Re: [PATCH v12 2/6] mm: Use zone and order instead of free area in free_list manipulators

[Alexander Duyck]

On Wed, 2019-10-23 at 10:26 +0200, David Hildenbrand wrote:
> On 23.10.19 00:28, Alexander Duyck wrote:
> > From: Alexander Duyck <alexander.h.duyck@xxxxxxxxxxxxxxx>
> > 
> > In order to enable the use of the zone from the list manipulator functions
> > I will need access to the zone pointer. As it turns out most of the
> > accessors were always just being directly passed &zone->free_area[order]
> > anyway so it would make sense to just fold that into the function itself
> > and pass the zone and order as arguments instead of the free area.
> > 
> > In order to be able to reference the zone we need to move the declaration
> > of the functions down so that we have the zone defined before we define the
> > list manipulation functions. Since the functions are only used in the file
> > mm/page_alloc.c we can just move them there to reduce noise in the header.
> > 
> > Reviewed-by: Dan Williams <dan.j.williams@xxxxxxxxx>
> > Reviewed-by: David Hildenbrand <david@xxxxxxxxxx>
> > Reviewed-by: Pankaj Gupta <pagupta@xxxxxxxxxx>
> > Signed-off-by: Alexander Duyck <alexander.h.duyck@xxxxxxxxxxxxxxx>
> > ---
> >   include/linux/mmzone.h |   32 -----------------------
> >   mm/page_alloc.c        |   67 +++++++++++++++++++++++++++++++++++-------------
> >   2 files changed, 49 insertions(+), 50 deletions(-)
> 
> Did you see
> 
> https://lore.kernel.org/lkml/20191001152928.27008.8178.stgit@xxxxxxxxxxxxxxxxxxxxx/T/#m4d2bc2f37bd7bdc3ae35c4f197905c275d0ad2f9
> 
> this time?
> 
> And the difference to the old patch is only an empty line.
> 

I saw the report. However I have not had much luck reproducing it in order
to get root cause. Here are my results for linux-next 20191021 with that
patch running page_fault2 over an average of 3 runs:

Baseline:   3734692.00
This patch: 3739878.67

Also I am not so sure about these results as the same patch had passed
previously before and instead it was patch 3 that was reported as having a
-1.2% regression[1]. All I changed in response to that report was to add
page_is_reported() which just wrapped the bit test for the reported flag
in a #ifdef to avoid testing it for the blocks that were already #ifdef
wrapped anyway.

I am still trying to see if I can get access to a system that would be a
better match for the one that reported the issue. My working theory is
that maybe it requires a high core count per node to reproduce. Either
that or it is some combination of the kernel being tested on and the patch
is causing some loop to go out of alignment and become more expensive.

I also included the page_fault2 results in my cover page as that seems to
show a slight improvement with all of the patches applied.

Thanks.

- Alex

[1]: https://lore.kernel.org/lkml/20190921152522.GU15734@shao2-debian/