Re: [PATCH 2/3] mm/memory.c: Update local TLB if PTE entry exists

[Andrew Morton]

On Fri, 15 May 2020 12:10:08 +0800 Bibo Mao <maobibo@xxxxxxxxxxx> wrote:

> If there are two threads hitting page fault at the same page,
> one thread updates PTE entry and local TLB, the other can
> update local tlb also, rather than give up and do page fault
> again.
>
> ...
>
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -1770,8 +1770,8 @@ static vm_fault_t insert_pfn(struct vm_area_struct *vma, unsigned long addr,
>  			}
>  			entry = pte_mkyoung(*pte);
>  			entry = maybe_mkwrite(pte_mkdirty(entry), vma);
> -			if (ptep_set_access_flags(vma, addr, pte, entry, 1))
> -				update_mmu_cache(vma, addr, pte);
> +			ptep_set_access_flags(vma, addr, pte, entry, 1);
> +			update_mmu_cache(vma, addr, pte);

Presumably these changes mean that other architectures will run
update_mmu_cache() more frequently than they used to.  How much more
frequently, and what will be the impact of this change?  (Please fully
explain all this in the changelog).

>  		}
>  		goto out_unlock;
>  	}
>
> ...
>
> @@ -2463,7 +2462,8 @@ static inline bool cow_user_page(struct page *dst, struct page *src,
>  		vmf->pte = pte_offset_map_lock(mm, vmf->pmd, addr, &vmf->ptl);
>  		locked = true;
>  		if (!likely(pte_same(*vmf->pte, vmf->orig_pte))) {
> -			/* The PTE changed under us. Retry page fault. */
> +			/* The PTE changed under us, update local tlb */
> +			pdate_mmu_cache(vma, addr, vmf->pte);

Missing a 'u' there.  Which tells me this patch isn't the one which you
tested!

>  			ret = false;
>  			goto pte_unlock;
>  		}
>
> ...
>