Re: [PATCH 0/4] RFC - ksm api change into madvise
From: Izik Eidus
Date: Wed Jun 10 2009 - 02:30:32 EST
On Tue, 9 Jun 2009 20:27:29 +0100 (BST)
Hugh Dickins <hugh.dickins@xxxxxxxxxxxxx> wrote:
> On Tue, 9 Jun 2009, Hugh Dickins wrote:
> > On Tue, 9 Jun 2009, Izik Eidus wrote:
> > > How does this look like?
> > One improvment to make now, though: you've elsewhere avoided
> > the pgd,pud,pmd,pte descent in ksm.c (using get_pte instead), and
> > page_check_address() is not static to rmap.c (filemap_xip wanted
> > it), so please continue to use that. It's not exported, right, but
> > I think Chris was already decisive that we should abandon modular
> > KSM, yes?
> I think you can simplify it further, can't you? Isn't the get_pte()
> preamble in try_to_merge_one_page() just unnecessary overhead now?
> See untested code below. Or even move the trylock/unlock of the page
> into write_protect_page if you prefer. Later on we'll uninline
> rmap.c's vma_address() so you can use it instead of your
> addr_in_vma() copy.
Great!, what you think about below? another thing we want to add or to
start sending it to Andrew?
btw may i add your signed-off to this patch?
(the only thing that i changed was taking down the *orig_pte = *ptep,
so we will merge write_protected pages, and add orig_pte = __pte(0) to
avoid annoying warning message about being used uninitialized)