Re: [PATCH 1/6] arm64: use hw_pte_t for HW PTE pointers

[Muhammad Usama Anjum <usama.anjum@arm.com>]

On 21/09/2026 8:59 am, Ryan Roberts wrote:
> On 14/09/2026 14:51, Muhammad Usama Anjum wrote:
>> With pte_t * used for HW PTE pointers, the compiler cannot distinguish
>> them from SW PTE value pointers. The generic HW PTE interfaces already
>> use hw_pte_t *, so arm64 HW PTE pointers must use that type before
>> ARCH_HAS_HW_PTE_T can make it distinct.
>>
>> Convert HW PTE pointers to hw_pte_t *, retaining pte_t for SW PTE
>> values. Helpers shared with PMD and PUD operations keep explicit casts
>> because this series does not introduce distinct entry types for those
>> levels.
>>
>> arm64 still uses the hw_pte_t alias in this patch. The new pointer
>> spelling therefore denotes the same type until the final opt-in.
>>
>> Include linux/pgtable_types.h in pi.h to make hw_pte_t visible.
>>
>> Signed-off-by: Muhammad Usama Anjum <usama.anjum@arm.com>
>> ---
>>  arch/arm64/include/asm/hugetlb.h  |  19 ++---
>>  arch/arm64/include/asm/pgalloc.h  |   2 +-
>>  arch/arm64/include/asm/pgtable.h  | 143 ++++++++++++++++++++------------------
>>  arch/arm64/include/asm/vmalloc.h  |   2 +-
>>  arch/arm64/kernel/efi.c           |   3 +-
>>  arch/arm64/kernel/pi/map_kernel.c |   4 +-
>>  arch/arm64/kernel/pi/map_range.c  |   9 +--
>>  arch/arm64/kernel/pi/pi.h         |   3 +-
>>  arch/arm64/mm/contpte.c           |  49 +++++++------
>>  arch/arm64/mm/fault.c             |   5 +-
>>  arch/arm64/mm/fixmap.c            |   6 +-
>>  arch/arm64/mm/hugetlbpage.c       |  48 +++++++------
>>  arch/arm64/mm/kasan_init.c        |   4 +-
>>  arch/arm64/mm/mmu.c               |  46 +++++++-----
>>  arch/arm64/mm/pageattr.c          |   4 +-
>>  arch/arm64/mm/trans_pgd.c         |   4 +-
>>  16 files changed, 190 insertions(+), 161 deletions(-)
>>
> 
> [...]
> 
>> diff --git a/arch/arm64/mm/fixmap.c b/arch/arm64/mm/fixmap.c
>> index f66a0016dd02d..237a9136bc73b 100644
>> --- a/arch/arm64/mm/fixmap.c
>> +++ b/arch/arm64/mm/fixmap.c
>> @@ -35,7 +35,7 @@ static pte_t bm_pte[NR_BM_PTE_TABLES][PTRS_PER_PTE] __bss_pgtbl;
>>  static pmd_t bm_pmd[PTRS_PER_PMD] __bss_pgtbl __maybe_unused;
>>  static pud_t bm_pud[PTRS_PER_PUD] __bss_pgtbl __maybe_unused;
>>  
>> -static inline pte_t *fixmap_pte(unsigned long addr)
>> +static inline hw_pte_t *fixmap_pte(unsigned long addr)
>>  {
>>  	return &bm_pte[BM_PTE_TABLE_IDX(addr)][pte_index(addr)];
> 
> How does this work, given you haven't yet converted bm_pte? You're returning a
> pte_t* where you need a hw_pte_t*. Either the compiler is implicitly converting
> (which would defeat the whole purpose of this series) or this patch doesn't
> compile until you convert bm_pte later on (which I suspect is more likely).

At this point, hw_pte_t is still a preprocessor alias for pte_t:

    #define hw_pte_t pte_t

After preprocessing, both expressions therefore have type pte_t *; no
implicit pointer conversion is involved.

Patch 6/6 then selects ARCH_HAS_HW_PTE_T, replacing the alias with the
distinct wrapper type. This ordering keeps every intermediate patch
type-correct and enables compiler enforcement only after the pointers and
storage have both been converted.

> 
>>  }
>> @@ -43,7 +43,7 @@ static inline pte_t *fixmap_pte(unsigned long addr)
>>  static void __init early_fixmap_init_pte(pmd_t *pmdp, unsigned long addr)
>>  {
>>  	pmd_t pmd = READ_ONCE(*pmdp);
>> -	pte_t *ptep;
>> +	hw_pte_t *ptep;
>>  
>>  	if (pmd_none(pmd)) {
>>  		ptep = bm_pte[BM_PTE_TABLE_IDX(addr)];
> 
> Same problem here?

The same applies here.

-- 
Thanks,
Usama