Re: [RESEND PATCH 1/2] arm64: change from CONT_PMD_SHIFT to CONT_PTE_SHIFT
On Wed, Dec 14, 2016 at 10:19:10PM +0800, zhongjiang wrote:
> From: zhong jiang
>
> I think that CONT_PTE_SHIFT is more reasonable even if they are some
> value. and the patch is not any functional change.
>
> Signed-off-by: zhong jiang
> ---
> arch/arm64/mm/hugetlbpage.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/arch/arm64/mm/hugetlbpage.c b/arch/arm64/mm/hugetlbpage.c
> index 2e49bd2..0a4c97b 100644
> --- a/arch/arm64/mm/hugetlbpage.c
> +++ b/arch/arm64/mm/hugetlbpage.c
> @@ -323,7 +323,7 @@ static __init int setup_hugepagesz(char *opt)
> static __init int add_default_hugepagesz(void)
> {
> if (size_to_hstate(CONT_PTES * PAGE_SIZE) == NULL)
> - hugetlb_add_hstate(CONT_PMD_SHIFT);
> + hugetlb_add_hstate(CONT_PTE_SHIFT);
> return 0;
> }
> arch_initcall(add_default_hugepagesz);
It looks like I fixed this already when I merged some other hugetlb
patches. See commit 6ed0038d5ecb ("arm64: Fix typo in
add_default_hugepagesz() for 64K pages").
--
Catalin
Re: [RESEND PATCH 1/2] arm64: change from CONT_PMD_SHIFT to CONT_PTE_SHIFT
On 2016/12/14 22:45, Ard Biesheuvel wrote:
> On 14 December 2016 at 14:19, zhongjiang wrote:
>> From: zhong jiang
>>
>> I think that CONT_PTE_SHIFT is more reasonable even if they are some
>> value. and the patch is not any functional change.
>>
> This may be the case for 64k pages, but not for 16k pages, and I
> actually think add_default_hugepagesz() could be made unconditional,
> given that both 64k on 4k kernels and 2 MB on 16k kernels are useful
> hugepage sizes that are not otherwise available by default.
I agree that we can make add_default_hugepagesz() to be unconditional.
but I do not know the history why it did so at that time. The patch
just is based on the current kernel.
by the way, please review the second patch if you have time. Any comment
will be welcomed.
Thanks
zhongjiang
>> Signed-off-by: zhong jiang
> Reviewed-by: Ard Biesheuvel
>
>> ---
>> arch/arm64/mm/hugetlbpage.c | 2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/arch/arm64/mm/hugetlbpage.c b/arch/arm64/mm/hugetlbpage.c
>> index 2e49bd2..0a4c97b 100644
>> --- a/arch/arm64/mm/hugetlbpage.c
>> +++ b/arch/arm64/mm/hugetlbpage.c
>> @@ -323,7 +323,7 @@ static __init int setup_hugepagesz(char *opt)
>> static __init int add_default_hugepagesz(void)
>> {
>> if (size_to_hstate(CONT_PTES * PAGE_SIZE) == NULL)
>> - hugetlb_add_hstate(CONT_PMD_SHIFT);
>> + hugetlb_add_hstate(CONT_PTE_SHIFT);
>> return 0;
>> }
>> arch_initcall(add_default_hugepagesz);
>> --
>> 1.8.3.1
>>
> .
>
Re: [RESEND PATCH 1/2] arm64: change from CONT_PMD_SHIFT to CONT_PTE_SHIFT
On 14 December 2016 at 14:19, zhongjiang wrote:
> From: zhong jiang
>
> I think that CONT_PTE_SHIFT is more reasonable even if they are some
> value. and the patch is not any functional change.
>
This may be the case for 64k pages, but not for 16k pages, and I
actually think add_default_hugepagesz() could be made unconditional,
given that both 64k on 4k kernels and 2 MB on 16k kernels are useful
hugepage sizes that are not otherwise available by default.
> Signed-off-by: zhong jiang
Reviewed-by: Ard Biesheuvel
> ---
> arch/arm64/mm/hugetlbpage.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/arch/arm64/mm/hugetlbpage.c b/arch/arm64/mm/hugetlbpage.c
> index 2e49bd2..0a4c97b 100644
> --- a/arch/arm64/mm/hugetlbpage.c
> +++ b/arch/arm64/mm/hugetlbpage.c
> @@ -323,7 +323,7 @@ static __init int setup_hugepagesz(char *opt)
> static __init int add_default_hugepagesz(void)
> {
> if (size_to_hstate(CONT_PTES * PAGE_SIZE) == NULL)
> - hugetlb_add_hstate(CONT_PMD_SHIFT);
> + hugetlb_add_hstate(CONT_PTE_SHIFT);
> return 0;
> }
> arch_initcall(add_default_hugepagesz);
> --
> 1.8.3.1
>
