Re: [RFC][PATCH] memcg remove css_get/put per pages v2

[Daisuke Nishimura]

> @@ -2432,15 +2463,18 @@ mem_cgroup_uncharge_swapcache(struct pag
>  	if (!swapout) /* this was a swap cache but the swap is unused ! */
>  		ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
>  
> -	memcg = __mem_cgroup_uncharge_common(page, ctype);
> +	memcg = try_get_mem_cgroup_from_page(page);
> +	if (!memcg)
> +		return;
> +
> +	__mem_cgroup_uncharge_common(page, ctype);
>  
>  	/* record memcg information */
> -	if (do_swap_account && swapout && memcg) {
> +	if (do_swap_account && swapout) {
>  		swap_cgroup_record(ent, css_id(&memcg->css));
>  		mem_cgroup_get(memcg);
>  	}
> -	if (swapout && memcg)
> -		css_put(&memcg->css);
> +	css_put(&memcg->css);
>  }
>  #endif
>  
hmm, this change seems to cause a problem.
I can see under flow of mem->memsw and "swap" field in memory.stat. 

I think doing swap_cgroup_record() against mem_cgroup which is not returned
by __mem_cgroup_uncharge_common() is a bad behavior.

How about doing like this ? We can safely access mem_cgroup while it has
memory.usage, iow, before we call res_counter_uncharge().
After this change, it seems to work well.

---
 mm/memcontrol.c |   22 +++++++++-------------
 1 files changed, 9 insertions(+), 13 deletions(-)

diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index 6e7c1c9..2fae26f 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -2362,10 +2362,6 @@ __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
 		break;
 	}
 
-	if (!mem_cgroup_is_root(mem))
-		__do_uncharge(mem, ctype);
-	if (ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
-		mem_cgroup_swap_statistics(mem, true);
 	mem_cgroup_charge_statistics(mem, pc, false);
 
 	ClearPageCgroupUsed(pc);
@@ -2379,6 +2375,12 @@ __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
 	unlock_page_cgroup(pc);
 
 	memcg_check_events(mem, page);
+	if (ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) {
+		mem_cgroup_swap_statistics(mem, true);
+		mem_cgroup_get(mem);
+	}
+	if (!mem_cgroup_is_root(mem))
+		__do_uncharge(mem, ctype);
 
 	return mem;
 
@@ -2463,18 +2465,12 @@ mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
 	if (!swapout) /* this was a swap cache but the swap is unused ! */
 		ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
 
-	memcg = try_get_mem_cgroup_from_page(page);
-	if (!memcg)
-		return;
-
-	__mem_cgroup_uncharge_common(page, ctype);
+	memcg = __mem_cgroup_uncharge_common(page, ctype);
 
 	/* record memcg information */
-	if (do_swap_account && swapout) {
+	if (do_swap_account && swapout && memcg)
+		/* We've already done mem_cgroup_get() in uncharge_common(). */
 		swap_cgroup_record(ent, css_id(&memcg->css));
-		mem_cgroup_get(memcg);
-	}
-	css_put(&memcg->css);
 }
 #endif
 
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/