Preemptable kernel (was Scheduling latencies news: less RAM = less latency)

Keith Owens (kaos@ocs.com.au)
Mon, 02 Aug 1999 08:49:01 +1000

On Mon, 2 Aug 1999 02:54:04 +1000,
Ralf Baechle <ralf@uni-koblenz.de> wrote:
>On Sun, Aug 01, 1999 at 12:23:43AM +0200, Benno Senoner wrote:
>
>> I think making the kernel fully preemptable is not an easy task and will
>> not happen very soon.
>
>SGI has made IRIX fully preemptable some years ago. By now some of their
>engineers believe that this was an overkill solution for their problems
>and not worth the resulting complexity.

MVS is preemptable, in addition 90% of the MVS "kernel" is pageable.
It allows both by the heavy use of spin and sleep locks, avoiding
deadlocks at the expense of a specific hierarchy of locks. For
example, to do anything to virtual storage tables, you must own the
VSM lock and to obtain the VSM lock, you are not allowed to already own
any locks lower in the hierarchy than VSM. Pretty standard deadlock
avoidance algorithm.

So it is possible to have a real and complex OS that is preemptable and
pageable. Although I have to say that the MVS internals to support
kernel preemption and paging are a lot more complex than the current
Linux internals.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/