Re: [PATCH] powerpc/64e: Fix wrong test in __ptep_test_and_clear_young()

[Christophe Leroy <christophe.leroy@csgroup.eu>]

Le 05/09/2023 à 04:36, Michael Ellerman a écrit :
> Christophe Leroy <christophe.leroy@csgroup.eu> writes:
>> Commit 45201c879469 ("powerpc/nohash: Remove hash related code from
>> nohash headers.") replaced:
>>
>>    if ((pte_val(*ptep) & (_PAGE_ACCESSED | _PAGE_HASHPTE)) == 0)
>> 	return 0;
>>
>> By:
>>
>>    if (pte_young(*ptep))
>> 	return 0;
>>
>> But it should be:
>>
>>    if (!pte_young(*ptep))
>> 	return 0;
> 
> 
> That seems bad :)
> 
> But I don't know off the top of my head where
> __ptep_test_and_clear_young() is used, and so what the symptoms could
> be. Presumably nothing too bad or someone would have noticed?
> 

The two uses in mm/vmscan.c are as follows:

	if (!ptep_test_and_clear_young(args->vma, addr, pte + i))
		VM_WARN_ON_ONCE(true);

So it seems to be expected to never happen.

The only useful place it is used seems to be folio_check_references() 
which is part of the reclaim process. So probably it messes up swap, but 
to what extent ? ppc64e is for embedded systems. Do embedded systems 
have swap at all ?

Also surprising that it is also called from mm/debug_vm_pgtable.c so 
shouldn't we have noticed earlier ? I'll check if it works.