Re: [PATCH 2/2] memcg: reduce usage at change limit

From: KAMEZAWA Hiroyuki
Date: Tue Jun 17 2008 - 05:55:30 EST


On Tue, 17 Jun 2008 13:06:56 +0900
KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx> wrote:

> On Tue, 17 Jun 2008 09:16:31 +0530
> Balbir Singh <balbir@xxxxxxxxxxxxxxxxxx> wrote:
>
> > KAMEZAWA Hiroyuki wrote:
> > > Reduce the usage of res_counter at the change of limit.
> > >
> > > Changelog v4 -> v5.
> > > - moved "feedback" alogrithm from res_counter to memcg.
> > >
> > > Background:
> > > - Now, mem->usage is not reduced at limit change. So, the users will see
> > > usage > limit case in memcg every time. This patch fixes it.
> > >
> > > Before:
> > > - no usage change at setting limit.
> > > - setting limit always returns 0 even if usage can never be less than zero.
> > > (This can happen when memory is locked or there is no swap.)
> > > - This is BUG, I think.
> > > After:
> > > - usage will be less than new limit at limit change.
> > > - set limit returns -EBUSY when the usage cannot be reduced.
> > >
> > >
> > > Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx>
> > >
> > > ---
> > > Documentation/controllers/memory.txt | 3 -
> > > mm/memcontrol.c | 68 ++++++++++++++++++++++++++++-------
> > > 2 files changed, 56 insertions(+), 15 deletions(-)
> > >
> > > Index: mm-2.6.26-rc5-mm3/mm/memcontrol.c
> > > ===================================================================
> > > --- mm-2.6.26-rc5-mm3.orig/mm/memcontrol.c
> > > +++ mm-2.6.26-rc5-mm3/mm/memcontrol.c
> > > @@ -852,18 +852,30 @@ out:
> > > css_put(&mem->css);
> > > return ret;
> > > }
> > > +/*
> > > + * try to set limit and reduce usage if necessary.
> > > + * returns 0 at success.
> > > + * returns -EBUSY if memory cannot be dropped.
> > > + */
> > >
> > > -static int mem_cgroup_write_strategy(char *buf, unsigned long long *tmp)
> > > +static inline int mem_cgroup_resize_limit(struct cgroup *cont,
> > > + unsigned long long val)
> > > {
> > > - *tmp = memparse(buf, &buf);
> > > - if (*buf != '\0')
> > > - return -EINVAL;
> > > + struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
> > > + int retry_count = 0;
> > > + int progress;
> > >
> > > - /*
> > > - * Round up the value to the closest page size
> > > - */
> > > - *tmp = ((*tmp + PAGE_SIZE - 1) >> PAGE_SHIFT) << PAGE_SHIFT;
> > > - return 0;
> > > +retry:
> > > + if (!res_counter_set_limit(&memcg->res, val))
> > > + return 0;
> > > + if (retry_count == MEM_CGROUP_RECLAIM_RETRIES)
> > > + return -EBUSY;
> > > +
> > > + cond_resched();
> >
> > Do we really need this? We do have cond_resched in shrink_page_list(),
> > shrink_active_list(), do we need it here as well?
> >
> I'd like to add this when adding a busy loop. But ok, will remove.
>
> > > + progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL);
> > > + if (!progress)
> > > + retry_count++;
> > > + goto retry;
> >
> > I don't like upward goto's. Can't we convert this to a nice do {} while or
> > while() loop?
> >
> Hmm, ok.
>
> I'll repost later, today.
>
I'll postpone this until -mm is settled ;)

Thanks,
-Kame

--
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/