Re: RT cache management

Alex P. Madarasz, Jr. (madarasz@erols.com)
Mon, 22 Jun 1998 07:27:47 -0400


At 6/22/98 04:18 AM , Rik van Riel wrote:
> On Sun, 21 Jun 1998, Jeff Millar wrote:
> > Recalling the questions: What cache control features do RTOS's have that
> > Linux doesn't? Does Linux have another approach to achieve the
benefits of
> > cache control? Do applications really need cache control?
>
> Linux has some cache controlling features, but they're not
> application program tunable. All Linux machines are
> automatically cache coherent, so programs don't really need
> that kind of control anyway...
>
> If memory is shared, you probably want to pay the price
> for cache coherency, if memory is not shared, you're not
> paying the price anyway...

I believe the original poster was discussing multi-processor "shared
memory" (also called "reflected" memory) hardware systems, where the shared
memory (usually on a peripheral card for PCs) is multi-ported via a
high-speed serial/parallel bus and has a "back-door" port from one system
to another which is not controlled / controllable by the mobo chipset.
See, for example, Systran's SCRAMnet series at :

<http://www.systran.com/scramnet.htm>

In such a system, the mobo chipset memory controller isn't integrated with
the shared memory, and a remote system can write to a local shared memory
location without the local system's chipset / CPU being notified,
invalidating the local / CPU cache without notifying the local system of
the fact. In RTOSs which allow for such a setup, you can tell the OS that
a block of memory is to be locked down (not pageable) and not cached, so
that the CPU / chipset is always forced to fetch data directly from the
shared memory rather than from cache.

---------------------------------------------
Alex P. Madarasz, Jr. -- madarasz@erols.com

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu