> Why does the end-user have to compile the kernel? Most distributions
> provide a kernel with no SCSI drivers in it, but use an initrd to get
> the root SCSI driver in (man mkinitrd on any Redhat box). Just
> distribute all SCSI drivers as modules and you won't have any problems.
>
That is not totally true. There are two problems here, one is where you have
different controllers in your system and the other is where you have multiples
of the same controller. What you list above solves the different controller
problem. By loading the drivers in the right order you will get predictable
results. However when having multiples of the same controller you are only
loading one driver so you are at the mercy of the way that driver was
developed. Some drivers give you ways to work around this others do not.
For example the aic7xxx.c (current one at least - I have not played with the
Beta one enough to know what it does) lets you play with the order by turning
BIOS off on the cards that you don't want to BOOT from. So the aic7xxx driver
sorts the controllers with BIOS enabled first. This solves the problem where
you have multiple adaptec controllers in the same box to make sure you have
the "boot" controller first. This, however, does not solve a third problem
where you have multiple disks on that controller. My recommendation is that
you always install on ID 0 since that will be the "first" one found. If you
install on ID 1 and you add ID 0 then you just broke your boot. If you
install on ID 1 where there was an ID 0 (so you install to sdb) then if ID 0
dies, get pulled, etc then you can boot because ID 1 is now ID 0.
So though I do agree that making all drivers modules usually simplifies
handling this there are still issues and solving these I do agree today is
beyond the scope for the unexperienced.
Eddie
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [EMAIL PROTECTED]
Please read the FAQ at http://www.tux.org/lkml/
