On 12/15/18 12:03 AM, Mel Gorman wrote: > Pages with no migration handler use a fallback hander which sometimes > works and sometimes persistently fails such as blockdev pages. Migration > will retry a number of times on these persistent pages which is wasteful > during compaction. This patch will fail migration immediately unless the > caller is in MIGRATE_SYNC mode which indicates the caller is willing to > wait while being persistent. Right. > This is not expected to help THP allocation success rates but it does > reduce latencies slightly. > > 1-socket thpfioscale > 4.20.0-rc6 4.20.0-rc6 > noreserved-v1r4 failfast-v1r4 > Amean fault-both-1 0.00 ( 0.00%) 0.00 * 0.00%* > Amean fault-both-3 2276.15 ( 0.00%) 3867.54 * -69.92%* This is rather weird. > Amean fault-both-5 4992.20 ( 0.00%) 5313.20 ( -6.43%) > Amean fault-both-7 7373.30 ( 0.00%) 7039.11 ( 4.53%) > Amean fault-both-12 11911.52 ( 0.00%) 11328.29 ( 4.90%) > Amean fault-both-18 17209.42 ( 0.00%) 16455.34 ( 4.38%) > Amean fault-both-24 20943.71 ( 0.00%) 20448.94 ( 2.36%) > Amean fault-both-30 22703.00 ( 0.00%) 21655.07 ( 4.62%) > Amean fault-both-32 22461.41 ( 0.00%) 21415.35 ( 4.66%) > > The 2-socket results are not materially different. Scan rates are > similar as expected. > > Signed-off-by: Mel Gorman <mgorman@xxxxxxxxxxxxxxxxxxx> Acked-by: Vlastimil Babka <vbabka@xxxxxxx> > --- > mm/migrate.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/mm/migrate.c b/mm/migrate.c > index df17a710e2c7..0e27a10429e2 100644 > --- a/mm/migrate.c > +++ b/mm/migrate.c > @@ -885,7 +885,7 @@ static int fallback_migrate_page(struct address_space *mapping, > */ > if (page_has_private(page) && > !try_to_release_page(page, GFP_KERNEL)) > - return -EAGAIN; > + return mode == MIGRATE_SYNC ? -EAGAIN : -EBUSY; > > return migrate_page(mapping, newpage, page, mode); > } >
