Re: Worlds greatest kernel

["Kyle" <kyleNOSPAM@xxxxxxxxxxxxxxxxxxxxxxxxxxxx>]

I can see interrupts being put into high-priority messages and maybe a
register type abstraction layer to enable bounds checking on registers.
Other problems relating to DMA, and other CPU circumvented functions would
be a challenge .. at least.
Kyle
"Nimrod Mesika" <nimrod-me@xxxxxxxxxxxx> wrote in message
news:20031008193507.GA779@xxxxxxxxxxxxxxxxxxxxxxx
> On Mon, Oct 06, 2003 at 10:57:56PM -0600, Kyle wrote:
> > In thinking about what would be the "best" kernel, the following come to
> > mind:
> ...
> > 5. Protected domain device drivers ( nothing crashes the core kernel
> > scheduler or loader unloader )
>
> That would be great *if* the hardware would be designed to support
> it.
>
> Right now even if you move drivers to userland (like QNX does, I
> believe), a buggy driver (or even buggy hardware) can take the
> system down.
>
> A badly formed bus cycle can kill the PCI bus and of course the
> hardware never heard of 'protected' memory...
>
> Maybe the high end server machines are designed this way. This
> usually violates the principle of pushing all complexity to the
> software layer.
>
> -- 
> Nimrod.
>
>