Re: [PATCH 6.12.y 3/3] s390/mm: Fix __ptep_rdp() inline assembly

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

On Mon, Nov 24, 2025 at 12:17:19PM -0500, Sasha Levin wrote:
> From: Heiko Carstens <hca@linux.ibm.com>
> 
> [ Upstream commit 31475b88110c4725b4f9a79c3a0d9bbf97e69e1c ]
> 
> When a zero ASCE is passed to the __ptep_rdp() inline assembly, the
> generated instruction should have the R3 field of the instruction set to
> zero. However the inline assembly is written incorrectly: for such cases a
> zero is loaded into a register allocated by the compiler and this register
> is then used by the instruction.
> 
> This means that selected TLB entries may not be flushed since the specified
> ASCE does not match the one which was used when the selected TLB entries
> were created.
> 
> Fix this by removing the asce and opt parameters of __ptep_rdp(), since
> all callers always pass zero, and use a hard-coded register zero for
> the R3 field.
> 
> Fixes: 0807b856521f ("s390/mm: add support for RDP (Reset DAT-Protection)")
> Cc: stable@vger.kernel.org
> Reviewed-by: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
> Signed-off-by: Sasha Levin <sashal@kernel.org>
> ---
>  arch/s390/include/asm/pgtable.h | 12 +++++-------
>  arch/s390/mm/pgtable.c          |  4 ++--
>  2 files changed, 7 insertions(+), 9 deletions(-)

...

> @@ -1304,7 +1302,7 @@ static inline void flush_tlb_fix_spurious_fault(struct vm_area_struct *vma,
>  	 * A local RDP can be used to do the flush.
>  	 */
>  	if (cpu_has_rdp() && !(pte_val(*ptep) & _PAGE_PROTECT))
> -		__ptep_rdp(address, ptep, 0, 0, 1);
> +		__ptep_rdp(address, ptep, 1);
>  }

I don't think it makes too much sense to backport only two of the many
cpu_has_xxx() conversion patches just to avoid the minimal difference
in context for this patch. From my point of view this puts the stable
branch into an inconsistent state wrt s390 and cpu features - old and
new interfaces are mixed.

I will provide a different stable backport patch which addresses only
the context diff. Then you can decide.