kswapd skips compaction if reclaim order drops to zero?

PATCH In review Hillf Danton 4 participants 9 messages Aug 15, 2013 – Aug 16, 2013

Subsystems (4)

Memory management Page allocator MGLRU MM reclaim

lore

Share & tools

lore

Hillf Danton

author

2013-08-15 10:02 UTC #0 lore raw

If the allocation order is not high, direct compaction does nothing. Can we skip compaction here if order drops to zero?

mm/vmscan.c Thu Aug 15 17:48:58 2013 renamed from mm/vmscan.c Thu Aug 15 17:47:26 2013 +1 -1

@@ -3034,7 +3034,7 @@ static unsigned long balance_pgdat(pg_da
  * Compact if necessary and kswapd is reclaiming at least the
  * high watermark number of pages as requsted
  */
- if (pgdat_needs_compaction && sc.nr_reclaimed > nr_attempted)
+ if (pgdat_needs_compaction && sc.nr_reclaimed > nr_attempted && order)
  compact_pgdat(pgdat, order);
 
  /*

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

Mel Gorman

2013-08-15 10:47 UTC #1 lore raw

Quoted 3 lines · Hillf Danton

If the allocation order is not high, direct compaction does nothing. Can we skip compaction here if order drops to zero?

If the allocation order is not high then

pgdat_needs_compaction == (order > 0) == false == no calling compact_pdatt

In the case where order is reset to 0 due to fragmentation then it does call compact_pgdat but it does no work due to the cc->order check in __compact_pgdat.

Signature · 8 lines

Mel Gorman
SUSE Labs

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

Minchan Kim

2013-08-15 13:41 UTC #2 lore raw

Hey Mel,

Quoted 5 lines · Mel Gorman

Quoted 3 lines · Hillf Danton
If the allocation order is not high, direct compaction does nothing. Can we skip compaction here if order drops to zero?

Quoted 8 lines

If the allocation order is not high then
pgdat_needs_compaction == (order > 0) == false == no calling compact_pdatt
In the case where order is reset to 0 due to fragmentation then it does call compact_pgdat but it does no work due to the cc->order check in __compact_pgdat.

I am looking at mmotm-2013-08-07-16-55 but couldn't find cc->order check right before compact_zone in __comact_pgdat. Could you pinpoint code piece?

Quoted 9 lines

Signature · 8 lines

Mel Gorman
SUSE Labs

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

Signature · 8 lines

Kind regards,
Minchan Kim

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

Mel Gorman

2013-08-15 13:56 UTC #3 lore raw

Quoted 12 lines · Minchan Kim

Hey Mel,
Quoted 9 lines · Mel Gorman
Quoted 3 lines · Hillf Danton
If the allocation order is not high, direct compaction does nothing. Can we skip compaction here if order drops to zero?
If the allocation order is not high then
pgdat_needs_compaction == (order > 0) == false == no calling compact_pdatt

Quoted 9 lines

Quoted 4 lines
In the case where order is reset to 0 due to fragmentation then it does call compact_pgdat but it does no work due to the cc->order check in __compact_pgdat.
I am looking at mmotm-2013-08-07-16-55 but couldn't find cc->order check right before compact_zone in __comact_pgdat. Could you pinpoint code piece?

Thanks, I screwed up as that check happens too late. However, it still ends up not mattering because it does this

compact_pgdat
-> __compact_pgdat

    -> compact_zone
      -> compaction_suitable

For order == 0, compaction_suitable will return either COMPACT_SKIPPED (if the watermarks are not met) and COMPACT_PARTIAL otherwise. Either way, compaction doesn't run.

Signature · 8 lines

Mel Gorman
SUSE Labs

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

Minchan Kim

2013-08-15 14:10 UTC #4 lore raw

Quoted 26 lines · Mel Gorman

Quoted 12 lines · Minchan Kim
Hey Mel,
Quoted 9 lines · Mel Gorman
Quoted 3 lines · Hillf Danton
If the allocation order is not high, direct compaction does nothing. Can we skip compaction here if order drops to zero?
If the allocation order is not high then
pgdat_needs_compaction == (order > 0) == false == no calling compact_pdatt
Quoted 9 lines
Quoted 4 lines
In the case where order is reset to 0 due to fragmentation then it does call compact_pgdat but it does no work due to the cc->order check in __compact_pgdat.
I am looking at mmotm-2013-08-07-16-55 but couldn't find cc->order check right before compact_zone in __comact_pgdat. Could you pinpoint code piece?
Thanks, I screwed up as that check happens too late. However, it still ends up not mattering because it does this

Quoted 8 lines

compact_pgdat
-> __compact_pgdat
    -> compact_zone
      -> compaction_suitable
For order == 0, compaction_suitable will return either COMPACT_SKIPPED (if the watermarks are not met) and COMPACT_PARTIAL otherwise. Either way, compaction doesn't run.

In compaction_suitable, it could pass first zone_watermark_ok but failed second zone_watermark_ok while fragindex is -1000 so compaction could run.

And we shouldn't depend on such coincidence.

Quoted 10 lines

Signature · 9 lines

-- 
Mel Gorman
SUSE Labs

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

Signature · 8 lines

Kind regards,
Minchan Kim

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

Mel Gorman

2013-08-15 15:39 UTC #5 lore raw

PATCH 1/1 mm: compaction: Do not compact pgdat for order-0

If kswapd was reclaiming for a high order and resets it to 0 due to fragmentation it will still call compact_pgdat. For the most part, this will fail a compaction_suitable() test and not compact but it is unnecessarily sloppy. It could be fixed in the caller but fix it in the API instead.

[[email protected]: Pointed out that it was a potential problem]

Signed-off-by Mel Gorman <[email protected]>

---
 mm/compaction.c | 3 +++
 1 file changed, 3 insertions(+)

mm/compaction.c +3 -0

index 05ccb4c..c437893 100644
@@ -1131,6 +1131,9 @@ void compact_pgdat(pg_data_t *pgdat, int order)
 		.sync = false,
 	};
 
+	if (!order)
+		return;
+
 	__compact_pgdat(pgdat, &cc);
 }

Minchan Kim

Acked-by on v1 1/1

2013-08-16 04:37 UTC #6 lore raw

Re: [PATCH] mm: compaction: Do not compact pgdat for order-0

Quoted 7 lines · Mel Gorman

If kswapd was reclaiming for a high order and resets it to 0 due to fragmentation it will still call compact_pgdat. For the most part, this will fail a compaction_suitable() test and not compact but it is unnecessarily sloppy. It could be fixed in the caller but fix it in the API instead.
[[email protected]: Pointed out that it was a potential problem]
Signed-off-by Mel Gorman <[email protected]>

Acked-by Minchan Kim <[email protected]>

Signature · 8 lines

Kind regards,
Minchan Kim

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

Wanpeng Li

2013-08-16 00:25 UTC #7 lore raw

In reply to Minchan Kim

Hi Minchan,

Quoted 32 lines · Minchan Kim

Quoted 31 lines · Mel Gorman
Quoted 12 lines · Minchan Kim
Hey Mel,
Quoted 9 lines · Mel Gorman
Quoted 3 lines · Hillf Danton
If the allocation order is not high, direct compaction does nothing. Can we skip compaction here if order drops to zero?
If the allocation order is not high then
pgdat_needs_compaction == (order > 0) == false == no calling compact_pdatt
Quoted 9 lines
Quoted 4 lines
In the case where order is reset to 0 due to fragmentation then it does call compact_pgdat but it does no work due to the cc->order check in __compact_pgdat.
I am looking at mmotm-2013-08-07-16-55 but couldn't find cc->order check right before compact_zone in __comact_pgdat. Could you pinpoint code piece?
Thanks, I screwed up as that check happens too late. However, it still ends up not mattering because it does this
compact_pgdat
-> __compact_pgdat
    -> compact_zone
      -> compaction_suitable

Quoted 8 lines

Quoted 3 lines
For order == 0, compaction_suitable will return either COMPACT_SKIPPED (if the watermarks are not met) and COMPACT_PARTIAL otherwise. Either way, compaction doesn't run.
In compaction_suitable, it could pass first zone_watermark_ok but failed second zone_watermark_ok while fragindex is -1000 so compaction could run.

I'm not sure why you said that second zone_watermark_ok failed while fragindex is -1000, actually they are the same check against order 0.

First:
watermark = low_wmark_pages(zone) + (2UL << order); zone_watermark_ok(zone, 0, watermark, 0, 0);

Second:
fragindex == -1000 && zone_watermark_ok(zone, order, watermark, 0, 0)

Both COMPACT_SKIPPED and COMPACT_PARTIAL will fail compaction.

Regards,
Wanpeng Li

Quoted 8 lines

And we shouldn't depend on such coincidence.
Quoted 5 lines
Signature · 3 lines
-- 
Mel Gorman
SUSE Labs

Quoted 15 lines

Quoted 5 lines
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Signature · 8 lines
Kind regards,
Minchan Kim

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

Wanpeng Li

2013-08-16 00:25 UTC #8 lore raw

In reply to Minchan Kim

Hi Minchan,

Quoted 32 lines · Minchan Kim

Quoted 31 lines · Mel Gorman
Quoted 12 lines · Minchan Kim
Hey Mel,
Quoted 9 lines · Mel Gorman
Quoted 3 lines · Hillf Danton
If the allocation order is not high, direct compaction does nothing. Can we skip compaction here if order drops to zero?
If the allocation order is not high then
pgdat_needs_compaction == (order > 0) == false == no calling compact_pdatt
Quoted 9 lines
Quoted 4 lines
In the case where order is reset to 0 due to fragmentation then it does call compact_pgdat but it does no work due to the cc->order check in __compact_pgdat.
I am looking at mmotm-2013-08-07-16-55 but couldn't find cc->order check right before compact_zone in __comact_pgdat. Could you pinpoint code piece?
Thanks, I screwed up as that check happens too late. However, it still ends up not mattering because it does this
compact_pgdat
-> __compact_pgdat
    -> compact_zone
      -> compaction_suitable

Quoted 8 lines

Quoted 3 lines
For order == 0, compaction_suitable will return either COMPACT_SKIPPED (if the watermarks are not met) and COMPACT_PARTIAL otherwise. Either way, compaction doesn't run.
In compaction_suitable, it could pass first zone_watermark_ok but failed second zone_watermark_ok while fragindex is -1000 so compaction could run.

I'm not sure why you said that second zone_watermark_ok failed while fragindex is -1000, actually they are the same check against order 0.

First:
watermark = low_wmark_pages(zone) + (2UL << order); zone_watermark_ok(zone, 0, watermark, 0, 0);

Second:
fragindex == -1000 && zone_watermark_ok(zone, order, watermark, 0, 0)

Both COMPACT_SKIPPED and COMPACT_PARTIAL will fail compaction.

Regards,
Wanpeng Li

Quoted 8 lines

And we shouldn't depend on such coincidence.
Quoted 5 lines
Signature · 3 lines
-- 
Mel Gorman
SUSE Labs

Quoted 15 lines

Quoted 5 lines
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Signature · 8 lines
Kind regards,
Minchan Kim

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>