lkml.org 
[lkml]   [2010]   [Jun]   [9]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [RFC][PATCH] memcg remove css_get/put per pages v2
I can't find any trivial bugs from my review at the moment.
I'll do some tests.

Some minor commens.

On Wed, 9 Jun 2010 15:59:40 +0900, KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> wrote:
> Still RFC, added lkml to CC: list.
> ==
> From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
>
> Now, memory cgroup increments css(cgroup subsys state)'s reference
> count per a charged page. And the reference count is kept until
> the page is uncharged. But this has 2 bad effect.
>
> 1. Because css_get/put calls atoimic_inc()/dec, heavy call of them
> on large smp will not scale well.
> 2. Because css's refcnt cannot be in a state as "ready-to-release",
> cgroup's notify_on_release handler can't work with memcg.
>
> This is a trial to remove css's refcnt per a page. Even if we remove
> refcnt, pre_destroy() does enough synchronization.
>
> After this patch, it seems css_get() is still called in try_charge().
> But the logic is.
>
> 1. task_lock(mm->owner)
There is no task_lock() in this version :)

(snip)
> @@ -4219,7 +4252,6 @@ static int mem_cgroup_do_precharge(unsig
> mc.precharge += count;
> VM_BUG_ON(test_bit(CSS_ROOT, &mem->css.flags));
> WARN_ON_ONCE(count > INT_MAX);
> - __css_get(&mem->css, (int)count);
> return ret;
> }
> one_by_one:
You can remove VM_BUG_ON() and WARN_ON_ONCE() here, too.

> @@ -4469,8 +4501,6 @@ static void mem_cgroup_clear_mc(void)
> */
> res_counter_uncharge(&mc.to->res,
> PAGE_SIZE * mc.moved_swap);
> - VM_BUG_ON(test_bit(CSS_ROOT, &mc.to->css.flags));
> - __css_put(&mc.to->css, mc.moved_swap);
> }
> /* we've already done mem_cgroup_get(mc.to) */
>
>
And, you can remove "WARN_ON_ONCE(mc.moved_swap > INT_MAX)" at the beginning
of this block, too.


Thanks,
Daisuke Nishimura.


\
 
 \ /
  Last update: 2010-06-10 04:53    [W:1.006 / U:0.100 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site