Re: [PATCH 1/4] mm/migrate: correct return value of migrate_pages()

[Joonsoo Kim]

On Fri, Dec 06, 2013 at 01:37:26PM -0500, Naoya Horiguchi wrote:
> On Fri, Dec 06, 2013 at 03:42:16PM +0100, Vlastimil Babka wrote:
> > On 12/06/2013 09:41 AM, Joonsoo Kim wrote:
> > >migrate_pages() should return number of pages not migrated or error code.
> > >When unmap_and_move return -EAGAIN, outer loop is re-execution without
> > >initialising nr_failed. This makes nr_failed over-counted.
> > >
> > >So this patch correct it by initialising nr_failed in outer loop.
> > >
> > >Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@xxxxxxx>
> > >
> > >diff --git a/mm/migrate.c b/mm/migrate.c
> > >index 3747fcd..1f59ccc 100644
> > >--- a/mm/migrate.c
> > >+++ b/mm/migrate.c
> > >@@ -1102,6 +1102,7 @@ int migrate_pages(struct list_head *from, new_page_t get_new_page,
> > >
> > >  	for(pass = 0; pass < 10 && retry; pass++) {
> > >  		retry = 0;
> > >+		nr_failed = 0;
> > >
> > >  		list_for_each_entry_safe(page, page2, from, lru) {
> > >  			cond_resched();
> > >
> > 
> > If I'm reading the code correctly, unmap_and_move() (and
> > unmap_and_move_huge_page() as well) deletes all pages from the
> > 'from' list, unless it fails with -EAGAIN. So the only pages you see
> > in subsequent passes are those that failed with -EAGAIN and those
> > are not counted as nr_failed. So there shouldn't be over-count, but
> > your patch could result in under-count.
> > 
> > Perhaps a comment somewhere would clarify this.
> 
> I agree and suggest the one below.
> Joonsoo, feel free to append it to your series:)
> 
> Thanks,
> Naoya Horiguchi
> ---
> From: Naoya Horiguchi <n-horiguchi@xxxxxxxxxxxxx>
> Date: Fri, 6 Dec 2013 13:08:15 -0500
> Subject: [PATCH] migrate: add comment about permanent failure path
> 
> Let's add a comment about where the failed page goes to, which makes
> code more readable.
> 
> Signed-off-by: Naoya Horiguchi <n-horiguchi@xxxxxxxxxxxxx>
> ---
>  mm/migrate.c | 7 ++++++-
>  1 file changed, 6 insertions(+), 1 deletion(-)
> 
> diff --git a/mm/migrate.c b/mm/migrate.c
> index 661ff5f66591..c01caafa0a6f 100644
> --- a/mm/migrate.c
> +++ b/mm/migrate.c
> @@ -1118,7 +1118,12 @@ int migrate_pages(struct list_head *from, new_page_t get_new_page,
>  				nr_succeeded++;
>  				break;
>  			default:
> -				/* Permanent failure */
> +				/*
> +				 * Permanent failure (-EBUSY, -ENOSYS, etc.):
> +				 * unlike -EAGAIN case, the failed page is
> +				 * removed from migration page list and not
> +				 * retried in the next outer loop.
> +				 */
>  				nr_failed++;
>  				break;
>  			}
> -- 
> 1.8.3.1

Hello, Naoya.

When I saw this new comment, I found that unmap_and_move_huge_page()
has a bug. It would not remove the page from the list if
hugepage_migration_support() is false.
I will include your patch into my series and send an additional patch to fix
this problem.

Thanks.
--
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/