lkml.org 
[lkml]   [2014]   [May]   [23]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [patch 8/9] mm: memcontrol: rewrite charge API
On Fri 23-05-14 16:54:13, Michal Hocko wrote:
> On Wed 30-04-14 16:25:42, Johannes Weiner wrote:
> > The memcg charge API charges pages before they are rmapped - i.e. have
> > an actual "type" - and so every callsite needs its own set of charge
> > and uncharge functions to know what type is being operated on.
> >
> > Rewrite the charge API to provide a generic set of try_charge(),
> > commit_charge() and cancel_charge() transaction operations, much like
> > what's currently done for swap-in:
> >
> > mem_cgroup_try_charge() attempts to reserve a charge, reclaiming
> > pages from the memcg if necessary.
> >
> > mem_cgroup_commit_charge() commits the page to the charge once it
> > has a valid page->mapping and PageAnon() reliably tells the type.
> >
> > mem_cgroup_cancel_charge() aborts the transaction.
> >
> > As pages need to be committed after rmap is established but before
> > they are added to the LRU, page_add_new_anon_rmap() must stop doing
> > LRU additions again. Factor lru_cache_add_active_or_unevictable().
> >
> > The order of functions in mm/memcontrol.c is entirely random, so this
> > new charge interface is implemented at the end of the file, where all
> > new or cleaned up, and documented code should go from now on.
>
> I would prefer moving them after refactoring because the reviewing is
> really harder this way. If such moving is needed at all.
>
> Anyway this is definitely not a Friday material...
>
> So only a first impression from a quick glance.
>
> size is saying the code is slightly bigger:
> text data bss dec hex filename
> 487977 84898 45984 618859 9716b mm/built-in.o.7
> 488276 84898 45984 619158 97296 mm/built-in.o.8
>
> No biggie though.
>
> It is true it get's rid of ~80LOC in memcontrol.c but it adds some more
> outside of memcg. Most of the charging paths didn't get any easier, they
> already know the type and they have to make sure they even commit the
> charge now.
>
> But maybe it is just me feeling that now that we have
> mem_cgroup_charge_{anon,file,swapin} the API doesn't look so insane
> anymore and so I am not tempted to change it that much.
>
> I will look at this with a Monday and fresh brain again.

And now that I got to 9/9 it is obvious it helps a lot to clean up the
uncharge path. But I am not in a mental state to dive into this today.
--
Michal Hocko
SUSE Labs


\
 
 \ /
  Last update: 2014-05-23 18:21    [W:0.074 / U:1.660 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site