Re: [PATCH v5 21/23] KVM: S390: Remove PGSTE code from linux/s390 mm

[Heiko Carstens <hca@linux.ibm.com>]

On Mon, Nov 24, 2025 at 12:55:52PM +0100, Claudio Imbrenda wrote:
> Remove the PGSTE config option.
> Remove all code from linux/s390 mm that involves PGSTEs.
> 
> Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
> ---
>  arch/s390/Kconfig               |   3 -
>  arch/s390/include/asm/mmu.h     |  13 -
>  arch/s390/include/asm/page.h    |   4 -
>  arch/s390/include/asm/pgalloc.h |   4 -
>  arch/s390/include/asm/pgtable.h | 121 +----
>  arch/s390/kvm/dat.h             |   1 +
>  arch/s390/mm/hugetlbpage.c      |  24 -
>  arch/s390/mm/pgalloc.c          |  24 -
>  arch/s390/mm/pgtable.c          | 829 +-------------------------------
>  mm/khugepaged.c                 |   9 -
>  10 files changed, 17 insertions(+), 1015 deletions(-)

...

>  pte_t ptep_modify_prot_start(struct vm_area_struct *vma, unsigned long addr,
>  			     pte_t *ptep)
>  {
> -	pgste_t pgste;
> -	pte_t old;
> -	int nodat;
> -	struct mm_struct *mm = vma->vm_mm;
> -
> -	pgste = ptep_xchg_start(mm, addr, ptep);
> -	nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
> -	old = ptep_flush_lazy(mm, addr, ptep, nodat);
> -	if (mm_has_pgste(mm)) {
> -		pgste = pgste_update_all(old, pgste, mm);
> -		pgste_set(ptep, pgste);
> -	}
> -	return old;
> +	preempt_disable();
> +	return ptep_flush_lazy(vma->vm_mm, addr, ptep, 1);
>  }
>  
>  void ptep_modify_prot_commit(struct vm_area_struct *vma, unsigned long addr,
>  			     pte_t *ptep, pte_t old_pte, pte_t pte)
>  {
> -	pgste_t pgste;
> -	struct mm_struct *mm = vma->vm_mm;
> -
> -	if (mm_has_pgste(mm)) {
> -		pgste = pgste_get(ptep);
> -		pgste_set_key(ptep, pgste, pte, mm);
> -		pgste = pgste_set_pte(ptep, pgste, pte);
> -		pgste_set_unlock(ptep, pgste);
> -	} else {
> -		set_pte(ptep, pte);
> -	}
> +	set_pte(ptep, pte);
> +	preempt_enable();
>  }

Why did you add the preempt_disable()/preempt_enable() pair?
This causes preempt_count overflows.

See modify_prot_start_ptes() + modify_prot_commit_ptes()...