stack alignment on i386

stack alignment on i386

[Mikulas Patocka]

>    [PATCH] x86: Drop cc-options call for all options supported in gcc 
>3.2+
>
>    The kernel only supports gcc 3.2+ now so it doesn't make sense
>    anymore to explicitely check for options this compiler version
>    already has.
>
>    This actually fixes a bug. The -mprefered-stack-boundary check
>    never worked because gcc rightly complains
>
>      CC      arch/i386/kernel/asm-offsets.s
>    cc1: -mpreferred-stack-boundary=2 is not between 4 and 12
>

Hi

Your patch for stack alignment on i386 is wrong. -mprefered-stack-boundary 
is log2 of the stack alignment (read gcc man page). On AMD64 16-byte 
alignment is enforced because of SSE, on i386 minimal alignment is 4 
bytes.

See the code in gcc that triggered your error:
       i = atoi (ix86_preferred_stack_boundary_string);
       if (i < (TARGET_64BIT ? 4 : 2) || i > 12)
         error ("-mpreferred-stack-boundary=%d is not between %d and 12", 
i,
                TARGET_64BIT ? 4 : 2);
       else
         ix86_preferred_stack_boundary = (1 << i) * BITS_PER_UNIT;

You are actually dealing with gcc bug, not Linux bug --- when gcc is 
compiled with AMD64 support it enforces 16-byte alignment even in 32-bit 
mode.

Rather let it be as it was, adding -mpreferred-stack-boundary=4 doesn't 
make sense, because it is default.

Mikulas

Signed-off-by: Mikulas Patocka <mikulas@artax.karlin.mff.cuni.cz>

--- linux-2.6.22-rc1/arch/i386/Makefile_	Thu May 17 23:47:31 2007
+++ linux-2.6.22-rc1/arch/i386/Makefile	Thu May 17 23:47:45 2007
@@ -34,7 +34,9 @@
  CFLAGS += -pipe -msoft-float -mregparm=3 -freg-struct-return

  # prevent gcc from keeping the stack 16 byte aligned
-CFLAGS += -mpreferred-stack-boundary=4
+# gcc has a bug that it doesn't allow 4-byte alignment when compiled with
+# 64-bit support and producing 32-bit code, so this option must be tested
+CFLAGS += $(call cc-option,-mpreferred-stack-boundary=2)

  # CPU-specific tuning. Anything which can be shared with UML should go here.
  include $(srctree)/arch/i386/Makefile.cpu

Re: stack alignment on i386

[Andi Kleen]

On Wednesday 16 May 2007 07:24, Mikulas Patocka wrote:
> >    [PATCH] x86: Drop cc-options call for all options supported in gcc
> >3.2+
> >
> >    The kernel only supports gcc 3.2+ now so it doesn't make sense
> >    anymore to explicitely check for options this compiler version
> >    already has.
> >
> >    This actually fixes a bug. The -mprefered-stack-boundary check
> >    never worked because gcc rightly complains
> >
> >      CC      arch/i386/kernel/asm-offsets.s
> >    cc1: -mpreferred-stack-boundary=2 is not between 4 and 12
>
> Hi
>
> Your patch for stack alignment on i386 is wrong. -mprefered-stack-boundary
> is log2 of the stack alignment (read gcc man page). 

Ah just using the real number would have been too easy I guess.

> On AMD64 16-byte 
> alignment is enforced because of SSE, on i386 minimal alignment is 4
> bytes.

Ok fixed now

> See the code in gcc that triggered your error:
>        i = atoi (ix86_preferred_stack_boundary_string);
>        if (i < (TARGET_64BIT ? 4 : 2) || i > 12)
>          error ("-mpreferred-stack-boundary=%d is not between %d and 12",
> i,
>                 TARGET_64BIT ? 4 : 2);
>        else
>          ix86_preferred_stack_boundary = (1 << i) * BITS_PER_UNIT;
>
> You are actually dealing with gcc bug, not Linux bug --- when gcc is
> compiled with AMD64 support it enforces 16-byte alignment even in 32-bit
> mode.

Honza? 

> Rather let it be as it was, adding -mpreferred-stack-boundary=4 doesn't
> make sense, because it is default.

Hmm, maybe that was different on older gcc versions. There must have been
some reason this option was originally added.

Linus can you please just revert  c8fdd247255a3a027cd9f66dcf93e6847d1d2f85

-Andi