Re: [PATCH v2] mm: use update_mmu_tlb() on the second thread

[maobibo <maobibo@loongson.cn>]

在 2022/9/29 11:47, Qi Zheng 写道:
> 
> 
> On 2022/9/29 11:27, maobibo wrote:
>> 在 2022/9/29 11:07, Qi Zheng 写道:
>>>
>>>
>>> On 2022/9/26 22:34, David Hildenbrand wrote:
>>>> On 26.09.22 13:56, Qi Zheng wrote:
>>>>> As message in commit 7df676974359 ("mm/memory.c: Update local TLB
>>>>> if PTE entry exists") said, we should update local TLB only on the
>>>>> second thread. So in the do_anonymous_page() here, we should use
>>>>> update_mmu_tlb() instead of update_mmu_cache() on the second thread.
>>>>>
>>>>> Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
>>>>> Reviewed-by: Muchun Song <songmuchun@bytedance.com>
>>>>> ---
>>>>> v1: https://lore.kernel.org/lkml/20220924053239.91661-1-zhengqi.arch@bytedance.com/
>>>>>
>>>>> Changelog in v1 -> v2:
>>>>>    - change the subject and commit message (David)
>>>>>
>>>>>    mm/memory.c | 2 +-
>>>>>    1 file changed, 1 insertion(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/mm/memory.c b/mm/memory.c
>>>>> index 118e5f023597..9e11c783ba0e 100644
>>>>> --- a/mm/memory.c
>>>>> +++ b/mm/memory.c
>>>>> @@ -4122,7 +4122,7 @@ static vm_fault_t do_anonymous_page(struct vm_fault *vmf)
>>>>>        vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address,
>>>>>                &vmf->ptl);
>>>>>        if (!pte_none(*vmf->pte)) {
>>>>> -        update_mmu_cache(vma, vmf->address, vmf->pte);
>>>>> +        update_mmu_tlb(vma, vmf->address, vmf->pte);
>>>>>            goto release;
>>>>>        }
>>>>
>>>>
>>>> Staring at 7df676974359, it indeed looks like an accidental use [nothing else in that patch uses update_mmu_cache].
>>>>
>>>> So it looks good to me, but a confirmation from Bibo Mao might be good.
>>>
>>> Thanks, and Hi Bibo, any comments here? :)
>>
>> update_mmu_tlb is defined as empty on loongarch, maybe some lines should
>> be added in file arch/loongarch/include/asm/pgtable.h like this:
> 
> Seems like a bug? Because there are many other places where
> update_mmu_tlb() is called.
> 
>>
>> +#define __HAVE_ARCH_UPDATE_MMU_TLB
>> +#define update_mmu_tlb  update_mmu_cache
> 
> If so, I can make the above as a separate fix patch.
It sounds good to me.  

Huacai, do you have any comments?

regards
bibo, mao
> 
> Thanks,
> Qi
> 
>>
>> regards
>> bibo mao
>>>
>>>>
>>>
>>
>