[Devel] Re: [PATCH -mmotm 3/4] memcg: dirty pages accounting and limiting infrastructure

[Daisuke Nishimura]

On Mon, 8 Mar 2010 10:56:41 +0900, KAMEZAWA Hiroyuki <kamezawa.hiroyu at jp.fujitsu.com> wrote:
> On Mon, 8 Mar 2010 10:44:47 +0900
> Daisuke Nishimura <nishimura at mxp.nes.nec.co.jp> wrote:
> 
> > > +/*
> > > + * mem_cgroup_update_page_stat_locked() - update memcg file cache's accounting
> > > + * @page:	the page involved in a file cache operation.
> > > + * @idx:	the particular file cache statistic.
> > > + * @charge:	true to increment, false to decrement the statistic specified
> > > + *		by @idx.
> > > + *
> > > + * Update memory cgroup file cache's accounting from a locked context.
> > > + *
> > > + * NOTE: must be called with mapping->tree_lock held.
> > > + */
> > > +void mem_cgroup_update_page_stat_locked(struct page *page,
> > > +			enum mem_cgroup_write_page_stat_item idx, bool charge)
> > > +{
> > > +	struct address_space *mapping = page_mapping(page);
> > > +	struct page_cgroup *pc;
> > > +
> > > +	if (mem_cgroup_disabled())
> > > +		return;
> > > +	WARN_ON_ONCE(!irqs_disabled());
> > > +	WARN_ON_ONCE(mapping && !spin_is_locked(&mapping->tree_lock));
> > > +
> > I think this is a wrong place to insert assertion.
> > The problem about page cgroup lock is that it can be interrupted in current implementation.
> > So,
> > 
> > a) it must not be aquired under another lock which can be aquired in interrupt context,
> >    such as mapping->tree_lock, to avoid:
> > 
> > 		context1			context2
> > 					lock_page_cgroup(pcA)
> > 	spin_lock_irq(&tree_lock)
> > 		lock_page_cgroup(pcA)		<interrupted>
> > 		=>fail				spin_lock_irqsave(&tree_lock)
> > 						=>fail
> > 
> > b) it must not be aquired in interrupt context to avoid:
> > 
> > 	lock_page_cgroup(pcA)
> > 		<interrupted>
> > 		lock_page_cgroup(pcA)
> > 		=>fail
> > 
> > I think something like this would be better:
> > 
> > @@ -83,8 +83,14 @@ static inline enum zone_type page_cgroup_zid(struct page_cgroup *pc)
> >         return page_zonenum(pc->page);
> >  }
> > 
> > +#include <linux/irqflags.h>
> > +#include <linux/hardirq.h>
> >  static inline void lock_page_cgroup(struct page_cgroup *pc)
> >  {
> > +#ifdef CONFIG_DEBUG_VM
> > +       WARN_ON_ONCE(irqs_disabled());
> > +       WARN_ON_ONCE(in_interrupt());
> > +#endif
> >         bit_spin_lock(PCG_LOCK, &pc->flags);
> >  }
> > 
> > > +	pc = lookup_page_cgroup(page);
> > > +	if (unlikely(!pc) || !PageCgroupUsed(pc))
> > > +		return;
> > > +	mem_cgroup_update_page_stat(pc, idx, charge);
> > > +}
> > > +EXPORT_SYMBOL_GPL(mem_cgroup_update_page_stat_locked);
> > > +
> > > +/*
> > > + * mem_cgroup_update_page_stat_unlocked() - update memcg file cache's accounting
> > > + * @page:	the page involved in a file cache operation.
> > > + * @idx:	the particular file cache statistic.
> > > + * @charge:	true to increment, false to decrement the statistic specified
> > > + *		by @idx.
> > > + *
> > > + * Update memory cgroup file cache's accounting from an unlocked context.
> > > + */
> > > +void mem_cgroup_update_page_stat_unlocked(struct page *page,
> > > +			enum mem_cgroup_write_page_stat_item idx, bool charge)
> > > +{
> > > +	struct page_cgroup *pc;
> > > +
> > > +	if (mem_cgroup_disabled())
> > > +		return;
> > > +	pc = lookup_page_cgroup(page);
> > > +	if (unlikely(!pc) || !PageCgroupUsed(pc))
> > > +		return;
> > > +	lock_page_cgroup(pc);
> > > +	mem_cgroup_update_page_stat(pc, idx, charge);
> > >  	unlock_page_cgroup(pc);
> > >  }
> > > +EXPORT_SYMBOL_GPL(mem_cgroup_update_page_stat_unlocked);
> > >  
> > IIUC, test_clear_page_writeback(at least) can be called under interrupt context.
> > This means lock_page_cgroup() is called under interrupt context, that is,
> > the case b) above can happen.
> > hmm... I don't have any good idea for now except disabling irq around page cgroup lock
> > to avoid all of these mess things.
> > 
> 
> Hmm...simply IRQ-off for all updates ?
I think so in current code.
But after these changes, we must use local_irq_save()/restore()
instead of local_irq_disable()/enable() in mem_cgroup_update_page_stat().

> But IIRC, clear_writeback is done under treelock.... No ?
> 
The place where NR_WRITEBACK is updated is out of tree_lock.

   1311 int test_clear_page_writeback(struct page *page)
   1312 {
   1313         struct address_space *mapping = page_mapping(page);
   1314         int ret;
   1315
   1316         if (mapping) {
   1317                 struct backing_dev_info *bdi = mapping->backing_dev_info;
   1318                 unsigned long flags;
   1319
   1320                 spin_lock_irqsave(&mapping->tree_lock, flags);
   1321                 ret = TestClearPageWriteback(page);
   1322                 if (ret) {
   1323                         radix_tree_tag_clear(&mapping->page_tree,
   1324                                                 page_index(page),
   1325                                                 PAGECACHE_TAG_WRITEBACK);
   1326                         if (bdi_cap_account_writeback(bdi)) {
   1327                                 __dec_bdi_stat(bdi, BDI_WRITEBACK);
   1328                                 __bdi_writeout_inc(bdi);
   1329                         }
   1330                 }
   1331                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
   1332         } else {
   1333                 ret = TestClearPageWriteback(page);
   1334         }
   1335         if (ret)
   1336                 dec_zone_page_state(page, NR_WRITEBACK);
   1337         return ret;
   1338 }
   1339


Thanks,
Daisuke Nishimura.
_______________________________________________
Containers mailing list
Containers at lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers