Re: [PATCH V3 1/3] arm64/mm: Drop __HAVE_ARCH_HUGE_PTEP_GET

[Anshuman Khandual]

On 05/09/2020 03:39 AM, Mike Kravetz wrote:
> On 5/7/20 8:07 PM, Anshuman Khandual wrote:
>> Platform specific huge_ptep_get() is required only when fetching the huge
>> PTE involves more than just dereferencing the page table pointer. This is
>> not the case on arm64 platform. Hence huge_ptep_pte() can be dropped along
>> with it's __HAVE_ARCH_HUGE_PTEP_GET subscription. Before that, it updates
>> the generic huge_ptep_get() with READ_ONCE() which will prevent known page
>> table issues with THP on arm64.
>>
>> https://lore.kernel.org/r/1506527369-19535-1-git-send-email-will.deacon@xxxxxxx/
>>
>> Cc: Catalin Marinas <catalin.marinas@xxxxxxx>
>> Cc: Will Deacon <will@xxxxxxxxxx>
>> Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
>> Cc: linux-arm-kernel@xxxxxxxxxxxxxxxxxxx
>> Cc: linux-mm@xxxxxxxxx
>> Cc: linux-kernel@xxxxxxxxxxxxxxx
>> Signed-off-by: Anshuman Khandual <anshuman.khandual@xxxxxxx>
>> ---
>>  arch/arm64/include/asm/hugetlb.h | 6 ------
>>  include/asm-generic/hugetlb.h    | 2 +-
>>  2 files changed, 1 insertion(+), 7 deletions(-)
>>
>> diff --git a/arch/arm64/include/asm/hugetlb.h b/arch/arm64/include/asm/hugetlb.h
>> index 2eb6c234d594..b88878ddc88b 100644
>> --- a/arch/arm64/include/asm/hugetlb.h
>> +++ b/arch/arm64/include/asm/hugetlb.h
>> @@ -17,12 +17,6 @@
>>  extern bool arch_hugetlb_migration_supported(struct hstate *h);
>>  #endif
>>  
>> -#define __HAVE_ARCH_HUGE_PTEP_GET
>> -static inline pte_t huge_ptep_get(pte_t *ptep)
>> -{
>> -	return READ_ONCE(*ptep);
>> -}
>> -
>>  static inline int is_hugepage_only_range(struct mm_struct *mm,
>>  					 unsigned long addr, unsigned long len)
>>  {
>> diff --git a/include/asm-generic/hugetlb.h b/include/asm-generic/hugetlb.h
>> index 822f433ac95c..40f85decc2ee 100644
>> --- a/include/asm-generic/hugetlb.h
>> +++ b/include/asm-generic/hugetlb.h
>> @@ -122,7 +122,7 @@ static inline int huge_ptep_set_access_flags(struct vm_area_struct *vma,
>>  #ifndef __HAVE_ARCH_HUGE_PTEP_GET
>>  static inline pte_t huge_ptep_get(pte_t *ptep)
>>  {
>> -	return *ptep;
>> +	return READ_ONCE(*ptep);
>>  }
>>  #endif
> 
> I know you made this change in response to Will's comment.  And, since
> changes were made to consistently use READ_ONCE in arm64 code, it makes
> sense for that architecture.
> 
> However, with this change to generic code, you introduce READ_ONCE to
> other architectures where it was not used before.  Could this possibly
> introduce inconsistencies in their use of READ_ONCE?  To be honest, I
> am not very good at identifying any possible issues this could cause.
> However, it does seem possible.

Could you please give some more details. Is there any particular problem
which might be caused by this new READ_ONCE() here, that you you are
concerned about. READ_ONCE() is already getting used in multiple places
in core MM which can not be configured out (like mm/gup.c). It is getting
used in core HugeTLB (mm/hugetlb.c) as well. AFAICS, there is no standard
for using READ_ONCE() while walking page tables entries. We have examples
in core MM for both ways.

> 
> Will was nervous about dropping this from arm64.  I'm just a little nervous
> about adding it to other architectures.
> 
AFAICS, __HAVE_ARCH_HUGE_PTEP_GET should be used on a platform only when
a HugeTLB entry could not constructed by dereferencing a page table entry
as in the case with ARM (32 bit). Using READ_ONCE() while dereferencing is
really not a special case that will need __HAVE_ARCH_HUGE_PTEP_GET. Moving
READ_ONCE() into generic definition solves the problem while also taking
care of a known problem on arm64. IMHO, it seems like the right thing to
do unless there is another problem that pops up some where else because of
READ_ONCE().