]> git.kernelconcepts.de Git - karo-tx-linux.git/commit
mm: thp: fix SMP race condition between THP page fault and MADV_DONTNEED
authorAndrea Arcangeli <aarcange@redhat.com>
Fri, 26 Feb 2016 23:19:28 +0000 (15:19 -0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 3 Mar 2016 23:07:10 +0000 (15:07 -0800)
commit915d02457e74344bcd99fe64b159de2f6074b2c6
tree4b14f1a5fb21fb4a8fb28b6ed9d94baeff69c7cb
parentc88edc09363243dd84d6e8ac7a36773213b26ac8
mm: thp: fix SMP race condition between THP page fault and MADV_DONTNEED

commit ad33bb04b2a6cee6c1f99fabb15cddbf93ff0433 upstream.

pmd_trans_unstable()/pmd_none_or_trans_huge_or_clear_bad() were
introduced to locklessy (but atomically) detect when a pmd is a regular
(stable) pmd or when the pmd is unstable and can infinitely transition
from pmd_none() and pmd_trans_huge() from under us, while only holding
the mmap_sem for reading (for writing not).

While holding the mmap_sem only for reading, MADV_DONTNEED can run from
under us and so before we can assume the pmd to be a regular stable pmd
we need to compare it against pmd_none() and pmd_trans_huge() in an
atomic way, with pmd_trans_unstable().  The old pmd_trans_huge() left a
tiny window for a race.

Useful applications are unlikely to notice the difference as doing
MADV_DONTNEED concurrently with a page fault would lead to undefined
behavior.

[akpm@linux-foundation.org: tidy up comment grammar/layout]
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Reported-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
mm/memory.c