Messages in this thread | | | From | Linus Torvalds <> | Date | Fri, 17 Jun 2011 11:41:55 -0700 | Subject | Re: REGRESSION: Performance regressions from switching anon_vma->lock to mutex |
| |
On Fri, Jun 17, 2011 at 11:39 AM, Linus Torvalds <torvalds@linux-foundation.org> wrote: > > Having gone over it a bit more, I actually think I prefer to just > special-case the allocation instead.
Just to explain my thinking: the thing I disliked most about doing an allocation while holding the lock wasn't that I thought we would deadlock on page reclaim. I don't claim that kind of far-sight.
No, the thing I disliked was that if we're low on memory and actually have to wait, I disliked having the lack of concurrency. I'm ok with holding the mutex over a few more CPU cycles, but anything longer might actually hurt throughput. So the patch I just sent out should fix both the page reclaim deadlock, and avoid any problems with delays due to holding the critical lock over an expensive allocation.
Linus
| |