Messages in this thread | | | Date | Wed, 16 Dec 2009 19:13:12 +0900 | From | KAMEZAWA Hiroyuki <> | Subject | Re: [mm][RFC][PATCH 0/11] mm accessor updates. |
| |
On Wed, 16 Dec 2009 11:11:07 +0100 Andi Kleen <andi@firstfloor.org> wrote:
> On Wed, Dec 16, 2009 at 12:00:11PM +0900, KAMEZAWA Hiroyuki wrote: > > This is from Christoph Lameter's mm_accessor patch posted 5/Nov. > > > > Replacing all access to mm->mmap_sem with mm-accessor functions as > > mm_read_lock, > > mm_write_lock, > > etc... > > > > This kind of function allows us to improve page fault performance etc.. > > For example, skil down_read(mmap_sem) in some situation. > > (as: http://marc.info/?l=linux-mm&m=125809791306459&w=2) > > The problem is that it also slows down the writers, and we have > some workloads where writing is the bottleneck. > > I don't think this is the right trade off at this time. > Maybe you don't see my patch. (above URL) no slow down to write side. My one doesn't use percpu counter.
> Also the patches didn't fare too well in testing unfortunately. > > I suspect we'll rather need multiple locks split per address > space range.
This set doesn't include any changes of the logic. Just replace all mmap_sem. I think this is good start point (for introducing another logic etc..)
Thanks, -Kame
| |