Re: [PATCH 1/4] [BUGFIX] enhance charge_statistics function forfixising issues

[Johannes Weiner]

On Fri, Jan 14, 2011 at 07:06:44PM +0900, KAMEZAWA Hiroyuki wrote:
> mem_cgroup_charge_staistics() was designed for charging a page but
> now, we have transparent hugepage. To fix problems (in following patch)
> it's required to change the function to get the number of pages
> as its arguments.
> 
> The new function gets following as argument.
>   - type of page rather than 'pc'
>   - size of page which is accounted.
> 
> Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx>

I agree with the patch in general, below are only a few nitpicks.

> --- mmotm-0107.orig/mm/memcontrol.c
> +++ mmotm-0107/mm/memcontrol.c
> @@ -600,23 +600,23 @@ static void mem_cgroup_swap_statistics(s
>  }
>  
>  static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
> -					 struct page_cgroup *pc,
> -					 bool charge)
> +					 bool file,
> +					 int pages)

I think 'nr_pages' would be a better name.  This makes me think of a
'struct page *[]'.

>  {
> -	int val = (charge) ? 1 : -1;
> -
>  	preempt_disable();
>  
> -	if (PageCgroupCache(pc))
> -		__this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_CACHE], val);
> +	if (file)
> +		__this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_CACHE], pages);
>  	else
> -		__this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_RSS], val);
> +		__this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_RSS], pages);
>  
> -	if (charge)
> +	/* pagein of a big page is an event. So, ignore page size */
> +	if (pages > 0)
>  		__this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGIN_COUNT]);
>  	else
>  		__this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGOUT_COUNT]);
> -	__this_cpu_inc(mem->stat->count[MEM_CGROUP_EVENTS]);
> +
> +	__this_cpu_add(mem->stat->count[MEM_CGROUP_EVENTS], pages);
>  
>  	preempt_enable();
>  }
> @@ -2092,6 +2092,7 @@ static void ____mem_cgroup_commit_charge
>  					 struct page_cgroup *pc,
>  					 enum charge_type ctype)
>  {
> +	bool file = false;
>  	pc->mem_cgroup = mem;
>  	/*
>  	 * We access a page_cgroup asynchronously without lock_page_cgroup().
> @@ -2106,6 +2107,7 @@ static void ____mem_cgroup_commit_charge
>  	case MEM_CGROUP_CHARGE_TYPE_SHMEM:
>  		SetPageCgroupCache(pc);
>  		SetPageCgroupUsed(pc);
> +		file = true;
>  		break;
>  	case MEM_CGROUP_CHARGE_TYPE_MAPPED:
>  		ClearPageCgroupCache(pc);
> @@ -2115,7 +2117,7 @@ static void ____mem_cgroup_commit_charge
>  		break;
>  	}
>  
> -	mem_cgroup_charge_statistics(mem, pc, true);
> +	mem_cgroup_charge_statistics(mem, file, 1);

The extra local variable is a bit awkward, since there are already
several sources of this information (ctype and pc->flags).

Could you keep it like the other sites, just pass PageCgroupCache()
here as well?

> @@ -2186,14 +2188,14 @@ static void __mem_cgroup_move_account(st
>  		__this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
>  		preempt_enable();
>  	}
> -	mem_cgroup_charge_statistics(from, pc, false);
> +	mem_cgroup_charge_statistics(from, PageCgroupCache(pc), -1);
>  	if (uncharge)
>  		/* This is not "cancel", but cancel_charge does all we need. */
>  		mem_cgroup_cancel_charge(from, PAGE_SIZE);
>  
>  	/* caller should have done css_get */
>  	pc->mem_cgroup = to;
> -	mem_cgroup_charge_statistics(to, pc, true);
> +	mem_cgroup_charge_statistics(to, PageCgroupCache(pc), 1);
>  	/*
>  	 * We charges against "to" which may not have any tasks. Then, "to"
>  	 * can be under rmdir(). But in current implementation, caller of
> @@ -2551,6 +2553,7 @@ __mem_cgroup_uncharge_common(struct page
>  	struct page_cgroup *pc;
>  	struct mem_cgroup *mem = NULL;
>  	int page_size = PAGE_SIZE;
> +	bool file = false;
>  
>  	if (mem_cgroup_disabled())
>  		return NULL;
> @@ -2578,6 +2581,9 @@ __mem_cgroup_uncharge_common(struct page
>  	if (!PageCgroupUsed(pc))
>  		goto unlock_out;
>  
> +	if (PageCgroupCache(pc))
> +		file = true;
> +
>  	switch (ctype) {
>  	case MEM_CGROUP_CHARGE_TYPE_MAPPED:
>  	case MEM_CGROUP_CHARGE_TYPE_DROP:
> @@ -2597,7 +2603,7 @@ __mem_cgroup_uncharge_common(struct page
>  	}
>  
>  	for (i = 0; i < count; i++)
> -		mem_cgroup_charge_statistics(mem, pc + i, false);
> +		mem_cgroup_charge_statistics(mem, file, -1);

I see you get rid of this loop in the next patch, anyway.  Can you
just use PageCgroupCache() instead of the extra variable?
--
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/