powerpc: Incorrect stw operand modifier in __set_pte_at

powerpc: Incorrect stw operand modifier in __set_pte_at

[Mathieu Desnoyers]

Hi,

Reviewing use of the patterns "Un%Xn" with lwz and stw instructions
(where n should be the operand number) within the Linux kernel led
me to spot those 2 weird cases:

arch/powerpc/include/asm/nohash/pgtable.h:__set_pte_at()

                __asm__ __volatile__("\
                        stw%U0%X0 %2,%0\n\
                        eieio\n\
                        stw%U0%X0 %L2,%1"
                : "=m" (*ptep), "=m" (*((unsigned char *)ptep+4))
                : "r" (pte) : "memory");

I would have expected the stw to be:

                        stw%U1%X1 %L2,%1"

and:
arch/powerpc/include/asm/book3s/32/pgtable.h:__set_pte_at()

        __asm__ __volatile__("\
                stw%U0%X0 %2,%0\n\
                eieio\n\
                stw%U0%X0 %L2,%1"
        : "=m" (*ptep), "=m" (*((unsigned char *)ptep+4))
        : "r" (pte) : "memory");

where I would have expected:

                stw%U1%X1 %L2,%1"

Is it a bug or am I missing something ?

Thanks,

Mathieu

-- 
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com

Re: powerpc: Incorrect stw operand modifier in __set_pte_at

[Christophe Leroy]

Le 08/07/2020 à 16:45, Mathieu Desnoyers a écrit :
> Hi,
> 
> Reviewing use of the patterns "Un%Xn" with lwz and stw instructions
> (where n should be the operand number) within the Linux kernel led
> me to spot those 2 weird cases:
> 
> arch/powerpc/include/asm/nohash/pgtable.h:__set_pte_at()
> 
>                  __asm__ __volatile__("\
>                          stw%U0%X0 %2,%0\n\
>                          eieio\n\
>                          stw%U0%X0 %L2,%1"
>                  : "=m" (*ptep), "=m" (*((unsigned char *)ptep+4))
>                  : "r" (pte) : "memory");
> 
> I would have expected the stw to be:
> 
>                          stw%U1%X1 %L2,%1"
> 
> and:
> arch/powerpc/include/asm/book3s/32/pgtable.h:__set_pte_at()
> 
>          __asm__ __volatile__("\
>                  stw%U0%X0 %2,%0\n\
>                  eieio\n\
>                  stw%U0%X0 %L2,%1"
>          : "=m" (*ptep), "=m" (*((unsigned char *)ptep+4))
>          : "r" (pte) : "memory");
> 
> where I would have expected:
> 
>                  stw%U1%X1 %L2,%1"
> 
> Is it a bug or am I missing something ?

Well spotted. I guess it's definitly a bug.

Introduced 12 years ago by commit 
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=9bf2b5cd 
("powerpc: Fixes for CONFIG_PTE_64BIT for SMP support").

It's gone unnoticed until now it seems.

Can you submit a patch for it ?

Christophe

Re: powerpc: Incorrect stw operand modifier in __set_pte_at

[Segher Boessenkool]

On Wed, Jul 08, 2020 at 06:16:54PM +0200, Christophe Leroy wrote:
> Le 08/07/2020 à 16:45, Mathieu Desnoyers a écrit :
> >Reviewing use of the patterns "Un%Xn" with lwz and stw instructions
> >(where n should be the operand number) within the Linux kernel led
> >me to spot those 2 weird cases:
> >
> >arch/powerpc/include/asm/nohash/pgtable.h:__set_pte_at()
> >
> >                 __asm__ __volatile__("\
> >                         stw%U0%X0 %2,%0\n\
> >                         eieio\n\
> >                         stw%U0%X0 %L2,%1"
> >                 : "=m" (*ptep), "=m" (*((unsigned char *)ptep+4))
> >                 : "r" (pte) : "memory");
> >
> >I would have expected the stw to be:
> >
> >                         stw%U1%X1 %L2,%1"
> >
> >and:
> >arch/powerpc/include/asm/book3s/32/pgtable.h:__set_pte_at()
> >
> >         __asm__ __volatile__("\
> >                 stw%U0%X0 %2,%0\n\
> >                 eieio\n\
> >                 stw%U0%X0 %L2,%1"
> >         : "=m" (*ptep), "=m" (*((unsigned char *)ptep+4))
> >         : "r" (pte) : "memory");
> >
> >where I would have expected:
> >
> >                 stw%U1%X1 %L2,%1"
> >
> >Is it a bug or am I missing something ?
> 
> Well spotted. I guess it's definitly a bug.

Yes :-)

> Introduced 12 years ago by commit 
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=9bf2b5cd 
> ("powerpc: Fixes for CONFIG_PTE_64BIT for SMP support").
> 
> It's gone unnoticed until now it seems.

Apparently it always could use offset form memory accesses?  Or even
when not, %0 and %1 are likely to use the same base register for
addressing :-)


Segher