Re: [PATCH 2/3] mm/compaction: add more trace to understand compaction start/finish condition

From: Joonsoo Kim
Date: Thu Jan 08 2015 - 03:21:12 EST


On Tue, Jan 06, 2015 at 12:04:28PM +0100, Vlastimil Babka wrote:
> On 12/03/2014 08:52 AM, Joonsoo Kim wrote:
> > It is not well analyzed that when compaction start and when compaction
> > finish. With this tracepoint for compaction start/finish condition, I can
> > find following bug.
> >
> > http://www.spinics.net/lists/linux-mm/msg81582.html
> >
> > Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@xxxxxxx>
> > ---
> > include/linux/compaction.h | 2 +
> > include/trace/events/compaction.h | 91 +++++++++++++++++++++++++++++++++++++
> > mm/compaction.c | 40 ++++++++++++++--
> > 3 files changed, 129 insertions(+), 4 deletions(-)
> >
> > diff --git a/include/linux/compaction.h b/include/linux/compaction.h
> > index a9547b6..bdb4b99 100644
> > --- a/include/linux/compaction.h
> > +++ b/include/linux/compaction.h
> > @@ -12,6 +12,8 @@
> > #define COMPACT_PARTIAL 3
> > /* The full zone was compacted */
> > #define COMPACT_COMPLETE 4
> > +/* For more detailed tracepoint output, will be converted to COMPACT_CONTINUE */
> > +#define COMPACT_NOT_SUITABLE 5
>
> So this makes it sound like the value means "compaction was not suitable to do
> in this zone", but later it means something different.
>
> > /* When adding new state, please change compaction_status_string, too */
> >
> > /* Used to signal whether compaction detected need_sched() or lock contention */
> > diff --git a/include/trace/events/compaction.h b/include/trace/events/compaction.h
> > index 139020b..5e47cb2 100644
> > --- a/include/trace/events/compaction.h
> > +++ b/include/trace/events/compaction.h
> > @@ -164,6 +164,97 @@ TRACE_EVENT(mm_compaction_end,
> > compaction_status_string[__entry->status])
> > );
> >
> > +TRACE_EVENT(mm_compaction_try_to_compact_pages,
> > +
> > + TP_PROTO(
> > + unsigned int order,
> > + gfp_t gfp_mask,
> > + enum migrate_mode mode,
> > + int alloc_flags,
> > + int classzone_idx),
> > +
> > + TP_ARGS(order, gfp_mask, mode, alloc_flags, classzone_idx),
> > +
> > + TP_STRUCT__entry(
> > + __field(unsigned int, order)
> > + __field(gfp_t, gfp_mask)
> > + __field(enum migrate_mode, mode)
> > + __field(int, alloc_flags)
> > + __field(int, classzone_idx)
> > + ),
> > +
> > + TP_fast_assign(
> > + __entry->order = order;
> > + __entry->gfp_mask = gfp_mask;
> > + __entry->mode = mode;
> > + __entry->alloc_flags = alloc_flags;
> > + __entry->classzone_idx = classzone_idx;
> > + ),
> > +
> > + TP_printk("order=%u gfp_mask=0x%x mode=%d alloc_flags=0x%x classzone_idx=%d",
> > + __entry->order,
> > + __entry->gfp_mask,
> > + (int)__entry->mode,
> > + __entry->alloc_flags,
> > + __entry->classzone_idx)
> > +);
> > +
> > +DECLARE_EVENT_CLASS(mm_compaction_suitable_template,
> > +
> > + TP_PROTO(struct zone *zone,
> > + unsigned int order,
> > + int alloc_flags,
> > + int classzone_idx,
> > + int ret),
> > +
> > + TP_ARGS(zone, order, alloc_flags, classzone_idx, ret),
> > +
> > + TP_STRUCT__entry(
> > + __field(char *, name)
> > + __field(unsigned int, order)
> > + __field(int, alloc_flags)
> > + __field(int, classzone_idx)
> > + __field(int, ret)
> > + ),
> > +
> > + TP_fast_assign(
> > + __entry->name = (char *)zone->name;
>
> This does not identify the NUMA node, just the zone type, isn't it?

Will fix.

>
> > + __entry->order = order;
> > + __entry->alloc_flags = alloc_flags;
> > + __entry->classzone_idx = classzone_idx;
> > + __entry->ret = ret;
> > + ),
> > +
> > + TP_printk("zone=%-8s order=%u alloc_flags=0x%x classzone_idx=%d ret=%s",
> > + __entry->name,
> > + __entry->order,
> > + __entry->alloc_flags,
> > + __entry->classzone_idx,
> > + compaction_status_string[__entry->ret])
> > +);
> > +
> > +DEFINE_EVENT(mm_compaction_suitable_template, mm_compaction_finished,
> > +
> > + TP_PROTO(struct zone *zone,
> > + unsigned int order,
> > + int alloc_flags,
> > + int classzone_idx,
> > + int ret),
> > +
> > + TP_ARGS(zone, order, alloc_flags, classzone_idx, ret)
> > +);
> > +
> > +DEFINE_EVENT(mm_compaction_suitable_template, mm_compaction_suitable,
> > +
> > + TP_PROTO(struct zone *zone,
> > + unsigned int order,
> > + int alloc_flags,
> > + int classzone_idx,
> > + int ret),
> > +
> > + TP_ARGS(zone, order, alloc_flags, classzone_idx, ret)
> > +);
> > +
> > #endif /* _TRACE_COMPACTION_H */
> >
> > /* This part must be outside protection */
> > diff --git a/mm/compaction.c b/mm/compaction.c
> > index 4c7b837..f5d2405 100644
> > --- a/mm/compaction.c
> > +++ b/mm/compaction.c
> > @@ -25,6 +25,7 @@ char *compaction_status_string[] = {
> > "continue",
> > "partial",
> > "complete",
> > + "not_suitable_page",
>
> So here COMPACT_NOT_SUITABLE is interpreted as "no suitable page was found".
>
> > };
> >
> > static inline void count_compact_event(enum vm_event_item item)
> > @@ -1048,7 +1049,7 @@ static isolate_migrate_t isolate_migratepages(struct zone *zone,
> > return cc->nr_migratepages ? ISOLATE_SUCCESS : ISOLATE_NONE;
> > }
> >
> > -static int compact_finished(struct zone *zone, struct compact_control *cc,
> > +static int __compact_finished(struct zone *zone, struct compact_control *cc,
> > const int migratetype)
> > {
> > unsigned int order;
> > @@ -1103,7 +1104,21 @@ static int compact_finished(struct zone *zone, struct compact_control *cc,
> > return COMPACT_PARTIAL;
> > }
> >
> > - return COMPACT_CONTINUE;
> > + return COMPACT_NOT_SUITABLE;
>
> So for compact_finished tracepoint you print "not_suitable_page" and it's what
> it really means - watermarks were met, but no suitable page was actually found.
> But you use "COMPACT_NOT_SUITABLE" which hints at a different meaning.
>
> > +}
> > +
> > +static int compact_finished(struct zone *zone, struct compact_control *cc,
> > + const int migratetype)
> > +{
> > + int ret;
> > +
> > + ret = __compact_finished(zone, cc, migratetype);
> > + trace_mm_compaction_finished(zone, cc->order, cc->alloc_flags,
> > + cc->classzone_idx, ret);
> > + if (ret == COMPACT_NOT_SUITABLE)
> > + ret = COMPACT_CONTINUE;
> > +
> > + return ret;
> > }
> >
> > /*
> > @@ -1113,7 +1128,7 @@ static int compact_finished(struct zone *zone, struct compact_control *cc,
> > * COMPACT_PARTIAL - If the allocation would succeed without compaction
> > * COMPACT_CONTINUE - If compaction should run now
> > */
> > -unsigned long compaction_suitable(struct zone *zone, int order,
> > +static unsigned long __compaction_suitable(struct zone *zone, int order,
> > int alloc_flags, int classzone_idx)
> > {
> > int fragindex;
> > @@ -1157,11 +1172,25 @@ unsigned long compaction_suitable(struct zone *zone, int order,
> > */
> > fragindex = fragmentation_index(zone, order);
> > if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold)
> > - return COMPACT_SKIPPED;
> > + return COMPACT_NOT_SUITABLE;
>
> But here in compaction_suitable, return here means that fragmentation seems to
> be low and it's unlikely that compaction will help. COMPACT_NOT_SUITABLE sounds
> like a good name, but then tracepoint prints "not_suitable_page" and that's
> something different.

Okay. How about adding one more like below?

#define COMPACT_NO_SUITABLE_PAGE
#define COMPACT_NOT_SUITABLE_ZONE

It will distiguish return value properly.

> > return COMPACT_CONTINUE;
> > }
> >
> > +unsigned long compaction_suitable(struct zone *zone, int order,
> > + int alloc_flags, int classzone_idx)
> > +{
> > + unsigned long ret;
> > +
> > + ret = __compaction_suitable(zone, order, alloc_flags, classzone_idx);
> > + trace_mm_compaction_suitable(zone, order, alloc_flags,
> > + classzone_idx, ret);
> > + if (ret == COMPACT_NOT_SUITABLE)
> > + ret = COMPACT_SKIPPED;
>
> I don't like this wrapping just for tracepints, but I don't know of a better way :/

Yes, I don't like it, too. :/

Thanks.
--
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/