Re: [PATCH 01/13] mm: Update ptep_get_lockless()s comment

[Peter Zijlstra <peterz@infradead.org>]

On Sun, Oct 30, 2022 at 03:47:36PM -0700, Linus Torvalds wrote:

>  include/linux/rmap.h |  1 +
>  mm/memory.c          |  3 ++-
>  mm/rmap.c            | 24 ++++++++++++++++++++++++
>  3 files changed, 27 insertions(+), 1 deletion(-)
> 
> diff --git a/include/linux/rmap.h b/include/linux/rmap.h
> index bd3504d11b15..f62af001707c 100644
> --- a/include/linux/rmap.h
> +++ b/include/linux/rmap.h
> @@ -196,6 +196,7 @@ void page_add_new_anon_rmap(struct page *, struct vm_area_struct *,
>  		unsigned long address);
>  void page_add_file_rmap(struct page *, struct vm_area_struct *,
>  		bool compound);
> +void page_zap_pte_rmap(struct page *);
>  void page_remove_rmap(struct page *, struct vm_area_struct *,
>  		bool compound);
>  
> diff --git a/mm/memory.c b/mm/memory.c
> index f88c351aecd4..c893f5ffc5a8 100644
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -1452,8 +1452,9 @@ static unsigned long zap_pte_range(struct mmu_gather *tlb,
>  				    likely(!(vma->vm_flags & VM_SEQ_READ)))
>  					mark_page_accessed(page);
>  			}
> +			page_zap_pte_rmap(page);
> +			munlock_vma_page(page, vma, false);
>  			rss[mm_counter(page)]--;
> -			page_remove_rmap(page, vma, false);
>  			if (unlikely(page_mapcount(page) < 0))
>  				print_bad_pte(vma, addr, ptent, page);
>  			if (unlikely(__tlb_remove_page(tlb, page))) {
> diff --git a/mm/rmap.c b/mm/rmap.c
> index 2ec925e5fa6a..28b51a31ebb0 100644
> --- a/mm/rmap.c
> +++ b/mm/rmap.c
> @@ -1412,6 +1412,30 @@ static void page_remove_anon_compound_rmap(struct page *page)
>  		__mod_lruvec_page_state(page, NR_ANON_MAPPED, -nr);
>  }
>  
> +/**
> + * page_zap_pte_rmap - take down a pte mapping from a page
> + * @page:	page to remove mapping from
> + *
> + * This is the simplified form of page_remove_rmap(), that only
> + * deals with last-level pages, so 'compound' is always false,
> + * and the caller does 'munlock_vma_page(page, vma, compound)'
> + * separately.
> + *
> + * This allows for a much simpler calling convention and code.
> + *
> + * The caller holds the pte lock.
> + */
> +void page_zap_pte_rmap(struct page *page)
> +{

One could consider adding something like:

#ifdef USE_SPLIT_PTE_PTLOCKS
	lockdep_assert_held(ptlock_ptr(page))
#endif


> +	if (!atomic_add_negative(-1, &page->_mapcount))
> +		return;
> +
> +	lock_page_memcg(page);
> +	__dec_lruvec_page_state(page,
> +		PageAnon(page) ? NR_ANON_MAPPED : NR_FILE_MAPPED);
> +	unlock_page_memcg(page);
> +}

Took me a little while, but yes, .compound=false seems to reduce to
this.