Re: [PATCH v4 10/10] thp: implement refcounting for huge zero page

[Kirill A. Shutemov]

On Fri, Oct 19, 2012 at 02:59:41AM +0300, Kirill A. Shutemov wrote:
> On Thu, Oct 18, 2012 at 04:45:02PM -0700, Andrew Morton wrote:
> > On Mon, 15 Oct 2012 09:00:59 +0300
> > "Kirill A. Shutemov" <kirill.shutemov@xxxxxxxxxxxxxxx> wrote:
> > 
> > > H. Peter Anvin doesn't like huge zero page which sticks in memory forever
> > > after the first allocation. Here's implementation of lockless refcounting
> > > for huge zero page.
> > > 
> > > We have two basic primitives: {get,put}_huge_zero_page(). They
> > > manipulate reference counter.
> > > 
> > > If counter is 0, get_huge_zero_page() allocates a new huge page and
> > > takes two references: one for caller and one for shrinker. We free the
> > > page only in shrinker callback if counter is 1 (only shrinker has the
> > > reference).
> > > 
> > > put_huge_zero_page() only decrements counter. Counter is never zero
> > > in put_huge_zero_page() since shrinker holds on reference.
> > > 
> > > Freeing huge zero page in shrinker callback helps to avoid frequent
> > > allocate-free.
> > 
> > I'd like more details on this please.  The cost of freeing then
> > reinstantiating that page is tremendous, because it has to be zeroed
> > out again.  If there is any way at all in which the kernel can be made
> > to enter a high-frequency free/reinstantiate pattern then I expect the
> > effects would be quite bad.
> > 
> > Do we have sufficient mechanisms in there to prevent this from
> > happening in all cases?  If so, what are they, because I'm not seeing
> > them?
> 
> We only free huge zero page in shrinker callback if nobody in the system
> uses it. Never on put_huge_zero_page(). Shrinker runs only under memory
> pressure or if user asks (drop_caches).
> Do you think we need an additional protection mechanism?

Andrew?

> > > Refcounting has cost. On 4 socket machine I observe ~1% slowdown on
> > > parallel (40 processes) read page faulting comparing to lazy huge page
> > > allocation.  I think it's pretty reasonable for synthetic benchmark.
> 
> -- 
>  Kirill A. Shutemov

-- 
 Kirill A. Shutemov
--
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/