Re: [PATCH 1/5] arm64/mm: Drop pte_mkhuge()

[Anshuman Khandual <anshuman.khandual@arm.com>]

On 10/14/24 16:38, Ryan Roberts wrote:
> On 14/10/2024 09:59, Anshuman Khandual wrote:
>>
>>
>> On 10/9/24 18:50, Ryan Roberts wrote:
>>> On 05/10/2024 13:38, Anshuman Khandual wrote:
>>>> Core HugeTLB defines arch_make_huge_pte() fallback definition, which calls
>>>> platform provided pte_mkhuge(). But if any platform already provides custom
>>>> arch_make_huge_pte(), then it does not need to provide pte_mkhuge(). arm64
>>>> defines arch_make_huge_pte(), but then also calls pte_mkhuge() internally.
>>>> This creates confusion as if both of these callbacks are being used in core
>>>> HugeTLB and required to be defined in the platform.
>>>>
>>>> This changes arch_make_huge_pte() to create block mapping directly and also
>>>> drops off now redundant helper pte_mkhuge(), making things clear. Also this
>>>> changes HugeTLB page creation from just clearing the PTE_TABLE_BIT (bit[1])
>>>> to actually setting bits[1:0] via PTE_TYPE_[MASK|SECT] instead.
>>>>
>>>> Cc: Catalin Marinas <catalin.marinas@arm.com>
>>>> Cc: Will Deacon <will@kernel.org>
>>>> Cc: Ard Biesheuvel <ardb@kernel.org>
>>>> Cc: Ryan Roberts <ryan.roberts@arm.com>
>>>> Cc: Mark Rutland <mark.rutland@arm.com>
>>>> Cc: linux-arm-kernel@lists.infradead.org
>>>> Cc: linux-kernel@vger.kernel.org
>>>> Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com>
>>>> ---
>>>>  arch/arm64/include/asm/pgtable-hwdef.h | 1 +
>>>>  arch/arm64/include/asm/pgtable.h       | 5 -----
>>>>  arch/arm64/mm/hugetlbpage.c            | 2 +-
>>>>  3 files changed, 2 insertions(+), 6 deletions(-)
>>>>
>>>> diff --git a/arch/arm64/include/asm/pgtable-hwdef.h b/arch/arm64/include/asm/pgtable-hwdef.h
>>>> index fd330c1db289..956a702cb532 100644
>>>> --- a/arch/arm64/include/asm/pgtable-hwdef.h
>>>> +++ b/arch/arm64/include/asm/pgtable-hwdef.h
>>>> @@ -158,6 +158,7 @@
>>>>  #define PTE_VALID		(_AT(pteval_t, 1) << 0)
>>>>  #define PTE_TYPE_MASK		(_AT(pteval_t, 3) << 0)
>>>>  #define PTE_TYPE_PAGE		(_AT(pteval_t, 3) << 0)
>>>> +#define PTE_TYPE_SECT		(_AT(pteval_t, 1) << 0)
>>>>  #define PTE_TABLE_BIT		(_AT(pteval_t, 1) << 1)
>>>>  #define PTE_USER		(_AT(pteval_t, 1) << 6)		/* AP[1] */
>>>>  #define PTE_RDONLY		(_AT(pteval_t, 1) << 7)		/* AP[2] */
>>>> diff --git a/arch/arm64/include/asm/pgtable.h b/arch/arm64/include/asm/pgtable.h
>>>> index c329ea061dc9..fa4c32a9f572 100644
>>>> --- a/arch/arm64/include/asm/pgtable.h
>>>> +++ b/arch/arm64/include/asm/pgtable.h
>>>> @@ -438,11 +438,6 @@ static inline void __set_ptes(struct mm_struct *mm,
>>>>  	}
>>>>  }
>>>>  
>>>> -/*
>>>> - * Huge pte definitions.
>>>> - */
>>>> -#define pte_mkhuge(pte)		(__pte(pte_val(pte) & ~PTE_TABLE_BIT))
>>>> -
>>>>  /*
>>>>   * Hugetlb definitions.
>>>>   */
>>>> diff --git a/arch/arm64/mm/hugetlbpage.c b/arch/arm64/mm/hugetlbpage.c
>>>> index 5f1e2103888b..5922c95630ad 100644
>>>> --- a/arch/arm64/mm/hugetlbpage.c
>>>> +++ b/arch/arm64/mm/hugetlbpage.c
>>>> @@ -361,7 +361,7 @@ pte_t arch_make_huge_pte(pte_t entry, unsigned int shift, vm_flags_t flags)
>>>>  {
>>>>  	size_t pagesize = 1UL << shift;
>>>>  
>>>> -	entry = pte_mkhuge(entry);
>>>> +	entry = __pte((pte_val(entry) & ~PTE_TYPE_MASK) | PTE_TYPE_SECT);
>>>
>>> I think there may be an existing bug here; if pagesize == CONT_PTE_SIZE, then
>>> entry will be placed in the level 3 table. In this case, shouldn't bit 1 remain
>>> set, because at level 3, a page mapping is denoted by bits[1:0] = 3 ? Currently
>>> its being unconditionally cleared.
>>
>> That's not a problem, pte_mkcont() brings back both the bits
>> via PTE_TYPE_PAGE along with PTE_CONT.
>>
>>         if (pagesize == CONT_PTE_SIZE) {
>>                 entry = pte_mkcont(entry);
>>         } else if (pagesize == CONT_PMD_SIZE) {
>>                 entry = pmd_pte(pmd_mkcont(pte_pmd(entry)));
>>         } else if (pagesize != PUD_SIZE && pagesize != PMD_SIZE) {
>>                 pr_warn("%s: unrecognized huge page size 0x%lx\n",
>>                         __func__, pagesize);
>>         }
>>
>> static inline pte_t pte_mkcont(pte_t pte)
>> {
>> 	pte = set_pte_bit(pte, __pgprot(PTE_CONT));
>>       	return set_pte_bit(pte, __pgprot(PTE_TYPE_PAGE));
> 
> Oh wow, that's pretty hacky. Good job we never call pte_mkcont() on a
> PTE_PRESENT_INVALID PTE. This would turn it valid again.

Ideally each individual HW page table helper should not do more than one thing
at a time. Here pte_mkcont() should just take a pte which is already a valid
one with PTE_TYPE_PAGE and just set PTE_CONT.

> 
>> }
>>
>> Although the same is not required for CONT_PMD_SIZE size huge
>> pages where only PTE_CONT is enough.
>>
>> static inline pmd_t pmd_mkcont(pmd_t pmd)
>> {
>>         return __pmd(pmd_val(pmd) | PMD_SECT_CONT);
>> }
>>  
>>>>  	if (pagesize == CONT_PTE_SIZE) {
>>>>  		entry = pte_mkcont(entry);
>>>>  	} else if (pagesize == CONT_PMD_SIZE) {
>>>
>