On Fri, May 17, 2024 at 09:00:05PM +0200, Christophe Leroy wrote:
> huge_pte_alloc() for non-HUGEPD targets is reserved for 8xx at the
> moment. In order to convert other targets for non-HUGEPD, complement
> huge_pte_alloc() to support any standard cont-PxD setup.
>
> Signed-off-by: Christophe Leroy <[email protected]>
> ---
> arch/powerpc/mm/hugetlbpage.c | 25 ++++++++++++++++++++++++-
> 1 file changed, 24 insertions(+), 1 deletion(-)
>
> diff --git a/arch/powerpc/mm/hugetlbpage.c b/arch/powerpc/mm/hugetlbpage.c
> index 42b12e1ec851..f8aefa1e7363 100644
> --- a/arch/powerpc/mm/hugetlbpage.c
> +++ b/arch/powerpc/mm/hugetlbpage.c
> @@ -195,11 +195,34 @@ pte_t *huge_pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma,
> pte_t *huge_pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma,
> unsigned long addr, unsigned long sz)
> {
> - pmd_t *pmd = pmd_off(mm, addr);
> + pgd_t *pgd;
> + p4d_t *p4d;
> + pud_t *pud;
> + pmd_t *pmd;
> +
> + addr &= ~(sz - 1);
> + pgd = pgd_offset(mm, addr);
> +
> + p4d = p4d_offset(pgd, addr);
> + if (sz >= PGDIR_SIZE)
> + return (pte_t *)p4d;
> +
> + pud = pud_alloc(mm, p4d, addr);
> + if (!pud)
> + return NULL;
> + if (sz >= PUD_SIZE)
> + return (pte_t *)pud;
> +
> + pmd = pmd_alloc(mm, pud, addr);
> + if (!pmd)
> + return NULL;
>
> if (sz < PMD_SIZE)
> return pte_alloc_huge(mm, pmd, addr, sz);
>
> + if (!IS_ENABLED(CONFIG_PPC_8xx))
> + return (pte_t *)pmd;
So only 8xx has cont-PMD for hugepages?
> +
> if (sz != SZ_8M)
> return NULL;
Since this function is the core for allocation huge pages, I think it would
benefit from a comment at the top explaining the possible layouts.
e.g: Who can have cont-{P4d,PUD,PMD} etc.
A brief explanation of the possible scheme for all powerpc platforms.
That would help people looking into this in a future.
--
Oscar Salvador
SUSE Labs
Le 25/05/2024 à 06:29, Oscar Salvador a écrit :
> On Fri, May 17, 2024 at 09:00:05PM +0200, Christophe Leroy wrote:
>> huge_pte_alloc() for non-HUGEPD targets is reserved for 8xx at the
>> moment. In order to convert other targets for non-HUGEPD, complement
>> huge_pte_alloc() to support any standard cont-PxD setup.
>>
>> Signed-off-by: Christophe Leroy <[email protected]>
>> ---
>> arch/powerpc/mm/hugetlbpage.c | 25 ++++++++++++++++++++++++-
>> 1 file changed, 24 insertions(+), 1 deletion(-)
>>
>> diff --git a/arch/powerpc/mm/hugetlbpage.c b/arch/powerpc/mm/hugetlbpage.c
>> index 42b12e1ec851..f8aefa1e7363 100644
>> --- a/arch/powerpc/mm/hugetlbpage.c
>> +++ b/arch/powerpc/mm/hugetlbpage.c
>> @@ -195,11 +195,34 @@ pte_t *huge_pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma,
>> pte_t *huge_pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma,
>> unsigned long addr, unsigned long sz)
>> {
>> - pmd_t *pmd = pmd_off(mm, addr);
>> + pgd_t *pgd;
>> + p4d_t *p4d;
>> + pud_t *pud;
>> + pmd_t *pmd;
>> +
>> + addr &= ~(sz - 1);
>> + pgd = pgd_offset(mm, addr);
>> +
>> + p4d = p4d_offset(pgd, addr);
>> + if (sz >= PGDIR_SIZE)
>> + return (pte_t *)p4d;
>> +
>> + pud = pud_alloc(mm, p4d, addr);
>> + if (!pud)
>> + return NULL;
>> + if (sz >= PUD_SIZE)
>> + return (pte_t *)pud;
>> +
>> + pmd = pmd_alloc(mm, pud, addr);
>> + if (!pmd)
>> + return NULL;
>>
>> if (sz < PMD_SIZE)
>> return pte_alloc_huge(mm, pmd, addr, sz);
>>
>> + if (!IS_ENABLED(CONFIG_PPC_8xx))
>> + return (pte_t *)pmd;
>
> So only 8xx has cont-PMD for hugepages?
No, all have cont-PMD but only 8xx handles pages greater than PMD_SIZE
as cont-PTE instead of cont-PMD.
>
>> +
>> if (sz != SZ_8M)
>> return NULL;
>
> Since this function is the core for allocation huge pages, I think it would
> benefit from a comment at the top explaining the possible layouts.
> e.g: Who can have cont-{P4d,PUD,PMD} etc.
> A brief explanation of the possible scheme for all powerpc platforms.
All is standard except 8xx, let's just have a comment for 8xx.
>
> That would help people looking into this in a future.
>
>
>
On Sat, May 25, 2024 at 06:44:06AM +0000, Christophe Leroy wrote:
> No, all have cont-PMD but only 8xx handles pages greater than PMD_SIZE
> as cont-PTE instead of cont-PMD.
Yes, sorry, I managed to confuse myself. It is obvious from the code.
--
Oscar Salvador
SUSE Labs