Re: [PATCH -V8 14/16] hugetlb/cgroup: add charge/uncharge calls forHugeTLB alloc/free

[Johannes Weiner]

On Sat, Jun 09, 2012 at 06:39:06PM +0530, Aneesh Kumar K.V wrote:
> Johannes Weiner <hannes@xxxxxxxxxxx> writes:
> 
> > On Sat, Jun 09, 2012 at 02:29:59PM +0530, Aneesh Kumar K.V wrote:
> >> From: "Aneesh Kumar K.V" <aneesh.kumar@xxxxxxxxxxxxxxxxxx>
> >> 
> >> This adds necessary charge/uncharge calls in the HugeTLB code.  We do
> >> hugetlb cgroup charge in page alloc and uncharge in compound page destructor.
> >> 
> >> Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@xxxxxxxxxxxxxxxxxx>
> >> ---
> >>  mm/hugetlb.c        |   16 +++++++++++++++-
> >>  mm/hugetlb_cgroup.c |    7 +------
> >>  2 files changed, 16 insertions(+), 7 deletions(-)
> >> 
> >> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> >> index bf79131..4ca92a9 100644
> >> --- a/mm/hugetlb.c
> >> +++ b/mm/hugetlb.c
> >> @@ -628,6 +628,8 @@ static void free_huge_page(struct page *page)
> >>  	BUG_ON(page_mapcount(page));
> >>  
> >>  	spin_lock(&hugetlb_lock);
> >> +	hugetlb_cgroup_uncharge_page(hstate_index(h),
> >> +				     pages_per_huge_page(h), page);
> >
> > hugetlb_cgroup_uncharge_page() takes the hugetlb_lock, no?
> 
> Yes, But this patch also modifies it to not take the lock, because we
> hold spin_lock just below in the call site. I didn't want to drop the
> lock and take it again.

Sorry, I missed that.

> > It's quite hard to review code that is split up like this.  Please
> > always keep the introduction of new functions in the same patch that
> > adds the callsite(s).
> 
> One of the reason I split the charge/uncharge routines and the callers
> in separate patches is to make it easier for review. Irrespective of
> the call site charge/uncharge routines should be correct with respect
> to locking and other details. What I did in this patch is a small
> optimization of avoiding dropping and taking the lock again. May be the
> right approach would have been to name it __hugetlb_cgroup_uncharge_page
> and make sure the hugetlb_cgroup_uncharge_page still takes spin_lock.
> But then we don't have any callers for that.

I think this makes it needlessly complicated and there is no correct
or incorrect locking in (initially) dead code :-)

The callsites are just a few lines.  It's harder to review if you
introduce an API and then change it again mid-patchset.

If there are no callers for a function that grabs the lock itself,
don't add it.  Just add a note to the kerneldoc that explains the
requirement or put VM_BUG_ON(!spin_is_locked(&hugetlb_lock)); in
there or so.
--
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/