vmscan.c: Reclaim unevictable pages.

[Vlastimil Babka]

On 4/6/19 7:59 AM, Pankaj Suryawanshi wrote:
> Hello ,
> 
> shrink_page_list() returns , number of pages reclaimed, when pages is
> unevictable it returns VM_BUG_ON_PAGE(PageLRU(page) ||
> PageUnevicatble(page),page);
> 
> We can add the unevictable pages in reclaim list in
> shrink_page_list(), return total number of reclaim pages including
> unevictable pages, let the caller handle unevictable pages.
> 
> I think the problem is shrink_page_list is awkard. If page is
> unevictable it goto activate_locked->keep_locked->keep lables, keep
> lable list_add the unevictable pages and throw the VM_BUG instead of
> passing it to caller while it relies on caller for
> non-reclaimed-non-unevictable  page's putback.
> I think we can make it consistent so that shrink_page_list could
> return non-reclaimed pages via page_list and caller can handle it. As
> an advance, it could try to migrate mlocked pages without retrial.
> 
> 
> Below is the issue i observed of CMA_ALLOC of large size buffer :
> (Kernel version - 4.14.65 With Android Pie.
> 
> [   24.718792] page dumped because: VM_BUG_ON_PAGE(PageLRU(page) ||
> PageUnevictable(page))
> [   24.726949] page->mem_cgroup:bd008c00
> [   24.730693] ------------[ cut here ]------------
> [   24.735304] kernel BUG at mm/vmscan.c:1350!
> [   24.739478] Internal error: Oops - BUG: 0 [#1] PREEMPT SMP ARM

Please include full report including the full stacktrace, kernel version
etc etc.

> 
> 
> Below is the patch which solved this issue :
> 
> diff --git a/mm/vmscan.c b/mm/vmscan.c
> index be56e2e..12ac353 100644
> --- a/mm/vmscan.c
> +++ b/mm/vmscan.c
> @@ -998,7 +998,7 @@ static unsigned long shrink_page_list(struct
> list_head *page_list,
>                 sc->nr_scanned++;
> 
>                 if (unlikely(!page_evictable(page)))
> -                       goto activate_locked;
> +                      goto cull_mlocked;
> 
>                 if (!sc->may_unmap && page_mapped(page))
>                         goto keep_locked;
> @@ -1331,7 +1331,12 @@ static unsigned long shrink_page_list(struct
> list_head *page_list,
>                 } else
>                         list_add(&page->lru, &free_pages);
>                 continue;
> -
> +cull_mlocked:
> +                if (PageSwapCache(page))
> +                        try_to_free_swap(page);
> +                unlock_page(page);
> +                list_add(&page->lru, &ret_pages);
> +                continue;
>  activate_locked:
>                 /* Not a candidate for swapping, so reclaim swap space. */
>                 if (PageSwapCache(page) && (mem_cgroup_swap_full(page) ||
> 
> 
> 
> 
> It fixes the below issue.
> 
> 1. Large size buffer allocation using cma_alloc successful with
> unevictable pages.
> 
> cma_alloc of current kernel will fail due to unevictable page
> 
> Please let me know if anything i am missing.
> 
> Regards,
> Pankaj
>