Messages in this thread |  | | | Subject | Re: REGRESSION: Performance regressions from switching anon_vma->lock to mutex | | From | Peter Zijlstra <> | | Date | Fri, 17 Jun 2011 20:18:08 +0200 |
| |
On Fri, 2011-06-17 at 11:01 -0700, Linus Torvalds wrote:
> So I do think that "page_referenced_anon()" should do a trylock, and > return "referenced" if the trylock fails. Comments?
The only problem I can immediately see with that is when a single process' anon memory is dominant, then such an allocation will never succeed in freeing these pages because the one lock will pin pretty much all anon. Then again, there's always a few file pages to drop.
That said, its rather unlikely, and iirc people were working on removing direct reclaim, or at least rely less on it.
|  |