Re: [PATCH v2 07/13] powerpc: add support for folded p4d page tables

[Christophe Leroy <christophe.leroy@c-s.fr>]

Le 26/02/2020 à 11:56, Mike Rapoport a écrit :
> On Wed, Feb 26, 2020 at 10:46:13AM +0100, Christophe Leroy wrote:
>>
>>
>> Le 26/02/2020 à 10:13, Mike Rapoport a écrit :
>>> On Tue, Feb 18, 2020 at 12:54:40PM +0200, Mike Rapoport wrote:
>>>> On Sun, Feb 16, 2020 at 11:41:07AM +0100, Christophe Leroy wrote:
>>>>>
>>>>>
>>>>> Le 16/02/2020 à 09:18, Mike Rapoport a écrit :
>>>>>> From: Mike Rapoport <rppt@linux.ibm.com>
>>>>>>
>>>>>> Implement primitives necessary for the 4th level folding, add walks of p4d
>>>>>> level where appropriate and replace 5level-fixup.h with pgtable-nop4d.h.
>>>>>
>>>>> I don't think it is worth adding all this additionnals walks of p4d, this
>>>>> patch could be limited to changes like:
>>>>>
>>>>> -		pud = pud_offset(pgd, gpa);
>>>>> +		pud = pud_offset(p4d_offset(pgd, gpa), gpa);
>>>>>
>>>>> The additionnal walks should be added through another patch the day powerpc
>>>>> need them.
>>>>
>>>> Ok, I'll update the patch to reduce walking the p4d.
>>>
>>> Here's what I have with more direct acceses from pgd to pud.
>>
>> I went quickly through. This looks promising.
>>
>> Do we need the walk_p4d() in arch/powerpc/mm/ptdump/hashpagetable.c ?
>> Can't we just do
>>
>> @@ -445,7 +459,7 @@ static void walk_pagetables(struct pg_state *st)
>>   		addr = KERN_VIRT_START + i * PGDIR_SIZE;
>>   		if (!pgd_none(*pgd))
>>   			/* pgd exists */
>> -			walk_pud(st, pgd, addr);
>> +			walk_pud(st, p4d_offset(pgd, addr), addr);
> 
> We can do
> 
> 	addr = KERN_VIRT_START + i * PGDIR_SIZE;
> 	p4d = p4d_offset(pgd, addr);
> 	if (!p4d_none(*pgd))
> 		walk_pud()
> 
> But I don't think this is really essential. Again, we are trading off code
> consistency vs line count. I don't think line count is that important.

Ok.

Christophe