Also note that locking down cache lines is almost never a win anyway,
unless you have some _very_ specific uses for the machine.
For example, even if we locked down a very commonly used line in the
kernel, what would happen to all those applications that are running
almost totally in user mode? To them, the cache would appear smaller
than it would otherwise be.
Locking down the cache line is usually worth it only if:
- you have a very specialized application or use for the CPU, and you
_know_ that something is always so important that it really is worth
it.
or
- you have real-time constraints that you cannot guarantee any other
way. You know that locking down the cache may be bad for overall
performance, but at least you'll get _repeatable_ performance or some
particular routine has such strict latency requirements that you
don't care that you're slowing down everything else.
So I applaud the larger cache in the Cyrix CPU, but I suspect it is
better used as a regular cache rather than anything else. It's not as
if a 64kB L1 is excessive these days - there are rather few applications
that wouldn't be happier with more. Often quite a lot more.
Linus
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu