Re: [PATCH v10 01/10] arm64/mm: add pte_set_huge() and pte_clear_huge()

[Dev Jain <dev.jain@arm.com>]

On 14/10/26 2:16 pm, Wen Jiang wrote:
> On Wed, 14 Oct 2026 at 14:03, Dev Jain <dev.jain@arm.com> wrote:
>>
>>
>>
>> On 02/10/26 2:39 pm, Wen Jiang wrote:
>>> vmap installs PTE-level block mappings (PTE_CONT on arm64) by reusing
>>> set_huge_pte_at() and huge_ptep_get_and_clear(), which are HugeTLB
>>> helpers gated by CONFIG_HUGETLB_PAGE. This makes the feature silently
>>> unavailable on CONFIG_HUGETLB_PAGE=n kernels and couples mm/vmalloc.c to
>>> HugeTLB internals it does not otherwise need.
>>>
>>> Add pte_set_huge()/pte_clear_huge() to arm64, joining the existing
>>> pmd_set_huge()/pud_set_huge() family in mm/mmu.c. They build on
>>> __set_ptes() and __get_and_clear_full_ptes() directly, so they do not
>>> depend on CONFIG_HUGETLB_PAGE.
>>>
>>> There is no caller yet: mm/vmalloc.c is converted later in this series,
>>> once the generic fallbacks are in place.
>>>
>>> Signed-off-by: Wen Jiang <jiangwen6@xiaomi.com>
>>> Reviewed-by: Barry Song <baohua@kernel.org>
>>> ---
>>>  arch/arm64/include/asm/pgtable.h |  6 ++++++
>>>  arch/arm64/mm/mmu.c              | 23 +++++++++++++++++++++++
>>>  2 files changed, 29 insertions(+)
>>>
>>> diff --git a/arch/arm64/include/asm/pgtable.h b/arch/arm64/include/asm/pgtable.h
>>> index 6000905a2e865..9b762eb68cff0 100644
>>> --- a/arch/arm64/include/asm/pgtable.h
>>> +++ b/arch/arm64/include/asm/pgtable.h
>>> @@ -1978,6 +1978,12 @@ static inline void clear_young_dirty_ptes(struct vm_area_struct *vma,
>>>
>>>  #endif /* CONFIG_ARM64_CONTPTE */
>>>
>>> +#define __HAVE_ARCH_PTE_SET_HUGE
>>> +void pte_set_huge(pte_t *ptep, unsigned long addr, phys_addr_t phys,
>>> +               pgprot_t prot, unsigned long size);
>>> +#define __HAVE_ARCH_PTE_CLEAR_HUGE
>>> +pte_t pte_clear_huge(pte_t *ptep, unsigned long addr, unsigned long size);
>>> +
>>>  #endif /* !__ASSEMBLER__ */
>>>
>>>  #endif /* __ASM_PGTABLE_H */
>>> diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
>>> index 79d90226fd5dc..ae2f15618c97d 100644
>>> --- a/arch/arm64/mm/mmu.c
>>> +++ b/arch/arm64/mm/mmu.c
>>> @@ -1912,6 +1912,29 @@ int pmd_clear_huge(pmd_t *pmdp)
>>>       return 1;
>>>  }
>>>
>>> +void pte_set_huge(pte_t *ptep, unsigned long addr, phys_addr_t phys,
>>> +               pgprot_t prot, unsigned long size)
>>> +{
>>> +     unsigned long pfn = __phys_to_pfn(phys);
>>> +     pte_t pte = pte_mkcont(pfn_pte(pfn, prot));
>>> +     unsigned int nr = size >> PAGE_SHIFT;
>>> +
>>> +     VM_WARN_ON(!IS_ALIGNED(size, CONT_PTE_SIZE));
>>> +     VM_WARN_ON(!IS_ALIGNED(addr, CONT_PTE_SIZE));
>>> +     VM_WARN_ON(!IS_ALIGNED(pfn, CONT_PTES));
>>> +
>>> +     /* For vmap, there is no valid -> valid transition */
>>> +     VM_WARN_ON(pte_valid(__ptep_get(ptep)));
>>> +
>>> +     __set_ptes(&init_mm, addr, ptep, pte, nr);
>>> +}
>>> +
>>> +pte_t pte_clear_huge(pte_t *ptep, unsigned long addr, unsigned long size)
>>> +{
>>> +     return __get_and_clear_full_ptes(&init_mm, addr, ptep,
>>> +                                      size >> PAGE_SHIFT, 0);
>>> +}
>>
>> I think we should have the same VM_WARN_ON() checks here, as for
>> pte_clear_huge(), in addition to also warning for
>> current->mm != &init_mm.
>>
>> We should make it clear that pte_clear_huge() gets called only for
>> kernel mappings, and the range we operate on is definitely an aligned
>> CONT range - which is why we do not go through get_and_clear_full_ptes ->
>> contpte_get_and_clear_full_ptes, and possibly contpte_try_unfold(), but
>> directly use __get_and_clear_full_ptes.
>>
> 
> Hi Dev,
> 
> I'll add the alignment VM_WARN_ON() checks to pte_clear_huge() to
> match pte_set_huge().
> 
> Regarding the current->mm != &init_mm check: these functions don't take
> an mm parameter -- &init_mm is hardcoded in the function body.
> current->mm would never equal &init_mm in practice: for user processes it
> points to the process's own mm, and for kernel threads it is NULL.
> So VM_WARN_ON(current->mm != &init_mm) would always fire.

Right.


> 
> The existing hugetlb helpers that vmalloc used before this decoupling
> (huge_ptep_get_and_clear -> get_clear_contig) also take mm as a
> parameter
> and don't check current->mm either -- the caller simply passes &init_mm.
> 
> How about adding a comment to make the "kernel mapping only" intent
> more explicit?
> Something like:

I think then we should just call the functions vmap_pte_set_huge and
vmap_pte_clear_huge. In arm64 we have init_pte() which sets the cont bit
at init time for kernel mappings. So to differentiate I think we should
prefix the function name by vmap.


> 
>  pte_t pte_clear_huge(pte_t *ptep, unsigned long addr, unsigned long size)
>  {
> +       /* Kernel mappings only */
> +
> +       VM_WARN_ON(!IS_ALIGNED(size, CONT_PTE_SIZE));
> +       VM_WARN_ON(!IS_ALIGNED(addr, CONT_PTE_SIZE));
> +
>         return __get_and_clear_full_ptes(&init_mm, addr, ptep,
>                                          size >> PAGE_SHIFT, 0);
>  }
> 
> Thanks,
> Wen