Re: [PATCH 2/3] mm/huge_memory.c: update tlb entry if pmd is changed

[Mike Kravetz]

On 6/24/20 2:26 AM, Bibo Mao wrote:
> When set_pmd_at is called in function do_huge_pmd_anonymous_page,
> new tlb entry can be added by software on MIPS platform.
> 
> Here add update_mmu_cache_pmd when pmd entry is set, and
> update_mmu_cache_pmd is defined as empty excepts arc/mips platform.
> This patch has no negative effect on other platforms except arc/mips
> system.

I am confused by this comment.  It appears that update_mmu_cache_pmd
is defined as non-empty on arc, mips, powerpc and sparc architectures.
Am I missing something?

If those architectures do provide update_mmu_cache_pmd, then the previous
patch and this one now call update_mmu_cache_pmd with the actual faulting
address instead of the huge page aligned address.  This was intentional
for mips.  However, are there any potential issues on the other architectures?
I am no expert in any of those architectures.  arc looks like it could be
problematic as update_mmu_cache_pmd calls update_mmu_cache and then
operates on (address & PAGE_MASK).  That could now be different.

-- 
Mike Kravetz

> 
> Signed-off-by: Bibo Mao <maobibo@xxxxxxxxxxx>
> ---
>  mm/huge_memory.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index 0f9187b..8b4ccf7 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -643,6 +643,7 @@ static vm_fault_t __do_huge_pmd_anonymous_page(struct vm_fault *vmf,
>  		lru_cache_add_active_or_unevictable(page, vma);
>  		pgtable_trans_huge_deposit(vma->vm_mm, vmf->pmd, pgtable);
>  		set_pmd_at(vma->vm_mm, haddr, vmf->pmd, entry);
> +		update_mmu_cache_pmd(vma, vmf->address, vmf->pmd);
>  		add_mm_counter(vma->vm_mm, MM_ANONPAGES, HPAGE_PMD_NR);
>  		mm_inc_nr_ptes(vma->vm_mm);
>  		spin_unlock(vmf->ptl);
> @@ -756,6 +757,7 @@ vm_fault_t do_huge_pmd_anonymous_page(struct vm_fault *vmf)
>  			} else {
>  				set_huge_zero_page(pgtable, vma->vm_mm, vma,
>  						   haddr, vmf->pmd, zero_page);
> +				update_mmu_cache_pmd(vma, vmf->address, vmf->pmd);
>  				spin_unlock(vmf->ptl);
>  				set = true;
>  			}
>