Re: [PATCH]mm/compation.c: checking page in lru twice

[Andrea Arcangeli]

On Fri, May 06, 2011 at 07:32:46PM +0800, Figo.zhang wrote:
> 
> in isolate_migratepages() have check page in LRU twice, the next one
> at _isolate_lru_page(). 

hugetlbfs or any other compound page won't have PageLRU set and they
may go away from under us leading to compound_order not being reliable
if we remove the PageLRU check before compound_order. So we need to
verify the page is in LRU before running compound_order safely. And if
we hold the lru_lock, the page won't be isolated under us, and we know
it's not going to get splitted either.

We might use compound_trans_order but that's only reliable if run on
the head page so it's not so reliable, and so far it's only used by
memory-failure to "diminish" the risk of races in reading the compound
order, but it's not the best having to use compound_trans_order (and
memory-failure remains unsafe w.r.t to hugetlbfs being released during
hwpoisoning, so compound_trans_order might have to be improved for
it).
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/