On Mon, Feb 08, 2010 at 05:14:59PM +0000, Mel Gorman wrote:
> On Mon, Feb 08, 2010 at 03:24:22PM +1100, David Gibson wrote:
> > On Thu, Feb 04, 2010 at 09:02:18PM +0000, Eric B. Munson wrote:
> > > elflink.c tests slice boundaries for all architectures, but this is
> > > not necessary on x86[_64] as there are no rules for placement of
> > > huge pages with respect to small pages. This patch addes a define
> > > that will skip this check on x86[_64] builds.
> >
> > Hrm. The x86 versions of hugetlb_slice_end() and so forth are already
> > supposed to make this check a nop. We shouldn't need to ifdef this
> > out.
> >
>
> I think the problem may be with the definition of ALIGN_UP
>
> ALIGN_UP(x,a) (((x) + (a)) & ~((a) - 1))
>
> If the address is already aligned, does this definition not mean that
> ALIGN_UP will return a value one hugepage higher than it should? Prehaps
> the real definition should have been
>
> ALIGN_UP(x,a) (((x) + (a) - 1) & ~((a) - 1))
Yes, yes it should. That's embarrassing we have ALIGN() correct, but
ALIGN_UP() is wrong.
> Eric, could you check what the old and rounded values are and see are
> the alignment macros screwing things up? If so and the problem is with
> ALIGN_UP, try this version.
Indeed. I certainly think just removing the check completely for x86
is the wrong fix - those slice functions should work correctly (if
trivially) for x86 too.
--
David Gibson | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_
| _way_ _around_!
http://www.ozlabs.org/~dgibson
------------------------------------------------------------------------------
The Planet: dedicated and managed hosting, cloud storage, colocation
Stay online with enterprise data centers and the best network in the business
Choose flexible plans and management services without long-term contracts
Personal 24x7 support from experience hosting pros just a phone call away.
http://p.sf.net/sfu/theplanet-com
_______________________________________________
Libhugetlbfs-devel mailing list
Libhugetlbfs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/libhugetlbfs-devel
