[Pmem] [PATCH 10/15] memory tiering: avoid kswap to be keep in failure state for long

From: Huang Ying <ying.huang(a)intel.com> ANBZ: #80 cherry-picked from https://git.kernel.org/pub/scm/linux/kernel/git/vishal/tiering.git/commit/?… In memory tiering benchmark testing with THP (Transparent Huge Page) enabled, we found that sometimes demotion/promotion between the fast and slow memory nodes may pause for several hundreds seconds, even if the THP failed to be isolated live lock has been fixed. Further digging found that sometimes the pages in the fast memory node becomes so hot, that kswapd fails to reclaim any page. Finally, the kswapd failure count (pgdat->kswapd_failures) reaches its max value. So kswapd will not been waken up until a successful direct reclaiming. For general use case, this isn't a big problem, because the memory users will do direct reclaim finally and trigger successful direct reclaiming or OOM to fix the issue. But in memory tiering system, the demotion and promotion will avoid to create too much memory pressure on the fast memory node, so direct reclaiming will not be triggered to resolve the issue. Even if the access pattern of the applications have changed, we can not adjust the placement of the hot/cold pages to optimize the performance. To resolve the issue, in this patch, in memory tiering mode, kswapd will be waken up even 10 seconds to try to free some pages to recover kswapd failures. Signed-off-by: "Huang, Ying" <ying.huang(a)intel.com> Signed-off-by: Baolin Wang <baolin.wang(a)linux.alibaba.com> --- mm/vmscan.c | 19 +++++++++++++++++-- 1 file changed, 17 insertions(+), 2 deletions(-) diff --git a/mm/vmscan.c b/mm/vmscan.c index fa9e268..abb4579 100644 --- a/mm/vmscan.c +++ b/mm/vmscan.c @@ -4002,8 +4002,23 @@ static void kswapd_try_to_sleep(pg_data_t *pgdat, int alloc_order, int reclaim_o */ set_pgdat_percpu_threshold(pgdat, calculate_normal_threshold); - if (!kthread_should_stop()) - schedule(); + if (!kthread_should_stop()) { + /* + * In memory tiering mode, try harder to recover from + * kswapd failures, because direct reclaiming may be + * not triggered. + */ + if (sysctl_numa_balancing_mode & NUMA_BALANCING_MEMORY_TIERING && + node_is_toptier(pgdat->node_id) && + pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES) { + remaining = schedule_timeout(10 * HZ); + if (!remaining) { + pgdat->kswapd_classzone_idx = ZONE_MOVABLE; + pgdat->kswapd_order = 0; + } + } else + schedule(); + } set_pgdat_percpu_threshold(pgdat, calculate_pressure_threshold); } else { -- 1.8.3.1