Re: [PATCH] mm: /prom/pid/clear_refs: avoid split_huge_page()

2014-05-22 Thread Cyrill Gorcunov
On Thu, May 22, 2014 at 09:32:47AM +0400, Cyrill Gorcunov wrote: > On Thu, May 22, 2014 at 04:11:10AM +0300, Kirill A. Shutemov wrote: > > Andrew Morton wrote: > > > On Wed, 21 May 2014 22:04:22 +0300 "Kirill A. Shutemov" > > > wrote: > > > > > > > Currently we split all THP pages on any clear_r

Re: [PATCH] mm: /prom/pid/clear_refs: avoid split_huge_page()

2014-05-21 Thread Cyrill Gorcunov
On Thu, May 22, 2014 at 04:11:10AM +0300, Kirill A. Shutemov wrote: > Andrew Morton wrote: > > On Wed, 21 May 2014 22:04:22 +0300 "Kirill A. Shutemov" > > wrote: > > > > > Currently we split all THP pages on any clear_refs request. It's not > > > necessary. We can handle this on PMD level. > > >

Re: [PATCH] mm: /prom/pid/clear_refs: avoid split_huge_page()

2014-05-21 Thread Kirill A. Shutemov
this a request for Cyrill & co to > help? Cyrill, Pavel, could you take care of this? > Perhaps this is post-3.15 material. Sure. From: "Kirill A. Shutemov" Date: Thu, 22 May 2014 03:44:38 +0300 Subject: [PATCH] mm: /prom/pid/clear_refs: avoid split_huge_page() Currently

Re: [PATCH] mm: /prom/pid/clear_refs: avoid split_huge_page()

2014-05-21 Thread Cyrill Gorcunov
On Wed, May 21, 2014 at 12:34:46PM -0700, Andrew Morton wrote: > On Wed, 21 May 2014 22:04:22 +0300 "Kirill A. Shutemov" > wrote: > > > Currently we split all THP pages on any clear_refs request. It's not > > necessary. We can handle this on PMD level. > > > > One side effect is that soft dirty

Re: [PATCH] mm: /prom/pid/clear_refs: avoid split_huge_page()

2014-05-21 Thread Andrew Morton
On Wed, 21 May 2014 22:04:22 +0300 "Kirill A. Shutemov" wrote: > Currently we split all THP pages on any clear_refs request. It's not > necessary. We can handle this on PMD level. > > One side effect is that soft dirty will potentially see more dirty > memory, since we will mark whole THP page

Re: [PATCH] mm: /prom/pid/clear_refs: avoid split_huge_page()

2014-05-21 Thread Cyrill Gorcunov
On Wed, May 21, 2014 at 10:04:22PM +0300, Kirill A. Shutemov wrote: > Currently we split all THP pages on any clear_refs request. It's not > necessary. We can handle this on PMD level. > > One side effect is that soft dirty will potentially see more dirty > memory, since we will mark whole THP pag

[PATCH] mm: /prom/pid/clear_refs: avoid split_huge_page()

2014-05-21 Thread Kirill A. Shutemov
Currently we split all THP pages on any clear_refs request. It's not necessary. We can handle this on PMD level. One side effect is that soft dirty will potentially see more dirty memory, since we will mark whole THP page dirty at once. Sanity checked with CRIU test suite. More testing is require