> > Well, kernelspace should not mean that we're no longer protected against
> > programming mistakes. It should just mean that we're priviledged to drop parts
> > of the protection in case we need to access the hardware.
>
> But we're not... Witness machines locking hard from a teardrop...
> Losing networking, and having to reboot cleanly, or reinsmod it or
> something would be vastly preferrable (at least to some) to just locking
> hard.
What if we'd make the kernel a bunch of ordinary processes which have maybe a
higher priority and access to the hardware, but nothing else? The only thing in the
kernel that needs to run out of process context is the actual task switching, irq
handling and the library functions that run in the caller's context.
This way, if something crashes, the corresponding task can be shut down, the memory
freed and the task reinitialized, so if IP defrag crashed, the only damage done
would be that the fragments in the wait queue would be lost... :-) The probably
most difficult task would be that the other tasks might need to react on such
things and that things should not be slowed down too much...
CU
Simon
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu