hm, as far as i can see, your patch dirties both cachelines of the target
struct page on every lookup (page_splay_tree_find()), correct? If correct,
is this just an implementational thing, or something more fundamental?
i've been thinking about getting rid of some of the cacheline dirtying in
the page lookup code, ie. something like this:
- #define SetPageReferenced(page) set_bit(PG_referenced, &(page)->flags)
+ #define SetPageReferenced(page) \
+ if (!test_bit(PG_referenced), &(page)->flags) \
+ set_bit(PG_referenced, &(page)->flags)
this would have the benefit of not touching the cacheline while doing a
simple read(), if the referenced bit is still set. (which is not cleared
too eagerly in most no-VM-pressure cases.) And it should not be a problem
that this is not race-free - missing to set the referenced bit is not a
big failure.
Ingo
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Fri Nov 30 2001 - 21:00:22 EST