Re: [PATCH] rs6000/test: Fix bswap64-4.c with has_arch_ppc64 [PR106680]

[Segher Boessenkool <segher@kernel.crashing.org>]

On Wed, Aug 31, 2022 at 04:38:02PM -0500, Peter Bergner wrote:
> On 8/31/22 4:07 PM, Segher Boessenkool wrote:
> > On Wed, Aug 31, 2022 at 02:53:07PM -0500, Peter Bergner wrote:
> >> Changing OS_MISSING_POWERPC64 as I mentioned would not add OPTION_MASK_POWERPC64
> >> to our cpu masks when -m32 is used.
> > 
> > So you say this is where the bug is?
> 
> For linux64.h which is what I think the powerpc64-linux build will use,
> we have:
> 
> linux64.h:#define OS_MISSING_POWERPC64 !TARGET_64BIT
> 
> Doing the macro expansion by hand into:
> 
>   set_masks = POWERPC_MASKS;
> #ifdef OS_MISSING_POWERPC64
>   if (OS_MISSING_POWERPC64)
>     set_masks &= ~OPTION_MASK_POWERPC64;
> #endif
> 
> 
> ...gives us:
> 
>   set_masks = POWERPC_MASKS;
>   if (!TARGET_64BIT)
>     set_masks &= ~OPTION_MASK_POWERPC64;
> 
> So if we handled a -mpowerpc64 earlier on the command line and added
> OPTION_MASK_POWERPC64 to our cpu mask, then a following -m32 use will
> remove it here.
> 
> So I mentioned doing:
> 
> linux64.h:
> - #define OS_MISSING_POWERPC64 !TARGET_64BIT
> + #define OS_MISSING_POWERPC64 0
> 
> ...which disables the above code only for powerpc64-linux builds and doesn't
> affect AIX, Darwin, BSD, etc. or a powerpc-linux build.

But it is incorrect as well.  Instead, we should look if -mpowerpc64 is
enabled explicitly, and not change it if so.

> > The kernel has.  But there are user space things (glibc) that haven't
> > been fixed, and those are default as well.
> 
> Sure, but someone who is using -m32 -mpowerpc64 should know that and
> relying on a 32-bit glibc to save/restore the full 64-bit registers
> is a user error in my book.  If you're using -m32 -mpower64, you
> better know what you are doing and the limitations you have to live under.

Of course.  But -mpowerpc64 can never be any kind of default for 32-bit
Linux.  Not indirectly either.


Segher