[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: ncr support?

Marcus Brinkmann <Marcus.Brinkmann@ruhr-uni-bochum.de> writes:

> On Mon, Jan 25, 1999 at 04:33:38PM +0900, OKUJI Yoshinori wrote:
> >
> >   In addition, it doesn't make sense to enable all scsi drivers, in
> > many cases. Because they often conflict when detecting scsi devices.
> > GNU Mach doesn't have loadable modules, so we have to prepare many
> > kernel images as Slackware does.
> Ouch!!! I hope this is not the last word on it. Isn't it the real benefit of
> a microkernel to be modular and extensible at run time? "many kernel images"
> is violent from my point of view.

As far as I understand, the entire notion of "device drivers" in
kernel space is more or less herecy when talking about micro kernels.
Ideally, all device drivers should be replaced with user-level tasks.

>From a practical point of view, I don't really know what's best. I
kind of like the principle that no user level program should ever be
able to crash the system or it's hardware seriously, which implies
severe restrictions on what a driver-like program can do (I really
don't like the linux-x86 way of having user programs like svgalib and
X write directly to the hardware). Perhaps having the kernel provide
reasonably safe but very primitive interfaces to the hardware (raw
scsi commands, for instance), and doing the rest by user level
programs would be a workable compromise? But then, there are the
efficiency issues.

For now, I guess that linux-compatibility, in order to borrow support
for common PC hardware, is more important than a theoretically
beatiful design.


Reply to: