Re: [patch 4/8] mm: memcg: lookup_page_cgroup (almost) never returnsNULL

[Johannes Weiner]

On Thu, Nov 24, 2011 at 11:26:06AM +0100, Michal Hocko wrote:
> On Thu 24-11-11 11:05:49, Johannes Weiner wrote:
> > On Thu, Nov 24, 2011 at 10:52:51AM +0100, Michal Hocko wrote:
> > > On Wed 23-11-11 16:42:27, Johannes Weiner wrote:
> > > > From: Johannes Weiner <jweiner@xxxxxxxxxx>
> > > > 
> > > > Pages have their corresponding page_cgroup descriptors set up before
> > > > they are used in userspace, and thus managed by a memory cgroup.
> > > > 
> > > > The only time where lookup_page_cgroup() can return NULL is in the
> > > > page sanity checking code that executes while feeding pages into the
> > > > page allocator for the first time.
> > > > 
> > > > Remove the NULL checks against lookup_page_cgroup() results from all
> > > > callsites where we know that corresponding page_cgroup descriptors
> > > > must be allocated.
> > > 
> > > OK, shouldn't we add
> > > 
> > > diff --git a/mm/page_cgroup.c b/mm/page_cgroup.c
> > > index 2d123f9..cb93f64 100644
> > > --- a/mm/page_cgroup.c
> > > +++ b/mm/page_cgroup.c
> > > @@ -35,8 +35,7 @@ struct page_cgroup *lookup_page_cgroup(struct page *page)
> > >  	struct page_cgroup *base;
> > >  
> > >  	base = NODE_DATA(page_to_nid(page))->node_page_cgroup;
> > > -	if (unlikely(!base))
> > > -		return NULL;
> > > +	BUG_ON(!base);
> > >  
> > >  	offset = pfn - NODE_DATA(page_to_nid(page))->node_start_pfn;
> > >  	return base + offset;
> > > @@ -112,8 +111,7 @@ struct page_cgroup *lookup_page_cgroup(struct page *page)
> > >  	unsigned long pfn = page_to_pfn(page);
> > >  	struct mem_section *section = __pfn_to_section(pfn);
> > >  
> > > -	if (!section->page_cgroup)
> > > -		return NULL;
> > > +	BUG_ON(!section->page_cgroup);
> > >  	return section->page_cgroup + pfn;
> > >  }
> > >  
> > > just to make it explicit?
> > 
> > No, see the last hunk in this patch.  It's actually possible for this
> > to run, although only while feeding fresh pages into the allocator:
> 
> Bahh. Yes, I have noticed the hunk but then I started thinking about
> how to make the NULL case explicit and totally forgot about that.
> Sorry about the noise.
> 
> > 
> > > > @@ -3326,6 +3321,7 @@ static struct page_cgroup *lookup_page_cgroup_used(struct page *page)
> > > >  	struct page_cgroup *pc;
> > > >  
> > > >  	pc = lookup_page_cgroup(page);
> > > > +	/* Can be NULL while bootstrapping the page allocator */
> > > >  	if (likely(pc) && PageCgroupUsed(pc))
> > > >  		return pc;
> > > >  	return NULL;
> > 
> > We could add a lookup_page_cgroup_safe() for this DEBUG_VM-only
> > callsite as an optimization separately and remove the NULL check from
> > lookup_page_cgroup() itself.  But this patch was purely about removing
> > the actively misleading checks.
> 
> Yes, but I am not sure whether code duplication is worth it. Let's just
> stick with current form. Maybe just move the comment when it can be NULL
> to the lookup_page_cgroup directly?

Don't underestimate it, this function is used quite heavily while the
case of the array being NULL is a minor fraction of all calls.  But
it's for another patch, anyway.

The case for when lookup_page_cgroup() returns NULL is kinda obvious
to me when directly looking at the function itself, because the arrays
are allocated just a few lines below.  But care to send a patch?
--
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/