" /etc/inittab was
installed), I didn't have *anything* anymore; the installed inittab assumes
tty1 through tty6 are available. I had to ssh in and change the inittab to be
able to log in at the console.
Perhaps base-config should configure an inittab that has a getty on a serial
port if we're configuring through a serial console.
* baseconfig-udeb didn't work. Rebooting the system and configuring it that way
was required.
* "Configure timezone" main-menu option in base-config didn't work.
* "Configure the keyboard" shouldn't be there on a serial console.
* "Select packages to install" sits there for at least 5 to 10 minutes with no
visual confirmation. This is probably to be expected on an m68k system, but it
would be nice if some sort of message would appear in the style of "doing foo,
this might take a while...". No big deal if it doesn't, though.
* tasksel's terminal setup is suboptimal on serial consoles; tasksel selects a
monochromatic colormap for some reason (even though cu does support colored
output, and the main menu does indeed show colors) with no difference in color
between a selected and a not-selected button, and, in fact, no way to see that
a button is selected. Only by accident did I discover that the <tab> key does,
indeed, work...
In short, the support for serial consoles isn't well tested and should be
improved a lot. At the moment, some udebs test whether we're working on a serial
console and handle it appropriately, but it's all done inconsistently, and the
physical test whether we're running on serial consoles is done a few times.
I think this shouldn't be necessary. As a suggestion, I propose the following:
* Add a "serial-support" udeb, which will detect whether we're running on a
serial console, and set the "debian-installer/serial-console" debconf value
(or whatever it was that some udeb sets) to true if it is. This should, of
course, be done before _anything_ that needs to know whether we're running on
a serial console is ran, i.e., pretty much near the beginning of the
installation process. This package should also disable langchooser the same
way lowmem-udeb does.
* userdevfs should be fixed to remove /dev/vc/[2-4] if we're running on serial
console. Right now, I've slightly hacked the Makefile here to not create those
device nodes, but they should of course be there if they are available.
Not removing the files if they're not available is not an option; busybox init
gives an error message every time it tries to start something which it can't,
and does /not/ disable anything if it respawns too fast. You can make the
logic.
* Things that want to do stuff with the console (keyboard configuration,
writing /etc/inittab, ...) should check whether
debian-installer/serial-console is set, and take appropriate action when we're
running serial.
* Many many more people testing on serial consoles :-)
Comments?
Reply to: