Hi Michal, I'm over a very long kernel bisection, and if I made no mistake in testing commits, this patch fixes a kernel bug which affects my HP Compaq dc5800 machine with 32 bit Ubuntu OS. The bug manifests itself with "NMI watchdog: BUG: soft lockup - CPU#0 stuck for 23s! [kswapd0:38]" messages in 4.8 kernels, and "page allocation stalls for 47608ms, order:1, mode:0x17000c0(GFP_KERNEL_ACCOUNT|__GFP_NOTRACK)" in 4.10 kernels up to this commit. Michal, can you confirm that this patch may fix issues like the ones I encountered? If so, I'll try to get the Ubuntu kernel staff to backport this commit to Yakkety's 4.8 kernel. On the other hand, I can't seem to be able to backport this commit to 4.8 with "git cherry-pick", so maybe I need to wait for your tweaks you mentioned. Anyway, thank you very much for the fix! Regards, MegaBrutal 2017-02-28 16:11 GMT+01:00 Michal Hocko <mhocko@xxxxxxxxxx>: > > From: Michal Hocko <mhocko@xxxxxxxx> > > commit 71ab6cfe88dcf9f6e6a65eb85cf2bda20a257682 upstream. > > get_scan_count() considers the whole node LRU size when > > - doing SCAN_FILE due to many page cache inactive pages > - calculating the number of pages to scan > > In both cases this might lead to unexpected behavior especially on 32b > systems where we can expect lowmem memory pressure very often. > > A large highmem zone can easily distort SCAN_FILE heuristic because > there might be only few file pages from the eligible zones on the node > lru and we would still enforce file lru scanning which can lead to > trashing while we could still scan anonymous pages. > > The later use of lruvec_lru_size can be problematic as well. Especially > when there are not many pages from the eligible zones. We would have to > skip over many pages to find anything to reclaim but shrink_node_memcg > would only reduce the remaining number to scan by SWAP_CLUSTER_MAX at > maximum. Therefore we can end up going over a large LRU many times > without actually having chance to reclaim much if anything at all. The > closer we are out of memory on lowmem zone the worse the problem will > be. > > Fix this by filtering out all the ineligible zones when calculating the > lru size for both paths and consider only sc->reclaim_idx zones. > > The patch would need to be tweaked a bit to apply to 4.10 and older but > I will do that as soon as it hits the Linus tree in the next merge > window. > > Link: http://lkml.kernel.org/r/20170117103702.28542-3-mhocko@xxxxxxxxxx > Fixes: b2e18757f2c9 ("mm, vmscan: begin reclaiming pages on a per-node basis") > Signed-off-by: Michal Hocko <mhocko@xxxxxxxx> > Tested-by: Trevor Cordes <trevor@xxxxxxxxxxxxx> > Acked-by: Minchan Kim <minchan@xxxxxxxxxx> > Acked-by: Hillf Danton <hillf.zj@xxxxxxxxxxxxxxx> > Acked-by: Mel Gorman <mgorman@xxxxxxx> > Acked-by: Johannes Weiner <hannes@xxxxxxxxxxx> > Cc: <stable@xxxxxxxxxxxxxxx> [4.8+] > Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> > Signed-off-by: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> > --- > mm/vmscan.c | 4 ++-- > 1 file changed, 2 insertions(+), 2 deletions(-) > > diff --git a/mm/vmscan.c b/mm/vmscan.c > index cd516c632e8f..30a88b945a44 100644 > --- a/mm/vmscan.c > +++ b/mm/vmscan.c > @@ -2205,7 +2205,7 @@ static void get_scan_count(struct lruvec *lruvec, struct mem_cgroup *memcg, > * system is under heavy pressure. > */ > if (!inactive_list_is_low(lruvec, true, sc) && > - lruvec_lru_size(lruvec, LRU_INACTIVE_FILE, MAX_NR_ZONES) >> sc->priority) { > + lruvec_lru_size(lruvec, LRU_INACTIVE_FILE, sc->reclaim_idx) >> sc->priority) { > scan_balance = SCAN_FILE; > goto out; > } > @@ -2272,7 +2272,7 @@ static void get_scan_count(struct lruvec *lruvec, struct mem_cgroup *memcg, > unsigned long size; > unsigned long scan; > > - size = lruvec_lru_size(lruvec, lru, MAX_NR_ZONES); > + size = lruvec_lru_size(lruvec, lru, sc->reclaim_idx); > scan = size >> sc->priority; > > if (!scan && pass && force_scan) > -- > 2.11.0 >