Re: [PATCH for mmotm 2/5]

From: Mel Gorman
Date: Fri Jun 12 2009 - 06:22:23 EST


On Thu, Jun 11, 2009 at 08:50:06PM +0900, KOSAKI Motohiro wrote:
> > On Thu, Jun 11, 2009 at 07:26:48PM +0900, KOSAKI Motohiro wrote:
> > > Changes since Wu's original patch
> > > - adding vmstat
> > > - rename NR_TMPFS_MAPPED to NR_SWAP_BACKED_FILE_MAPPED
> > >
> > >
> > > ----------------------
> > > Subject: [PATCH] introduce NR_SWAP_BACKED_FILE_MAPPED zone stat
> >
> > This got lost in the actual subject line.
> >
> > > Desirable zone reclaim implementaion want to know the number of
> > > file-backed and unmapped pages.
> > >
> >
> > There needs to be more justification for this. We need an example
> > failure case that this addresses. For example, Patch 1 of my series was
> > to address the following problem included with the patchset leader
> >
> > "The reported problem was that malloc() stalled for a long time (minutes
> > in some cases) if a large tmpfs mount was occupying a large percentage of
> > memory overall. The pages did not get cleaned or reclaimed by zone_reclaim()
> > because the zone_reclaim_mode was unsuitable, but the lists are uselessly
> > scanned frequencly making the CPU spin at near 100%."
> >
> > We should have a similar case.
> >
> > What "desirable" zone_reclaim() should be spelled out as well. Minimally
> > something like
> >
> > "For zone_reclaim() to be efficient, it must be able to detect in advance
> > if the LRU scan will reclaim the necessary pages with the limitations of
> > the current zone_reclaim_mode. Otherwise, the CPU usage is increases as
> > zone_reclaim() uselessly scans the LRU list.
> >
> > The problem with the heuristic is ....
> >
> > This patch fixes the heuristic by ...."
> >
> > etc?
> >
> > I'm not trying to be awkward. I believe I provided similar reasoning
> > with my own patchset.
>
> You are right. my intention is not actual issue, it only fix
> documentation lie.
>
> Documentation/sysctl/vm.txt says
> =============================================================
>
> min_unmapped_ratio:
>
> This is available only on NUMA kernels.
>
> A percentage of the total pages in each zone. Zone reclaim will only
> occur if more than this percentage of pages are file backed and unmapped.
> This is to insure that a minimal amount of local pages is still available for
> file I/O even if the node is overallocated.
>
> The default is 1 percent.
> ==============================================================
>
> but actual code don't account "percentage of file backed and unmapped".
> Administrator can't imazine current implementation form this documentation.
>

That's a good point. I've suggested alternative documentation in another
thread.

> Plus, I don't think this patch is too messy. thus I did decide to make
> this fix.
>
> if anyone provide good documentation fix, my worry will vanish.
>

Hopefully your worry has vanished.

While I have no objection to the patch as such, I would like to know
what it's fixing. Believe me, if the scan-heuristic breaks again, this
patch would be one of the first things I considered as a fix :/

>
>
> > > Thus, we need to know number of swap-backed mapped pages for
> > > calculate above number.
> > >
> > >
> > > Cc: Mel Gorman <mel@xxxxxxxxx>
> > > Signed-off-by: Wu Fengguang <fengguang.wu@xxxxxxxxx>
> > > Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@xxxxxxxxxxxxxx>
> > > ---
> > > include/linux/mmzone.h | 2 ++
> > > mm/rmap.c | 7 +++++++
> > > mm/vmstat.c | 1 +
> > > 3 files changed, 10 insertions(+)
> > >
> > > Index: b/include/linux/mmzone.h
> > > ===================================================================
> > > --- a/include/linux/mmzone.h
> > > +++ b/include/linux/mmzone.h
> > > @@ -88,6 +88,8 @@ enum zone_stat_item {
> > > NR_ANON_PAGES, /* Mapped anonymous pages */
> > > NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
> > > only modified from process context */
> > > + NR_SWAP_BACKED_FILE_MAPPED, /* Similar to NR_FILE_MAPPED. but
> > > + only account swap-backed pages */
> > > NR_FILE_PAGES,
> > > NR_FILE_DIRTY,
> > > NR_WRITEBACK,
> > > Index: b/mm/rmap.c
> > > ===================================================================
> > > --- a/mm/rmap.c
> > > +++ b/mm/rmap.c
> > > @@ -829,6 +829,10 @@ void page_add_file_rmap(struct page *pag
> > > {
> > > if (atomic_inc_and_test(&page->_mapcount)) {
> > > __inc_zone_page_state(page, NR_FILE_MAPPED);
> > > + if (PageSwapBacked(page))
> > > + __inc_zone_page_state(page,
> > > + NR_SWAP_BACKED_FILE_MAPPED);
> > > +
> > > mem_cgroup_update_mapped_file_stat(page, 1);
> > > }
> > > }
> > > @@ -884,6 +888,9 @@ void page_remove_rmap(struct page *page)
> > > __dec_zone_page_state(page, NR_ANON_PAGES);
> > > } else {
> > > __dec_zone_page_state(page, NR_FILE_MAPPED);
> > > + if (PageSwapBacked(page))
> > > + __dec_zone_page_state(page,
> > > + NR_SWAP_BACKED_FILE_MAPPED);
> > > }
> > > mem_cgroup_update_mapped_file_stat(page, -1);
> > > /*
> > > Index: b/mm/vmstat.c
> > > ===================================================================
> > > --- a/mm/vmstat.c
> > > +++ b/mm/vmstat.c
> > > @@ -633,6 +633,7 @@ static const char * const vmstat_text[]
> > > "nr_mlock",
> > > "nr_anon_pages",
> > > "nr_mapped",
> > > + "nr_swap_backed_file_mapped",
> > > "nr_file_pages",
> > > "nr_dirty",
> > > "nr_writeback",
> > >
> >
> > Otherwise the patch seems reasonable.
> >
> > --
> > Mel Gorman
> > Part-time Phd Student Linux Technology Center
> > University of Limerick IBM Dublin Software Lab
>
>
>

--
Mel Gorman
Part-time Phd Student Linux Technology Center
University of Limerick IBM Dublin Software Lab
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/