> would be good to try to predict how long it will take to design the board
> and its operating system, and to try to predict what the cost of the comp
> will be at that time, and design on the basis of that. I realize that th
> approach is full of risks, and would be difficult to do, but I am afraid
> if this approach is not taken, the board my be obsolete by the time it is
Neil made the assumption that it will take a bit more than half a man-year
to port the kernel and drivers. This depends on how many filesystems we wil
support. For me, ext2fs and the ISO fs would be enough.
It has little to do with how many filesystems we support, as the filesystems
should be devoid of any system dependant code. Endianess is pretty easy to
deal with where we need binary compatability with existing filesystems (ie
MSDOS since you may want to read DOS floppies).
As far as how long device drivers take :
IDE, serial, parallel, etc. where we've got the same parts as in the PC -
we simply add appropriate #ifdef LINUX_MIPS or whatever to the kernel sources
around code that does memory mapped I/O instead of using the port commands.
SCSI - it looks like the SCSI driver I'm writing right now (NCR5380)
will have taken about 20 man-hours (including me reading the NCR5380
databook, designing the driver architecture, writing the code, etc) by
the time it goes into ALPHA test. The 5390's are even easier to program.
Ie : device drivers are not a significant problem.
I think the hard parts will be taking care of the major 386isms, ie the
scheduler and VM which is really prone to race conditions. Since people have
access to other MIPS architecture machines (ie, Decstations) and sample MIPS
code to work with (BSD4.4 runs on the Decstations) this can happen in parallel
with the hardware development.
Everything else *should* compile out of the box, except the X11-server stuf
That sounds not too bad, isn't it ?
I don't think there will be any significant technical problems getting
the software up.
Speaking of kernel development for the MIPS : we need to start a discussion
on how to best make the kernel sources multi-architecture, so that one
release of Linux runs on MIPS, Intel, and Motorola platforms.
The AMIGA people didn't bother to do this, so they need a different
kernel distribution and can never take advantage of new features
when they're added.