Message-Id: <1417809545-4540-1-git-send-email-vbabka@suse.cz>
Date:	Fri,  5 Dec 2014 20:59:01 +0100
From:	Vlastimil Babka <vbabka@...e.cz>
To:	linux-mm@...ck.org
Cc:	linux-kernel@...r.kernel.org, Joonsoo Kim <iamjoonsoo.kim@....com>,
	Minchan Kim <minchan@...nel.org>, Mel Gorman <mgorman@...e.de>,
	Rik van Riel <riel@...hat.com>,
	David Rientjes <rientjes@...gle.com>,
	Andrew Morton <akpm@...ux-foundation.org>,
	Hugh Dickins <hughd@...gle.com>,
	Linus Torvalds <torvalds@...ux-foundation.org>,
	Vlastimil Babka <vbabka@...e.cz>
Subject: [RFC PATCH V2 0/4] Reducing parameters of alloc_pages* family of functions

Hey all,

this is a V2 of attempting something that has been discussed when Minchan
proposed to expand the x86 kernel stack [1], namely the reduction of huge
number of parameters that the alloc_pages* family and get_page_from_freelist()
functions have.

The result is this series, ordered in the subjective importance of the patches.
Note that it's only build-tested and considered RFC. I would like some feedback
whether this is worth finishing and posting properly, and I welcome testing
with different configs/arches/gcc versions. I also welcome feedback and
suggestions on the evaluation methodology as this probably doesn't tell the
whole picture.

The series is based on mmotm-2014-12-02-15-55 and I use gcc 4.8.3 20140627 on
openSUSE 13.2. Config includess NUMA and COMPACTION, I can post whole if
needed.

The core is a new struct alloc_context, which looks like this:

struct alloc_context {
        struct zonelist *zonelist;
        nodemask_t *nodemask;
        struct zone *preferred_zone;

        unsigned int order;
        int classzone_idx;
        int migratetype;
        enum zone_type high_zoneidx;
};

All the contents is mostly constant, except that __alloc_pages_slowpath()
changes preferred_zone, classzone_idx and potentially zonelist. But that's not
a problem in case control returns to retry_cpuset: in __alloc_pages_nodemask(),
those will be reset to initial values again (although it's subtle).
On the other hand, gfp_flags and alloc_info mutate so much that it doesn't
make sense to put them into alloc_context. Still, the result is one parameter
instead of up to 7. This is all in Patch 2.

Patch 3 is a step to expand alloc_context usage out of page_alloc.c itself.
The function try_to_compact_pages() can also much benefit from the parameter
reduction, but it means the struct definition has to be moved to a shared
header.

Patch 1 should IMHO be included even if the rest is deemed not useful enough.
It improves maintainability and also has some code/stack reduction. Patch 4
is OTOH a tiny optimization.

First, let's look at the code size savings by bloat-o-meter, as the patches
stack up:

Patch 1 (vs mmotm):

function                                     old     new   delta
get_page_from_freelist                      2554    2490     -64

Patch 2:

function                                     old     new   delta
__alloc_pages_nodemask                       571    2220   +1649
get_page_from_freelist                      2490    2560     +70
__alloc_pages_direct_compact                 332     302     -30
__alloc_pages_slowpath                      1878       -   -1878

Here gcc decided to inline _slowpath, so let's try comparing with Patch 1
plus forced inline of _slowpath:

add/remove: 0/0 grow/shrink: 1/2 up/down: 70/-428 (-358)
function                                     old     new   delta
get_page_from_freelist                      2490    2560     +70
__alloc_pages_direct_compact                 332     302     -30
__alloc_pages_nodemask                      2618    2220    -398

Looks like get_page_from_freelist() did benefit from getting the parameters
separately, but overal it's a win.

Patch 3:

__alloc_pages_direct_compact                 302     254     -48
try_to_compact_pages                         582     598     +16

A tiny overal win.

Patch 4:

function                                     old     new   delta
__alloc_pages_nodemask                      2220    2217      -3
nr_free_zone_pages                           129     115     -14
get_page_from_freelist                      2560    2521     -39
try_to_compact_pages                         598     592      -6

Small but clear win. A few more object files should be also affected
but were not tested.


Now stack sizes per ./scripts/checkstack.pl:

                        mmotm    P1   P2
__alloc_pages_slowpath    176   176    -
get_page_from_freelist:   160   152  160
__alloc_pages_nodemask    104   104  168 
__alloc_pages_direct_c     32    32   24

Patch 1 saves a bit, Patch 2 result muddled by inlining.
Again, let's use Patch 1 + forced inline as baseline for the rest:

                          P1i    P2    P3    P4
__alloc_pages_nodemask    240   168   168   168
get_page_from_freelist:   152   160   160   160
try_to_compact_pages       64    64    64    64
__alloc_pages_direct_c     32    24     -     -

Again, Patch 2 bloats get_page_from_freelist(), but overal is a win. The
rest doesn't affect stack usage.

[1] http://marc.info/?l=linux-mm&m=140142462528257&w=2


Vlastimil Babka (4):
  mm: set page->pfmemalloc in prep_new_page()
  mm, page_alloc: reduce number of alloc_pages* functions' parameters
  mm: reduce try_to_compact_pages parameters
  mm: microoptimize zonelist operations

 include/linux/compaction.h |  14 ++-
 include/linux/mm.h         |  11 ++
 include/linux/mmzone.h     |  12 +--
 mm/compaction.c            |  16 +--
 mm/mmzone.c                |   4 +-
 mm/page_alloc.c            | 256 ++++++++++++++++++---------------------------
 6 files changed, 136 insertions(+), 177 deletions(-)

-- 
2.1.2

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists