On 28 Nov, Todd Whitesel wrote:
I like the idea of having the BIOS know how to load
the kernel itself.
Most filesystems can be read from with a fairly small amount of code.
Let people compile into their BIOS however much code they want depending
on which filesystems they actually plan to use.
I don't like the idea of
putting OS-specific knowledge in the firmware,
I think the firmware should load a file from the disk to a well-known
place in memory. This file could be the OS kernel if the kernel can cope
with this environment directly or it could be a secondary loader with
some OS specific knowledge that loads and perhaps relocated the kernel
in some arcane way.
If we can allow the kernel and BIOS to use much of the
same source code
for drivers, then there is no need to ever call into the BIOS once the
kernel is loaded. This eliminates stupid problems caused by the BIOS
wanting parts of the system to be one way and the kernel wanting them
to be another.
I don't think that the OS should have to call the FW after boot
but during the boot, however...
The main reason I do not like OpenFirmware is that it
presumes a slow,
frozen driver in ROM with its own frozen idea of how interrupts and DMAs
should work will be good enough that my kernel will never want to replace
it. Yeah, right.
Well, that was a property to OF that I was unaware of. What I like
OF is the Forth code, the device tree and that is is a standard.