Re: [PATCH v5 14/19] mm, swap: cleanup swap entry management workflow

[Chris Mason]

Kairui Song <ryncsn@xxxxxxxxx> wrote:
> From: Kairui Song <kasong@xxxxxxxxxxx>
> 
> The current swap entry allocation/freeing workflow has never had a clear
> definition. This makes it hard to debug or add new optimizations.
> 
> This commit introduces a proper definition of how swap entries would be
> allocated and freed. Now, most operations are folio based, so they will
> never exceed one swap cluster, and we now have a cleaner border between
> swap and the rest of mm, making it much easier to follow and debug,
> especially with new added sanity checks. Also making more optimization
> possible.
> 
> Swap entry will be mostly allocated and free with a folio bound.
> The folio lock will be useful for resolving many swap ralated races.
> 
> Now swap allocation (except hibernation) always starts with a folio in
> the swap cache, and gets duped/freed protected by the folio lock:

[ ... ]

Hi everyone,

Apologies for looping back around on this one, but I've got some prompt
changes that break down larger commits into sections and run them
independently through review.  I reran it on mm-new, and it found something
new:

> diff --git a/mm/swapfile.c b/mm/swapfile.c
> index 64970ee11fcf7..85bf4f7d9ae7b 100644
> --- a/mm/swapfile.c
> +++ b/mm/swapfile.c

[ ... ]

> @@ -2195,7 +2265,7 @@ static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
>  	/*
>  	 * Some architectures may have to restore extra metadata to the page
>  	 * when reading from swap. This metadata may be indexed by swap entry
> -	 * so this must be called before swap_free().
> +	 * so this must be called before folio_put_swap().
>  	 */
>  	arch_swap_restore(folio_swap(entry, folio), folio);
>
> @@ -2236,7 +2306,7 @@ static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
>  		new_pte = pte_mkuffd_wp(new_pte);
>  setpte:
>  	set_pte_at(vma->vm_mm, addr, pte, new_pte);
> -	swap_free(entry);
> +	folio_put_swap(folio, page);

Does this corrupt the swap map when ksm_might_need_to_copy() returns a
new folio? In that case, folio != swapcache, and the new folio is NOT in
the swap cache with folio->swap = 0. This would trigger
VM_WARN_ON_FOLIO(!folio_test_swapcache(folio), folio) and call
swap_entries_put_map() with entry.val = 0.

Compare with do_swap_page() which correctly uses folio_put_swap(swapcache,
NULL) when folio != swapcache. Should this use the original entry parameter
or the swapcache folio instead?