Re: [PATCH] memcg: remove trylock_page_cgroup

[KAMEZAWA Hiroyuki]

On Tue, 21 Apr 2009 20:41:04 -0700
Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:

> On Wed, 22 Apr 2009 12:16:41 +0900 KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx> wrote:
> 
> > How about this ? worth to be tested, I think.
> > -Kame
> > ==
> > From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx>
> > 
> > Before synchronized-LRU patch, mem cgroup had its own LRU lock.
> > And there was a code which does
> > # assume mz as per zone struct of memcg. 
> > 
> >    spin_lock mz->lru_lock
> > 	lock_page_cgroup(pc).
> >    and
> >    lock_page_cgroup(pc)
> > 	spin_lock mz->lru_lock
> > 
> > because we cannot locate "mz" until we see pc->page_cgroup, we used
> > trylock(). But now, we don't have mz->lru_lock. All cgroup
> > uses zone->lru_lock for handling list. Moreover, manipulation of
> > LRU depends on global LRU now and we can isolate page from LRU by
> > very generic way.(isolate_lru_page()).
> > So, this kind of trylock is not necessary now.
> > 
> > I thought I removed all trylock in synchronized-LRU patch but there
> > is still one. This patch removes trylock used in memcontrol.c and
> > its definition. If someone needs, he should add this again with enough
> > reason.
> > 
> > Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx>
> > ---
> >  include/linux/page_cgroup.h |    5 -----
> >  mm/memcontrol.c             |    3 +--
> >  2 files changed, 1 insertion(+), 7 deletions(-)
> > 
> > Index: mmotm-2.6.30-Apr21/include/linux/page_cgroup.h
> > ===================================================================
> > --- mmotm-2.6.30-Apr21.orig/include/linux/page_cgroup.h
> > +++ mmotm-2.6.30-Apr21/include/linux/page_cgroup.h
> > @@ -61,11 +61,6 @@ static inline void lock_page_cgroup(stru
> >  	bit_spin_lock(PCG_LOCK, &pc->flags);
> >  }
> >  
> > -static inline int trylock_page_cgroup(struct page_cgroup *pc)
> > -{
> > -	return bit_spin_trylock(PCG_LOCK, &pc->flags);
> > -}
> > -
> >  static inline void unlock_page_cgroup(struct page_cgroup *pc)
> >  {
> >  	bit_spin_unlock(PCG_LOCK, &pc->flags);
> > Index: mmotm-2.6.30-Apr21/mm/memcontrol.c
> > ===================================================================
> > --- mmotm-2.6.30-Apr21.orig/mm/memcontrol.c
> > +++ mmotm-2.6.30-Apr21/mm/memcontrol.c
> > @@ -1148,8 +1148,7 @@ static int mem_cgroup_move_account(struc
> >  	from_mz =  mem_cgroup_zoneinfo(from, nid, zid);
> >  	to_mz =  mem_cgroup_zoneinfo(to, nid, zid);
> >  
> > -	if (!trylock_page_cgroup(pc))
> > -		return ret;
> > +	lock_page_cgroup(pc);
> >  
> >  	if (!PageCgroupUsed(pc))
> >  		goto out;
> 
> But we can't remove that nasty `while (loop--)' thing?
> 
every call which use isolate_lru_page() should handle isolatation failure.
But its ok to remove force_empty_list()'s loop-- becasue we do retry
in force_empty()
    force_empty()                   # does retry.
      -> force_empty_list()         # does retry.

> I expect that it will reliably fail if the caller is running as
> SCHED_FIFO and the machine is single-CPU, or if we're trying to yield
> to a SCHED_OTHER task which is pinned to this CPU, etc.  The cond_resched()
> won't work.
> 
Hm, signal_pending() is supported now (so special user scan use alaram())
I used yield() before cond_resched() but I was told don't use it.
Should I replace cond_resched() with congestion_wait(HZ/10) or some ?

But I'd like to do that in other patch than this patch bacause it
chages force_empty()'s logic.

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/