Re: [PATCH 08/31] mm/page_vma_mapped: pte_offset_map_nolock() not pte_lockptr()

[Qi Zheng <qi.zheng@linux.dev>]

On 2023/5/22 12:58, Hugh Dickins wrote:
> map_pte() use pte_offset_map_nolock(), to make sure of the ptl belonging
> to pte, even if pmd entry is then changed racily: page_vma_mapped_walk()
> use that instead of getting pte_lockptr() later, or restart if map_pte()
> found no page table.
> 
> Signed-off-by: Hugh Dickins <hughd@google.com>
> ---
>   mm/page_vma_mapped.c | 28 ++++++++++++++++++++++------
>   1 file changed, 22 insertions(+), 6 deletions(-)
> 
> diff --git a/mm/page_vma_mapped.c b/mm/page_vma_mapped.c
> index 947dc7491815..2af734274073 100644
> --- a/mm/page_vma_mapped.c
> +++ b/mm/page_vma_mapped.c
> @@ -13,16 +13,28 @@ static inline bool not_found(struct page_vma_mapped_walk *pvmw)
>   	return false;
>   }
>   
> -static bool map_pte(struct page_vma_mapped_walk *pvmw)
> +static bool map_pte(struct page_vma_mapped_walk *pvmw, spinlock_t **ptlp)
>   {
>   	if (pvmw->flags & PVMW_SYNC) {
>   		/* Use the stricter lookup */
>   		pvmw->pte = pte_offset_map_lock(pvmw->vma->vm_mm, pvmw->pmd,
>   						pvmw->address, &pvmw->ptl);
> -		return true;
> +		*ptlp = pvmw->ptl;
> +		return !!pvmw->pte;
>   	}
>   
> -	pvmw->pte = pte_offset_map(pvmw->pmd, pvmw->address);
> +	/*
> +	 * It is important to return the ptl corresponding to pte,
> +	 * in case *pvmw->pmd changes underneath us; so we need to
> +	 * return it even when choosing not to lock, in case caller
> +	 * proceeds to loop over next ptes, and finds a match later.
> +	 * Though, in most cases, page lock already protects this.
> +	 */
> +	pvmw->pte = pte_offset_map_nolock(pvmw->vma->vm_mm, pvmw->pmd,
> +					  pvmw->address, ptlp);
> +	if (!pvmw->pte)
> +		return false;
> +
>   	if (pvmw->flags & PVMW_MIGRATION) {
>   		if (!is_swap_pte(*pvmw->pte))
>   			return false;
> @@ -51,7 +63,7 @@ static bool map_pte(struct page_vma_mapped_walk *pvmw)
>   	} else if (!pte_present(*pvmw->pte)) {
>   		return false;
>   	}
> -	pvmw->ptl = pte_lockptr(pvmw->vma->vm_mm, pvmw->pmd);
> +	pvmw->ptl = *ptlp;
>   	spin_lock(pvmw->ptl);
>   	return true;
>   }
> @@ -156,6 +168,7 @@ bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw)
>   	struct vm_area_struct *vma = pvmw->vma;
>   	struct mm_struct *mm = vma->vm_mm;
>   	unsigned long end;
> +	spinlock_t *ptl;
>   	pgd_t *pgd;
>   	p4d_t *p4d;
>   	pud_t *pud;
> @@ -257,8 +270,11 @@ bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw)
>   			step_forward(pvmw, PMD_SIZE);
>   			continue;
>   		}
> -		if (!map_pte(pvmw))
> +		if (!map_pte(pvmw, &ptl)) {
> +			if (!pvmw->pte)
> +				goto restart;

Could pvmw->pmd be changed? Otherwise, how about just jumping to the
retry label below?

@@ -205,6 +205,8 @@ bool page_vma_mapped_walk(struct 
page_vma_mapped_walk *pvmw)
                 }

                 pvmw->pmd = pmd_offset(pud, pvmw->address);
+
+retry:
                 /*
                  * Make sure the pmd value isn't cached in a register 
by the
                  * compiler and used as a stale value after we've 
observed a

>   			goto next_pte;
> +		}
>   this_pte:
>   		if (check_pte(pvmw))
>   			return true;
> @@ -281,7 +297,7 @@ bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw)
>   		} while (pte_none(*pvmw->pte));
>   
>   		if (!pvmw->ptl) {
> -			pvmw->ptl = pte_lockptr(mm, pvmw->pmd);
> +			pvmw->ptl = ptl;
>   			spin_lock(pvmw->ptl);
>   		}
>   		goto this_pte;

-- 
Thanks,
Qi