Re: [PATCH v2 2/3] mm: rmap: support batched checks of the references for large folios

[Ryan Roberts <ryan.roberts@arm.com>]

>>>   +#define clear_flush_young_ptes clear_flush_young_ptes
>>> +static inline int clear_flush_young_ptes(struct vm_area_struct *vma,
>>> +                    unsigned long addr, pte_t *ptep,
>>> +                    unsigned int nr)
>>> +{
>>> +    if (likely(nr == 1))
>>> +        return __ptep_clear_flush_young(vma, addr, ptep);
>>
>> Bug: This is broken if core-mm tries to call this for nr=1 on a pte that is part
>> of a contpte mapping.
>>
>> The similar fastpaths are here to prevent regressing the common small folio case.
> 
> Thanks for catching this. I had considered this before, but I still missed it.
> 
>> I guess here the best approach is (note no leading underscores):
>>
>>     if (likely(nr == 1))
>>         return ptep_clear_flush_young(vma, addr, ptep);
> 
> However, I prefer to use pte_cont() to check it. Later, I plan to clean up the
> ptep_clear_flush_young().
> 
>     if (nr == 1 && !pte_cont(__ptep_get(ptep))
>         return __ptep_clear_flush_young(vma, addr, ptep);

Sure. That would follow the pattern in clear_young_dirty_ptes(). Please use the
likely() hint as is done everywhere else:

	if (likely(nr == 1 && !pte_cont(__ptep_get(ptep))))

I notice that ptep_test_and_clear_young() and ptep_clear_flush_young() are both
testing aginst pte_valid_cont(). These could probably be relaxed to pte_cont()
since it is implicit the the pte must be valid?