[PATCH v2] memcg: reduce lock time at move charge (Was Re: [PATCH 04/10] memcg: disable local interrupts in lock_page_cgroup()

[KAMEZAWA Hiroyuki]

On Tue, 12 Oct 2010 09:09:15 +0530
Balbir Singh <balbir at linux.vnet.ibm.com> wrote:

> * KAMEZAWA Hiroyuki <kamezawa.hiroyu at jp.fujitsu.com> [2010-10-08 19:41:31]:
> 
> > On Fri, 8 Oct 2010 14:12:01 +0900
> > KAMEZAWA Hiroyuki <kamezawa.hiroyu at jp.fujitsu.com> wrote:
> > > > Sure.  It walks the same data three times, potentially causing
> > > > thrashing in the L1 cache.
> > > 
> > > Hmm, make this 2 times, at least.
> > > 
> > How about this ?
> > ==
> > From: KAMEZAWA Hiroyuki <kamezawa.hiroyu at jp.fujitsu.com>
> > 
> > Presently, at task migration among cgroups, memory cgroup scans page tables and
> > moves accounting if flags are properly set.
> > 
> > 
> > The core code, mem_cgroup_move_charge_pte_range() does
> > 
> >  	pte_offset_map_lock();
> > 	for all ptes in a page table:
> > 		1. look into page table, find_and_get a page
> > 		2. remove it from LRU.
> > 		3. move charge.
> > 		4. putback to LRU. put_page()
> > 	pte_offset_map_unlock();
> > 
> > for pte entries on a 3rd level? page table.
> > 
> > As a planned updates, we'll support dirty-page accounting. Because move_charge()
> > is highly race, we need to add more check in move_charge.
> > For example, lock_page();-> wait_on_page_writeback();-> unlock_page();
> > is an candidate for new check.
> >
> 
> 
> Is this a change to help dirty limits or is it a generic bug fix.
>  
Not a bug fix. This for adding lock_page() to moge_charge(). It helps us
to remove "irq disable" in update_stat().


Thanks,
-Kame