Messages in this thread | | | Date | Tue, 06 Jul 2010 10:26:24 +0800 | From | Xiaotian Feng <> | Subject | Re: [regression] commit 6a4f3b52 causes invalid memtype |
| |
On 07/05/2010 07:38 PM, Ali Gholami Rudi wrote: > Ingo Molnar<mingo@elte.hu> wrote: >>>>>> commit 2463eb8b3093995e09a0d41b3d78ee0cf5fb4249 >>>>>> Author: Peter Zijlstra<peterz@infradead.org> >>>>>> Date: Sat May 29 15:31:43 2010 +0200 >>>>>> >>>>>> rbtree: Undo augmented trees performance damage >>>>> >>>>> Where can I get it? I couldn't find it either in >>>>> Linus' or x86 repo tip. >>>> http://git.kernel.org/tip/2463eb8b3093995e09a0d41b3d78ee0cf5fb4249 >>> >>> It does fix it. >> >> Hm, interesting. That's a pure performance fix - so why does it fix a >> regression? >> >> Does Peter's augmented rbtrees improvements fix a bug too, as a side-effect? >> If yes then we need to accelerate that commit - right now it's aimed for >> v2.6.36. > > Since the bug appears after commit 6a4f3b52, it seems to be > related to the value of memtype->subtree_max_end. Peter's > patch seems to change the way memtype->subtree_max_end is > updated, which seems to fix the problem too.
Actually, this exposed a bug in augmented rbtree (http://lkml.org/lkml/2010/6/21/459). With commit 6a4f3b52, the bug was easier to trigger :-)
> > Thanks, > Ali > >
| |