Re: [Lhms-devel] [PATCH 4/7] Direct Migration V5: migrate_pages()extension

[Kamezawa Hiroyuki]

Christoph Lameter wrote:
> The current page migration functions in mempolicy.c do not migrate shmem 
> vmas to be safe. In the future we surely would like to support migration 
> of shmem. I'd be glad if you could make sure that this works.
Okay, shmem is not problem now.

> > Problem is:
> > 1. a page of shmem(tmpfs)'s generic file is in page-cache. assume page is
> > diry.
> > 2. When it passed to migrate_page(), it reaches pageout() in the middle of
> > migrate_page().
> > 3. pageout calls shmem_writepage(), and the page turns to be swap-cache page.
> >   At this point, page->mapping becomes NULL (see move_to_swapcache())
>
>
> A swapcache page would have page->mapping pointing to swapper space. 
> move_to_swap_cache does not set page->mapping == NULL.
int move_to_swap_cache(struct page *page, swp_entry_t entry)
{
        int err = __add_to_swap_cache(page, entry, GFP_ATOMIC);
        if (!err) {
                remove_from_page_cache(page);<------------------------this
                page_cache_release(page);       /* pagecache ref */
                if (!swap_duplicate(entry))
                        BUG();
                SetPageDirty(page);
                INC_CACHE_INFO(add_total);
        } else if (err == -EEXIST)
                INC_CACHE_INFO(exist_race);
        return err;
}

remove_from_page_cache(page) sets page->mapping == NULL.


> If page->mapping would be NULL then migrate_page() could not 
> have been called. The mapping is used to obtain the address of the 
> function to call,
What you say is here.
==
                /*
                 * Pages are properly locked and writeback is complete.
                 * Try to migrate the page.
                 */
                mapping = page_mapping(page);
                if (!mapping) <-------------------------------------------this check.
                        goto unlock_both;

                if (mapping->a_ops->migratepage) {
                        rc = mapping->a_ops->migratepage(newpage, page);
                        goto unlock_both;
                }
==
But, see page_mapping() .....
==
static inline struct address_space *page_mapping(struct page *page)
{
        struct address_space *mapping = page->mapping;

        if (unlikely(PageSwapCache(page)))
                mapping = &swapper_space;
        else if (unlikely((unsigned long)mapping & PAGE_MAPPING_ANON))
                mapping = NULL;
        return mapping;
}
==

Even if page->mapping == NULL, page_mapping() can return &swapper_space if PageSwapCache()
is true. (Note: a shmem page here is not page-cache, not anon, but swap-cache)

I'm now considering to add a_ops->migrate_page() to shmem is sane way...
But migration doesn't manage shmem, so this is just memory hot-remove's problem.


-- Kame




-
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/