[PATCH -mm] mm: hugetlbfs: Correctly populate shared pmd

[PATCH -mm] mm: hugetlbfs: Correctly populate shared pmd

[Michal Hocko]

Hi Andrew,
the following patch fixes yet-another race in the hugetlb pte sharing
code reported by Larry. It is based on top of the current -mm tree but
it cleanly applies to linus tree as well. It should go to stable as
well. The bug is there for ages but this fix is possible only since 3.0
because i_mmap_lock used to be a spinlock until 3d48ae45 which turned it
into mutex and so we can call pmd_alloc.
There was another candidate for the same issue by Mel
(https://lkml.org/lkml/2012/7/31/275) but we considered this one to be
better because it is more focused on the arch specific code and it also
highers chances for sharing.
---

Re: [PATCH -mm] mm: hugetlbfs: Correctly populate shared pmd

[Rik van Riel]

On 08/02/2012 10:16 AM, Michal Hocko wrote:
> Hi Andrew,
> the following patch fixes yet-another race in the hugetlb pte sharing
> code reported by Larry. It is based on top of the current -mm tree but
> it cleanly applies to linus tree as well. It should go to stable as
> well. The bug is there for ages but this fix is possible only since 3.0
> because i_mmap_lock used to be a spinlock until 3d48ae45 which turned it
> into mutex and so we can call pmd_alloc.

> This patch addresses the issue by moving pmd_alloc into huge_pmd_share
> which guarantees that the shared pud is populated in the same
> critical section as pmd. This also means that huge_pte_offset test in
> huge_pmd_share is serialized correctly now which in turn means that
> the success of the sharing will be higher as the racing tasks see the
> pud and pmd populated together.
>
> Race identified and changelog written mostly by Mel Gorman
> Reported-and-tested-by: Larry Woodman <lwoodman@redhat.com>
> Reviewed-by: Mel Gorman <mgorman@suse.de>
> Signed-off-by: Michal Hocko <mhocko@suse.cz>

Reviewed-by: Rik van Riel <riel@redhat.com>


--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH -mm] mm: hugetlbfs: Correctly populate shared pmd

[Hillf Danton]

On Thu, Aug 2, 2012 at 10:16 PM, Michal Hocko <mhocko@suse.cz> wrote:
> This patch addresses the issue by moving pmd_alloc into huge_pmd_share
> which guarantees that the shared pud is populated in the same
> critical section as pmd.

Is i_mmap_mutex for guarding new pmd allocation?
Is regression introduced if sharing is unavailable?

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH -mm] mm: hugetlbfs: Correctly populate shared pmd

[Michal Hocko]

On Fri 03-08-12 22:16:52, Hillf Danton wrote:
> On Thu, Aug 2, 2012 at 10:16 PM, Michal Hocko <mhocko@suse.cz> wrote:
> > This patch addresses the issue by moving pmd_alloc into huge_pmd_share
> > which guarantees that the shared pud is populated in the same
> > critical section as pmd.
> 
> Is i_mmap_mutex for guarding new pmd allocation?

It doesn't guard the pmd allocation itself it just makes sure that pud
population and pmd_allocation are done atomicaly wrt. other processes to
share the same pmd because sharing is synchronized by i_mmap_mutex.

> Is regression introduced if sharing is unavailable?

No. The bug is about the sharing as the changelog describes.

-- 
Michal Hocko
SUSE Labs

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>