Messages in this thread | | | Subject | Re: [PATCH 00/25] mm: Preemptibility -v7 | From | Peter Zijlstra <> | Date | Tue, 25 Jan 2011 22:09:30 +0100 |
| |
On Tue, 2011-01-25 at 12:47 -0800, Andi Kleen wrote: > > I thought the reason for the preempt off inside the mmu gather region was > to stay on the same CPU for local/remote flushes. How would it change that?
afaik its been preempt-off solely because it was always inside a number of spinlocks, I know both Hugh and BenH worked on making it preemptible far before I started this.
I remember Hugh and Nick talking about this at OLS'06 or 07, can't really remember.
As to local/remote flushes, there is no real saying where the pages came from due to on-demand paging and the scheduler never having had a notion of home-node. Therefore freeing them wouldn't be more of less local if that is exposed to the same migration as allocation was.
| |