Re: [BUGFIX] memcg: fix res_counter_read_u64 lock aware (Was Re:[PATCH] oom: handle overflow in mem_cgroup_out_of_memory()

[KAMEZAWA Hiroyuki]

On Wed, 26 Jan 2011 17:08:24 -0800
Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:

> On Thu, 27 Jan 2011 09:53:42 +0900
> KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx> wrote:
> 
> > res_counter_read_u64 reads u64 value without lock. It's dangerous
> > in 32bit environment. This patch adds lock.
> > 
> > Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx>
> > ---
> >  include/linux/res_counter.h |   13 ++++++++++++-
> >  kernel/res_counter.c        |    2 +-
> >  2 files changed, 13 insertions(+), 2 deletions(-)
> > 
> > Index: mmotm-0125/include/linux/res_counter.h
> > ===================================================================
> > --- mmotm-0125.orig/include/linux/res_counter.h
> > +++ mmotm-0125/include/linux/res_counter.h
> > @@ -68,7 +68,18 @@ struct res_counter {
> >   * @pos:     and the offset.
> >   */
> >  
> > -u64 res_counter_read_u64(struct res_counter *counter, int member);
> > +u64 res_counter_read_u64_locked(struct res_counter *counter, int member);
> > +
> > +static inline u64 res_counter_read_u64(struct res_counter *counter, int member)
> > +{
> > +	unsigned long flags;
> > +	u64 ret;
> > +
> > +	spin_lock_irqsave(&counter->lock, flags);
> > +	ret = res_counter_read_u64_locked(counter, member);
> > +	spin_unlock_irqrestore(&counter->lock, flags);
> > +	return ret;
> > +}
> >  
> >  ssize_t res_counter_read(struct res_counter *counter, int member,
> >  		const char __user *buf, size_t nbytes, loff_t *pos,
> > Index: mmotm-0125/kernel/res_counter.c
> > ===================================================================
> > --- mmotm-0125.orig/kernel/res_counter.c
> > +++ mmotm-0125/kernel/res_counter.c
> > @@ -126,7 +126,7 @@ ssize_t res_counter_read(struct res_coun
> >  			pos, buf, s - buf);
> >  }
> >  
> > -u64 res_counter_read_u64(struct res_counter *counter, int member)
> > +u64 res_counter_read_u64_locked(struct res_counter *counter, int member)
> >  {
> >  	return *res_counter_member(counter, member);
> >  }
> 
> We don't need the lock on 64-bit platforms!
> 
> And there's zero benefit to inlining the spin_lock/unlock(), given that
> the function will always be making a function call anyway.
> 
> See i_size_read() for inspiration.
> 

seq_counter can't be used for res_counter because I don't want to update
seq_counter at changing ->usae of res_counter. So, I'd like to just add
spinlock for this time.

I wonder making memcg's counter to use 32bit and record usage in the number
of pages may be a simple way...

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/