From: "Rik van Riel" <riel@conectiva.com.br>
> >
> > |who has added the PG_AGE_ stuff to filemap.c? It's behaving quite badly
on
> > |SMP - a simple stream of reads cause 20-30% kswapd activity, with the
> > |majority of the profiling hits getting the
'PageTestandClearRreferenced()'
> > |line.
>
> This was the case *before* ac10. The patchlet that went into
> ac10 always *resets* the referenced bit before adding a page
> to the lru list...
>
Is that bit often set, or usually zero?
If the bit is usually zero, we might replace TestAndClear with
TestAndTestAndClear: we avoid the lock cycle if the bit was already 0.
Atomic operations are extremely slow, e.g. the ia64 spinlock asm code
internally uses TestAndTestAndSet in spin_lock();
--
Manfred
-
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/
This archive was generated by hypermail 2b29 : Fri Jun 23 2000 - 21:00:11 EST