On 7/14/20 7:37 PM, Roman Gushchin wrote: > I've noticed a number of warnings like "vmstat_refresh: nr_free_cma > -5" or "vmstat_refresh: nr_zone_write_pending -11" on our production > hosts. The numbers of these warnings were relatively low and stable, > so it didn't look like we are systematically leaking the counters. > The corresponding vmstat counters also looked sane. > > These warnings are generated by the vmstat_refresh() function, which > assumes that atomic zone and numa counters can't go below zero. > However, on a SMP machine it's not quite right: due to per-cpu > caching it can in theory be as low as -(zone threshold) * NR_CPUs. > > For instance, let's say all cma pages are in use and NR_FREE_CMA_PAGES > reached 0. Then we've reclaimed a small number of cma pages on each > CPU except CPU0, so that most percpu NR_FREE_CMA_PAGES counters are > slightly positive (the atomic counter is still 0). Then somebody on > CPU0 consumes all these pages. The number of pages can easily exceed > the threshold and a negative value will be committed to the atomic > counter. > > To fix the problem and avoid generating false warnings, let's just > relax the condition and warn only if the value is less than minus > the maximum theoretically possible drift value, which is 125 * > number of online CPUs. It will still allow to catch systematic leaks, > but will not generate bogus warnings. > > Signed-off-by: Roman Gushchin <guro@xxxxxx> > Cc: Hugh Dickins <hughd@xxxxxxxxxx> > Signed-off-by: Roman Gushchin <guro@xxxxxx> Acked-by: Vlastimil Babka <vbabka@xxxxxxx>
