On Fri, 9 Oct 2020 09:30:59 -0400 Miaohe Lin <[email protected]> wrote:
> swap_ra_info() may leave ra_info untouched in non_swap_entry() case as page
> table lock is not held. In this case, we have ra_info.nr_pte == 0 and it is
> meaningless to continue with swap cache readahead. Skip such ops by init
> ra_info.win = 1.
>
>
> --- a/mm/swap_state.c
> +++ b/mm/swap_state.c
> @@ -839,7 +839,7 @@ static struct page *swap_vma_readahead(swp_entry_t
> fentry, gfp_t gfp_mask,
> swp_entry_t entry;
> unsigned int i;
> bool page_allocated;
> - struct vma_swap_readahead ra_info = {0,};
> + struct vma_swap_readahead ra_info = {1,};
>
> swap_ra_info(vmf, &ra_info);
> if (ra_info.win == 1)
Looks right. But I think it would be better to fix this in
swap_ra_info() - make it more consistent in its return state.
Otherwise, let's modernify that struct initializer:
---
a/mm/swap_state.c~mm-swap_state-skip-meaningless-swap-cache-readahead-when-ra_infowin-==-0-fix
+++ a/mm/swap_state.c
@@ -839,7 +839,9 @@ static struct page *swap_vma_readahead(s
swp_entry_t entry;
unsigned int i;
bool page_allocated;
- struct vma_swap_readahead ra_info = {1,};
+ struct vma_swap_readahead ra_info = {
+ .win = 1,
+ };
swap_ra_info(vmf, &ra_info);
if (ra_info.win == 1)
_
