LinuxSA Mailing list archives

  From: Mark Newton <newton@atdot.dotat.org>
  To  : Rick Marshall <rjm@herzfeld.com.au>
  Date: Fri, 12 Mar 1999 14:45:10 +1030 (CST)

Re: Big disk drive + small PC + linux

Rick Marshall wrote:

 > Can't speak for Linux because I haven't tried it, but in the old days SCO,
 > Interactive, and Solaris could boot any size disk irrespective of the
 > BIOS. This is because Unix and (I presume) Linux store the disk config in
 > the boot sector. All you should need is any BIOS def to get going. We used
 > to use type 1 and let Unix take over from there.

Linux is rather different from most other UNIXes in the way it boots.

Here's the lowdown:  If you attempt to use the BIOS, you cannot reasonably
expect to access any part of the disk past cylinder number 1023.  This
is why disks go through that stupid translation stuff to lie about their
geometry:  can you really imagine a disk that's 15mm high having 32
platters (64 heads) in it?  No?  So why do so many systems report so 
many disks as having 64 heads?  Because if you articifically inflate the
number of heads you can articifically compensate by lowering the number
of cylinders you pretend to have. 

So the fireware on the disk works it all out to compensate for the 
short-sighted limiting design developed by the original BIOS writers
at IBM.

You end up with a problem, however, when you have a disk that's so large
that even with geometry remapping you can't fit the disk into the BIOS's
constraints.

On a Linux system you must split your disk up into multiple partitions
so that your boot partition is less than 1024 cylinders in size.  I think
Windows is also limited accordingly.  Most other UNIXes, however, aren't.

Most UNIXes use a multiple stage boot

  1st stage:  Boot sector at the start of your disk
  2nd stage:  Larger program which is limited in size like the 1st stage
              usually is, which can contain "real" device drivers for 
              disks and knowledge of UNIX filesystems
  kernel   :  you get the idea.

The first stage boot doesn't know how to load kernels;  it just knows
about the second stage bootloader, which is usually kept somewhere in
a reserved portion of the disk (which, on PC UNIXes, is always inside
the first 1024 cylinders).  The first stage's job is to load the second
stage and execute it.

The second stage bootloader can put the CPU into protected mode (which
removes the stupid 640k or 1 Mbyte limit on memory accessing at the 
cost of rendering BIOS device drivers unusable) and use a built-in
(non-BIOS) disk device driver and built-in knowledge of the structure
of a UNIX filesystem to load a kernel, which is usually called /unix
or /stand/unix or something like that.  This is the system employed
by SCO, Solaris, UnixSwear, et al.

Linux takes an alternative approach: It uses the BIOS for I/O, which
gives you your 1024-cylinder limit.  It also means the secondary
bootstrap (LILO) can't run in protected mode, which gives you the
kernel size limitation (ever built kernels that are too large to
boot?).  The advantage that this gives is that LILO can be substantially
simpler to implement.

FreeBSD has recently taken yet another approach.  It uses the two-stage
boot system described above to load a third-stage bootstrap, called
/boot/loader. /boot/loader implements a shell-like interface to system
configuration and a forth interpreter which can be used for customization.
Based on instructions in a config file, /boot/loader takes care of
loading the kernel and any boot-time modules.  /boot/loader runs entirely
in protected mode and doesn't care how big your disk is.

Linux could benefit from that third approach: Rather than using LILO
to boot /vmLinuz or whatever, it should be possible to write a tertiary
bootstrap and have LILO boot that instead.  This would have the benefit
of introducing significant extra functionality, usability and 
configurability without having any impact whatsoever on the existing
boot-time infrastructure employed by Linux.

Does anyone want to take on a holiday project? :-)

In the short term, you need to make sure that the disk you're booting
from features a boot partition which falls entirely within the first
1024 cylinders.

    - mark

--------------------------------------------------------------------
I tried an internal modem,                    newton@atdot.dotat.org
     but it hurt when I walked.                          Mark Newton
----- Voice: +61-4-1958-3414 ------------- Fax: +61-8-83034403 -----

-- 
Check out the LinuxSA web pages at http://www.linuxsa.org.au/
To unsubscribe from the LinuxSA list:
  mail linuxsa-request@linuxsa.org.au with "unsubscribe" as the subject