On Fri, 26 Mar 2010, Andrea Arcangeli wrote: > On Fri, Mar 26, 2010 at 01:16:40PM -0400, Rik van Riel wrote: > > On 03/26/2010 01:00 PM, Andrea Arcangeli wrote: > > > From: Andrea Arcangeli<aarcange@xxxxxxxxxx> > > > > > > When swapcache is replaced by a ksm page don't leave orhpaned swap cache. > > > > Why is this part of the hugepage series? > > This is a not relevant for hugepages. There's another ksm change so I > thought I could sneak it in. It's still separated so it can be pulled > off separately as needed. It's a nice little catch, but you certainly shouldn't have buried it amidst 40 other patches unrelated to it! I was going to ack that patch and urge you to forward it to Andrew separately, but did you test whether it works? Isn't it actually a no-op? Because KSM is holding page lock across replace_page() (maybe that's something I added after you were last there - at the time I did it just from instinct, and to make a block look prettier; but later I believe it turned out to be necessary), and free_swap_cache() only works if trylock_page() succeeds. So if we want the fix, I think it would have to be reworked, perhaps slightly messier. Something needed for -stable? No, I think not, it's a situation that gradually increases memory pressure, and then memory pressure frees it (vmscan.c's __delete_from_swap_cache); it never prevents swapoff. But it would be nice to fix it all the same: thanks for spotting. Hugh -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxxx For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>