compaction: check lock contention first before taking lock

isolate_migratepages_range will take zone->lru_lock first and check if the lock is contented, if yes, it will release the lock. This isn't efficient. If the lock is truly contented, a lock/unlock pair will increase the lock contention. We'd better check if the lock is contended first. compact_trylock_irqsave perfectly meets the requirement.

Signed-off-by Shaohua Li <[email protected]>

---
 mm/compaction.c |    7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

Index: linux/mm/compaction.c
===================================================================

mm/compaction.c +1 -0

 	for (; low_pfn < end_pfn; low_pfn++) {
 		struct page *page;
 

@@ -400,6 +400,7 @@ isolate_migratepages_range(struct zone *
 	if (locked)
 		spin_unlock_irqrestore(&zone->lru_lock, flags);
 
+skip:
 	trace_mm_compaction_isolate_migratepages(nr_scanned, nr_isolated);
 
 	if (!nr_isolated)

Quoted 18 lines · Shaohua Li

isolate_migratepages_range will take zone->lru_lock first and check if the lock is contented, if yes, it will release the lock. This isn't efficient. If the lock is truly contented, a lock/unlock pair will increase the lock contention. We'd better check if the lock is contended first. compact_trylock_irqsave perfectly meets the requirement.

Signed-off-by Shaohua Li <[email protected]>

---
 mm/compaction.c |    7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

Index: linux/mm/compaction.c
===================================================================

mm/compaction.c +0 -0

@@ -295,9 +295,9 @@ isolate_migratepages_range(struct zone *
 	}
 

mm/compaction.c isolate_migratepages_range(struct zone * · line 298

Quoted 2 lines

mm/compaction.c +0 -1

 	/* Time to isolate some pages for migration */
-	cond_resched();

Why did you remove the cond_resched()? I expect it's because compact_checklock_irqsave() does a need_resched() check and if it is true will either call cond_resched() or abort compaction. If it is aborting it will not call cond_resched() but there is a reasonable expectation that the caller will schedule soon. If this is the reasoning then it should be included in the changelog. If it's an accident then leave the cond_resched() where it is.

mm/compaction.c isolate_migratepages_range(struct zone * · line 300

Quoted 5 lines

mm/compaction.c +3 -2

-	spin_lock_irqsave(&zone->lru_lock, flags);
-	locked = true;
+	locked = compact_trylock_irqsave(&zone->lru_lock, &flags, cc);
+	if (!locked)
+		goto skip;

There is no need for the goto. No useful work has taken place at this point and there is no need to even trigger the tracepoint. Just return 0.