Re: [PATCH] mm: fix migration hangs on anon_vma lock

From: Mel Gorman
Date: Tue Jan 11 2011 - 06:45:30 EST


On Mon, Jan 10, 2011 at 11:08:04PM -0800, Hugh Dickins wrote:
> Increased usage of page migration in mmotm reveals that the anon_vma
> locking in unmap_and_move() has been deficient since 2.6.36 (or even
> earlier).

Hmm, a certain amount of the compaction work was spent fixing migration
bugs. I wonder if there are mysterious bug reports out there related to the
use of move_pages() that are only getting fixed now.

> Review at the time of f18194275c39835cb84563500995e0d503a32d9a
> "mm: fix hang on anon_vma->root->lock" missed the issue here: the anon_vma
> to which we get a reference may already have been freed back to its slab
> (it is in use when we check page_mapped, but that can change), and so its
> anon_vma->root may be switched at any moment by reuse in anon_vma_prepare.
>
> Perhaps we could fix that with a get_anon_vma_unless_zero(), but let's not:
> just rely on page_lock_anon_vma() to do all the hard thinking for us, then
> we don't need any rcu read locking over here.
>
> In removing the rcu_unlock label: since PageAnon is a bit in page->mapping,
> it's impossible for a !page->mapping page to be anon; but insert VM_BUG_ON
> in case the implementation ever changes.
>
> Signed-off-by: Hugh Dickins <hughd@xxxxxxxxxx>
> Cc: stable@xxxxxxxxxx [2.6.37, 2.6.36]

Reasoning and patch look correct. Light testing did not show up any
obvious problems.

Reviewed-by: Mel Gorman <mel@xxxxxxxxx>

--
Mel Gorman
Part-time Phd Student Linux Technology Center
University of Limerick IBM Dublin Software Lab
--
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/