On Wed, 31 May 2017 14:31:16 +0800 kernel test robot <xiaolong...@intel.com>
wrote:
>
> FYI, we noticed the following commit:
>
> commit: aefd950b83d2d8cf4d3c270546c8725f866da191 ("mm: make kswapd try harder
> to keep active pages in cache")
> https://git.kernel.org/cgit/linux/kernel/git/next/linux-next.git master
>
> in testcase: boot
>
> ...
>
> [ 160.541829] divide error: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC
>
> ...
>
> [ 160.587334] RIP: shrink_node+0x47f/0x5a0 RSP: ffffc900001bbd78
>
> ...
hm. This?
--- a/mm/vmscan.c~mm-make-kswapd-try-harder-to-keep-active-pages-in-cache-fix-2
+++ a/mm/vmscan.c
@@ -2724,7 +2724,7 @@ static bool shrink_node(pg_data_t *pgdat
if (sc->nr_reclaimed - nr_reclaimed) {
reclaimable = true;
} else if (sc->inactive_only && !skip_slab) {
- unsigned long percent;
+ unsigned long percent = 100;
/*
* We didn't reclaim anything this go around, so the
@@ -2735,7 +2735,8 @@ static bool shrink_node(pg_data_t *pgdat
* hoping that eventually we'll start freeing enough
* objects to reclaim space.
*/
- percent = (slab_reclaimed * 100 / slab_scanned);
+ if (slab_scanned)
+ percent = (slab_reclaimed * 100 / slab_scanned);
if (percent < 50)
sc->inactive_only = 0;
else
Or this?
--- a/mm/vmscan.c~mm-make-kswapd-try-harder-to-keep-active-pages-in-cache-fix-3
+++ a/mm/vmscan.c
@@ -2628,7 +2628,7 @@ static bool shrink_node(pg_data_t *pgdat
};
unsigned long node_lru_pages = 0;
unsigned long slab_reclaimed = 0;
- unsigned long slab_scanned = 0;
+ unsigned long slab_scanned = 1; /* Avoid div-by-zero */
struct mem_cgroup *memcg;
nr_reclaimed = sc->nr_reclaimed;
_
