Re: [IPCop-devel] Update on the Cobalt not booting situation.

[David W S. <avi...@ai...>]

Olaf Westrik wrote:
> On 2010-09-06 03:21, David W Studeman wrote:
>
>
>>> I have a diff uploaded at least now. Basically 3000 series (Raq3/4 and
>>> Qube3) will boot the kernel fine but crashes when userspace (init) crashes.
>>> The Raq550 which is Pentium III boots and runs fine.
>   >
>> I do have a 512mb image of 4889 uploaded now and as I already mentioned,
>> the diff file.
>
>
> A few hours ago I pulled your 4995 diff and started building a RaqCop.
> Apart from prefetch failing to load the 2 .tar.gz files from raqcop.com
> (some http 406 error) and myself struggling with patch, all went smoothly.
>
> Before fixing^w attempting to narrow down the booting problem, I took a
> closer look at your modifications and thought I'd write something down
> for further discussion:
>
>    - logoutput to /dev/tty12, no point to do that as there is no tty12 on
> a headless box. But this would apply to other headless boxes as well,
> would it not? Question: do we need/want logging on tty12 to start with,
> or can we drop that completely? Make some provisions in rc.sysinit to
> add/remove it?

This should apply to any headless device. I directed it to null for the 
time being. One could actually send it to a second serial terminal if 
one wanted to see it.

>    - update script for PCI/USB IDs, useful enough to have available for
> everybody?

Definitely but of course I am biased. This became important to me 
especially with Raqcop 1.4 where I integrated the sysinfo addon but 
really merged it with the OS and had everything built from lfs including 
usbutils but old enough versions that the included ids were very 
outdated. A long time ago you and I touched on the possibility of being 
able to invoke it on the updates page of the gui. On my daily Raqcop 
1.4.21 I have cron invoke it.
>
>    - LCD 'support', surely we can't make that fully universal (your LCD
> is smaller/bigger then mine ;-)), but we could make for some standard
> provisions to have at least some minimalistic output from our boot scripts.

This originally came from Bob Fisch of the French IPCop forum. It 
originally was designed for a 20x2 parallel connected HD44780 compatible 
display. I had to hack a few functions to make them friendlier to the 
16x2 display of the Cobalts. Also one of my forum members added the 
ability to invoke text where empty spaces can be used but preserved the 
old lcd -text output. What you see in rc.sysinit is using the new better 
output. This should be usable in many displays if not most.
>
>
> Then there are some rather specific RaQ patches, for which there are 3
> options:
> 1) keep as is
> 2) create a branch in SVN (or revive the already existing 'build' branch)
> 3) attempt a full merge and something like cobalt setting in our .config
> file.
>
> Reasoning behind 2/3 to make interaction slightly easier, surely the RaQ
> might seem exotic, but I do have a personal interest (:-)) and I also
> think that making things work on a RaQ (same as for ALIX boards) is
> benificial for the IPCop project. Not because of the millions of
> additional users, but simply because the RaQ/ALIX are not that far away
> from your typical IPCop box.
>
>
> Opinions, thoughts, anyone?
>
>
> Olaf

Many of the Cobalt mods should be used on any headless device such as 
the inittab, securetty, and commenting out everything in rc.sysinit that 
looks for any of the tty's. It's also beneficial to have a specific 
kernel for headless devices that has the VT's and keyboard as well as 
mouse support deselected. You'll get a much cleaner boot output and 
won't have to field a ton of questions about the annoying yet harmless 
errors in the boot output and dmesg.

The partition and installer sources would be the same on everything 
except the Cobalts. Notice that I save networking for first boot since 
without a NFS installer you use a donor machine or VMWare to install it 
and you need to rerun setup anyway for networking. I also do not include 
anything related to extlinux or the standard pc kernel but I do use the 
standard installer kernel since as of yet, we use a standard PC or 
emulator to get it installed to a drive. A nullmodem cable IS required 
to setup networking the first time.
-- 
Dave Studeman
http:/www.raqcop.com