Re: [PATCH -next] mm/vmscan: fix some -Wenum-conversion warnings
From: Qian Cai
Date: Fri Nov 15 2019 - 11:22:11 EST
On Fri, 2019-11-15 at 11:17 -0500, Johannes Weiner wrote:
> On Fri, Nov 15, 2019 at 09:04:57AM -0500, Qian Cai wrote:
> > The -next commit "mm: vmscan: enforce inactive:active ratio at the
> > reclaim root" [1] introduced some GCC -Wenum-conversion warnings,
> >
> > mm/vmscan.c:2216:39: warning: implicit conversion from enumeration type
> > 'enum lru_list' to different enumeration type 'enum node_stat_item'
> > [-Wenum-conversion]
> > inactive = lruvec_page_state(lruvec, inactive_lru);
> > ~~~~~~~~~~~~~~~~~ ^~~~~~~~~~~~
> > mm/vmscan.c:2217:37: warning: implicit conversion from enumeration type
> > 'enum lru_list' to different enumeration type 'enum node_stat_item'
> > [-Wenum-conversion]
> > active = lruvec_page_state(lruvec, active_lru);
> > ~~~~~~~~~~~~~~~~~ ^~~~~~~~~~
> > mm/vmscan.c:2746:42: warning: implicit conversion from enumeration type
> > 'enum lru_list' to different enumeration type 'enum node_stat_item'
> > [-Wenum-conversion]
> > file = lruvec_page_state(target_lruvec, LRU_INACTIVE_FILE);
>
> Interesting, it doesn't show these for me with gcc-9.2.0.
Sorry, I meant to say it is clang.
>
> We should definitely fix these!
>
> > @@ -2213,8 +2213,9 @@ static bool inactive_is_low(struct lruvec *lruvec, enum lru_list inactive_lru)
> > unsigned long inactive_ratio;
> > unsigned long gb;
> >
> > - inactive = lruvec_page_state(lruvec, inactive_lru);
> > - active = lruvec_page_state(lruvec, active_lru);
> > + inactive = lruvec_page_state(lruvec,
> > + (enum node_stat_item)inactive_lru);
> > + active = lruvec_page_state(lruvec, (enum node_stat_item)active_lru);
>
> This is fragile, as it relies on the absolute values being identical,
> which we don't guarantee. However, we do guarantee the relative order
> between the LRU items, and we use NR_LRU_BASE for the translation.
>
> Please use NR_LRU_BASE + (in)active_lru here.
>
> > @@ -2743,7 +2744,8 @@ static bool shrink_node(pg_data_t *pgdat, struct scan_control *sc)
> > * thrashing, try to reclaim those first before touching
> > * anonymous pages.
> > */
> > - file = lruvec_page_state(target_lruvec, LRU_INACTIVE_FILE);
> > + file = lruvec_page_state(target_lruvec,
> > + (enum node_stat_item)LRU_INACTIVE_FILE);
>
> This should just directly use NR_INACTIVE_FILE instead.
>
> Thanks