[PATCH V3 1/6] powerpc/mm/book3s: Update pmd_present to look at _PAGE_PRESENT bit

[Christophe LEROY]

Le 21/09/2018 à 12:26, Aneesh Kumar K.V a écrit :
> On 9/21/18 11:25 AM, Christophe LEROY wrote:
>>
>>
>> Le 20/09/2018 à 20:09, Aneesh Kumar K.V a écrit :
> 
>>> diff --git a/arch/powerpc/mm/pgtable.c b/arch/powerpc/mm/pgtable.c
>>> index d71c7777669c..aee04b209b51 100644
>>> --- a/arch/powerpc/mm/pgtable.c
>>> +++ b/arch/powerpc/mm/pgtable.c
>>> @@ -188,11 +188,10 @@ void set_pte_at(struct mm_struct *mm, unsigned 
>>> long addr, pte_t *ptep,
>>>           pte_t pte)
>>>   {
>>>       /*
>>> -     * When handling numa faults, we already have the pte marked
>>> -     * _PAGE_PRESENT, but we can be sure that it is not in hpte.
>>> -     * Hence we can use set_pte_at for them.
>>> +     * Make sure hardware valid bit is not set. We don't do
>>> +     * tlb flush for this update.
>>>        */
>>> -    VM_WARN_ON(pte_present(*ptep) && !pte_protnone(*ptep));
>>> +    VM_WARN_ON(pte_val(*ptep) & _PAGE_PRESENT);
>>
>> Why not using pte_present() anymore ?
>>
>> Also, you are removing the pte_protnone() check, won't it change the 
>> behaviour ?
>>
>> If we can't use pte_present(), can we create a new helper for that 
>> (allthough _PAGE_PRESENT exists on all platforms).
>>
>> Christophe
>>
> 
> This patch update a page table clear to clear _PAGE_PRESENT and mark it 
> invalid via _PAGE_INVALID. The pte_present now looks at both the flag. 
> That is we want these transient clear of pte to be considered as present 
> pte even if _PAGE_PRESENT is cleared. What we are catching by the debug 
> BUG_ON in these function is we are not using them to set a pte where the 
> old entry has a hadware valid bit set. This is because we don't do any 
> tlb flush with set_pte_at.
> 
> 
> So the reason for pte_present -> pte_val() & _PAGE_PRESENT is because we 
> swtiched the clear to clear _PAGE_PRESENT and set _PAGE_INVALID and 
> pte_present now check both.
> 
> The reason for the removal of pte_protnone is because we dropped that 
> set_pte_at usage from core autonuma code long time back.
> 
> Now Considering we are calling this from mm/pgtable.c With your approach 
> of not using pte flags directly in core code we could switch this to 
> pte_hw_valid(). May be we can do that as an addon patch?
> 

Ok, depending on which serie goes first, I'll add it in mine if I have 
to rebase.

Christophe