Re: [PATCH] [RFC] mm/compaction: initialize compaction information

[Gioh Kim]

2015-03-19 ìí 5:41ì Vlastimil Babka ì(ê) ì ê:
> On 03/19/2015 06:30 AM, Gioh Kim wrote:
> > I tried to start compaction via /proc/sys/vm/compact_memory
> > as soon as I turned on my ARM-based platform.
> > But the compaction didn't start.
> > I found some variables in struct zone are not initalized.
> >
> > I think zone->compact_cached_free_pfn and some cache values for compaction
> > are initalized when the kernel starts compaction, not via
> > /proc/sys/vm/compact_memory.
> > If my guess is correct, an initialization are needed for that case.
> >
> >
> > Signed-off-by: Gioh Kim <gioh.kim@xxxxxxx>
> > ---
> >   mm/compaction.c |    8 ++++++++
> >   1 file changed, 8 insertions(+)
> >
> > diff --git a/mm/compaction.c b/mm/compaction.c
> > index 8c0d945..944a9cc 100644
> > --- a/mm/compaction.c
> > +++ b/mm/compaction.c
> > @@ -1299,6 +1299,14 @@ static int compact_zone(struct zone *zone, struct compact_control *cc)
> >   		__reset_isolation_suitable(zone);
> >
> >   	/*
> > +	 * If this is activated by /proc/sys/vm/compact_memory
> > +	 * and the first try, cached information for compaction is not
> > +	 * initialized.
> > +	 */
> > +	if (cc->order == -1 && zone->compact_cached_free_pfn == 0)
> > +		__reset_isolation_suitable(zone);
> > +
> > +	/*
> >   	 * Setup to move all movable pages to the end of the zone. Used cached
> >   	 * information on where the scanners should start but check that it
> >   	 * is initialised by ensuring the values are within zone boundaries.
>
> The code below this comment already does the initialization if the cached values
> are outside zone boundaries (e.g. due to not being initialized). So if I go
> through what your __reset_isolation_suitable(zone) call possibly fixes:
>
> - the code below comment should take care of zone->compact_cached_migrate_pfn
> and zone->compact_cached_free_pfn.
> - the value of zone->compact_blockskip_flush shouldn't affect whether compaction
> is done.
> - the state of pageblock_skip bits shouldn't matter for compaction via
> /proc/sys... as that sets ignore_skip_hint = true
>
> It might be perhaps possible that the cached scanner positions are close to
> meeting and compaction occurs but doesn't process much. That would be also true
> if both were zero, but at least on my x86 system, lowest zone's start_pfn is 1
> so that would be detected and corrected. Maybe it is zero on yours though? (ARM?).

YES, it is. As comment above, my platform is based on ARM.
zone's start_pfn is 0.

> So in any case, the problem should be identified in more detail so we know the
> fix is not accidental. It could be also worthwile to always reset scanner
> positions when doing a /proc triggered compaction, so it's not depending on what
> happened before.

Excuse my poor english.
I cannot catch exactly what you want.
Is this what you want? This resets the position if compaction is started via /proc.


diff --git a/mm/compaction.c b/mm/compaction.c
index 8c0d945..827ec06 100644
--- a/mm/compaction.c
+++ b/mm/compaction.c
@@ -1587,8 +1587,10 @@ static void __compact_pgdat(pg_data_t *pgdat, struct compact_control *cc)
                INIT_LIST_HEAD(&cc->freepages);
                INIT_LIST_HEAD(&cc->migratepages);

-               if (cc->order == -1 || !compaction_deferred(zone, cc->order))
+               if (cc->order == -1 || !compaction_deferred(zone, cc->order)) {
+                       __reset_isolation_suitable(zone);
                        compact_zone(zone, cc);
+               }

                if (cc->order > 0) {
                        if (zone_watermark_ok(zone, cc->order,

--
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/