lkml.org 
[lkml]   [2013]   [Dec]   [11]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v2 7/7] mm/migrate: remove result argument on page allocation function for migration
On Mon, Dec 09, 2013 at 04:40:06PM +0000, Christoph Lameter wrote:
> On Mon, 9 Dec 2013, Joonsoo Kim wrote:
>
> > First, we don't use error number in fail case. Call-path related to
> > new_page_node() is shown in the following.
> >
> > do_move_page_to_node_array() -> migrate_pages() -> unmap_and_move()
> > -> new_page_node()
> >
> > If unmap_and_move() failed, migrate_pages() also returns err, and then
> > do_move_page_to_node_array() skips to set page's status to user buffer.
> > So we don't need to set error number to each pages on failure case.
>
> I dont get this. new_page_node() sets the error condition in the
> page_to_node array before this patch. There is no post processing in
> do_move_page_to_node_array(). The function simply returns and relies on
> new_page_node() to have set the page status. do_move_pages() then returns
> the page status back to userspace. How does the change preserve these
> diagnostics?

Hello, Christoph.

In do_move_pages(), if error occurs, 'goto out_pm' is executed and the
page status doesn't back to userspace. So we don't need to store err number.

Perhaps my description should be changed something like below.

*do_move_pages()* -> do_move_page_to_node_array() -> migrate_pages()
-> unmap_and_move() -> new_page_node()

If unmap_and_move() failed, migrate_pages() also returns err, and then
*do_move_pages()* skips to set page's status to user buffer.
So we don't need to set error number to each pages on failure case.

Is it sufficient explanation?

>
> > Next, we don't need to set node id of the new page in unmap_and_move(),
> > since it cannot be different with pm->node. In new_page_node(), we always
> > try to allocate the page in exact node by referencing pm->node. So it is
> > sufficient to set node id of the new page in new_page_node(), instead of
> > unmap_and_move().
>
> Thats a good thought.

Thanks!


\
 
 \ /
  Last update: 2013-12-11 10:01    [W:0.063 / U:2.052 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site