Re: [PATCH, PowerPC] Fix PR57949 (ABI alignment issue)

Re: [PATCH, PowerPC] Fix PR57949 (ABI alignment issue)

[Alan Modra]

On Wed, Aug 14, 2013 at 10:32:01AM -0500, Bill Schmidt wrote:
> This fixes a long-standing problem with GCC's implementation of the
> PPC64 ELF ABI.  If a structure contains a member requiring 128-bit
> alignment, and that structure is passed as a parameter, the parameter
> currently receives only 64-bit alignment.  This is an error, and is
> incompatible with correct code generated by the IBM XL compilers.

This caused multiple failures in the libffi testsuite:
libffi.call/cls_align_longdouble.c
libffi.call/cls_align_longdouble_split.c
libffi.call/cls_align_longdouble_split2.c
libffi.call/nested_struct5.c

Fixed by making the same alignment adjustment in libffi to structures
passed by value.  Bill, I think your patch needs to go on all active
gcc branches as otherwise we'll need different versions of libffi for
the next gcc releases.

The following was bootstrapped and regression checked powerpc64-linux.
OK for mainline, and the 4.7 and 4.8 branches when/if Bill's patch
goes in there?

	* src/powerpc/ffi.c (ffi_prep_args64): Align FFI_TYPE_STRUCT.
	(ffi_closure_helper_LINUX64): Likewise.

Index: libffi/src/powerpc/ffi.c
===================================================================
--- libffi/src/powerpc/ffi.c	(revision 202428)
+++ libffi/src/powerpc/ffi.c	(working copy)
@@ -462,6 +462,7 @@ ffi_prep_args64 (extended_cif *ecif, unsigned long
     double **d;
   } p_argv;
   unsigned long gprvalue;
+  unsigned long align;
 
   stacktop.c = (char *) stack + bytes;
   gpr_base.ul = stacktop.ul - ASM_NEEDS_REGISTERS64 - NUM_GPR_ARG_REGISTERS64;
@@ -532,6 +533,10 @@ ffi_prep_args64 (extended_cif *ecif, unsigned long
 #endif
 
 	case FFI_TYPE_STRUCT:
+	  align = (*ptr)->alignment;
+	  if (align > 16)
+	    align = 16;
+	  next_arg.ul = ALIGN (next_arg.ul, align);
 	  words = ((*ptr)->size + 7) / 8;
 	  if (next_arg.ul >= gpr_base.ul && next_arg.ul + words > gpr_end.ul)
 	    {
@@ -1349,6 +1354,7 @@ ffi_closure_helper_LINUX64 (ffi_closure *closure,
   long i, avn;
   ffi_cif *cif;
   ffi_dblfl *end_pfr = pfr + NUM_FPR_ARG_REGISTERS64;
+  unsigned long align;
 
   cif = closure->cif;
   avalue = alloca (cif->nargs * sizeof (void *));
@@ -1399,6 +1405,10 @@ ffi_closure_helper_LINUX64 (ffi_closure *closure,
 	  break;
 
 	case FFI_TYPE_STRUCT:
+	  align = arg_types[i]->alignment;
+	  if (align > 16)
+	    align = 16;
+	  pst = ALIGN (pst, align);
 #ifndef __LITTLE_ENDIAN__
 	  /* Structures with size less than eight bytes are passed
 	     left-padded.  */


-- 
Alan Modra
Australia Development Lab, IBM

Re: [PATCH, PowerPC] Fix PR57949 (ABI alignment issue)

[Jakub Jelinek]

On Wed, Sep 11, 2013 at 09:08:46PM +0930, Alan Modra wrote:
> The following was bootstrapped and regression checked powerpc64-linux.
> OK for mainline, and the 4.7 and 4.8 branches when/if Bill's patch
> goes in there?

IMHO ABI changing patches shouldn't be backported to release branches.

	Jakub

Re: [PATCH, PowerPC] Fix PR57949 (ABI alignment issue)

[Bill Schmidt]

On Wed, 2013-09-11 at 21:08 +0930, Alan Modra wrote:
> On Wed, Aug 14, 2013 at 10:32:01AM -0500, Bill Schmidt wrote:
> > This fixes a long-standing problem with GCC's implementation of the
> > PPC64 ELF ABI.  If a structure contains a member requiring 128-bit
> > alignment, and that structure is passed as a parameter, the parameter
> > currently receives only 64-bit alignment.  This is an error, and is
> > incompatible with correct code generated by the IBM XL compilers.
> 
> This caused multiple failures in the libffi testsuite:
> libffi.call/cls_align_longdouble.c
> libffi.call/cls_align_longdouble_split.c
> libffi.call/cls_align_longdouble_split2.c
> libffi.call/nested_struct5.c
> 
> Fixed by making the same alignment adjustment in libffi to structures
> passed by value.  Bill, I think your patch needs to go on all active
> gcc branches as otherwise we'll need different versions of libffi for
> the next gcc releases.

Hm, the libffi case is unfortunate. :(

The alternative is to leave libffi alone, and require code that calls
these interfaces with "bad" structs passed by value to be built using
-mcompat-align-parm, which was provided for such compatibility issues.
Hopefully there is a small number of cases where this can happen, and
this could be documented with libffi and gcc.  What do you think?

Thanks,
Bill

> 
> The following was bootstrapped and regression checked powerpc64-linux.
> OK for mainline, and the 4.7 and 4.8 branches when/if Bill's patch
> goes in there?
> 
> 	* src/powerpc/ffi.c (ffi_prep_args64): Align FFI_TYPE_STRUCT.
> 	(ffi_closure_helper_LINUX64): Likewise.
> 
> Index: libffi/src/powerpc/ffi.c
> ===================================================================
> --- libffi/src/powerpc/ffi.c	(revision 202428)
> +++ libffi/src/powerpc/ffi.c	(working copy)
> @@ -462,6 +462,7 @@ ffi_prep_args64 (extended_cif *ecif, unsigned long
>      double **d;
>    } p_argv;
>    unsigned long gprvalue;
> +  unsigned long align;
> 
>    stacktop.c = (char *) stack + bytes;
>    gpr_base.ul = stacktop.ul - ASM_NEEDS_REGISTERS64 - NUM_GPR_ARG_REGISTERS64;
> @@ -532,6 +533,10 @@ ffi_prep_args64 (extended_cif *ecif, unsigned long
>  #endif
> 
>  	case FFI_TYPE_STRUCT:
> +	  align = (*ptr)->alignment;
> +	  if (align > 16)
> +	    align = 16;
> +	  next_arg.ul = ALIGN (next_arg.ul, align);
>  	  words = ((*ptr)->size + 7) / 8;
>  	  if (next_arg.ul >= gpr_base.ul && next_arg.ul + words > gpr_end.ul)
>  	    {
> @@ -1349,6 +1354,7 @@ ffi_closure_helper_LINUX64 (ffi_closure *closure,
>    long i, avn;
>    ffi_cif *cif;
>    ffi_dblfl *end_pfr = pfr + NUM_FPR_ARG_REGISTERS64;
> +  unsigned long align;
> 
>    cif = closure->cif;
>    avalue = alloca (cif->nargs * sizeof (void *));
> @@ -1399,6 +1405,10 @@ ffi_closure_helper_LINUX64 (ffi_closure *closure,
>  	  break;
> 
>  	case FFI_TYPE_STRUCT:
> +	  align = arg_types[i]->alignment;
> +	  if (align > 16)
> +	    align = 16;
> +	  pst = ALIGN (pst, align);
>  #ifndef __LITTLE_ENDIAN__
>  	  /* Structures with size less than eight bytes are passed
>  	     left-padded.  */
> 
> 

Re: [PATCH, PowerPC] Fix PR57949 (ABI alignment issue)

[Jay]

Isn't mixing and matching and mismatching somewhat inevitable? Libffi & gcc don't always come along with each other? One must never change the ABI?

 - Jay

On Sep 11, 2013, at 5:55 AM, Bill Schmidt <wschmidt@linux.vnet.ibm.com> wrote:

> On Wed, 2013-09-11 at 21:08 +0930, Alan Modra wrote:
>> On Wed, Aug 14, 2013 at 10:32:01AM -0500, Bill Schmidt wrote:
>>> This fixes a long-standing problem with GCC's implementation of the
>>> PPC64 ELF ABI.  If a structure contains a member requiring 128-bit
>>> alignment, and that structure is passed as a parameter, the parameter
>>> currently receives only 64-bit alignment.  This is an error, and is
>>> incompatible with correct code generated by the IBM XL compilers.
>> 
>> This caused multiple failures in the libffi testsuite:
>> libffi.call/cls_align_longdouble.c
>> libffi.call/cls_align_longdouble_split.c
>> libffi.call/cls_align_longdouble_split2.c
>> libffi.call/nested_struct5.c
>> 
>> Fixed by making the same alignment adjustment in libffi to structures
>> passed by value.  Bill, I think your patch needs to go on all active
>> gcc branches as otherwise we'll need different versions of libffi for
>> the next gcc releases.
> 
> Hm, the libffi case is unfortunate. :(
> 
> The alternative is to leave libffi alone, and require code that calls
> these interfaces with "bad" structs passed by value to be built using
> -mcompat-align-parm, which was provided for such compatibility issues.
> Hopefully there is a small number of cases where this can happen, and
> this could be documented with libffi and gcc.  What do you think?
> 
> Thanks,
> Bill
> 
>> 
>> The following was bootstrapped and regression checked powerpc64-linux.
>> OK for mainline, and the 4.7 and 4.8 branches when/if Bill's patch
>> goes in there?
>> 
>>    * src/powerpc/ffi.c (ffi_prep_args64): Align FFI_TYPE_STRUCT.
>>    (ffi_closure_helper_LINUX64): Likewise.
>> 
>> Index: libffi/src/powerpc/ffi.c
>> ===================================================================
>> --- libffi/src/powerpc/ffi.c    (revision 202428)
>> +++ libffi/src/powerpc/ffi.c    (working copy)
>> @@ -462,6 +462,7 @@ ffi_prep_args64 (extended_cif *ecif, unsigned long
>>     double **d;
>>   } p_argv;
>>   unsigned long gprvalue;
>> +  unsigned long align;
>> 
>>   stacktop.c = (char *) stack + bytes;
>>   gpr_base.ul = stacktop.ul - ASM_NEEDS_REGISTERS64 - NUM_GPR_ARG_REGISTERS64;
>> @@ -532,6 +533,10 @@ ffi_prep_args64 (extended_cif *ecif, unsigned long
>> #endif
>> 
>>    case FFI_TYPE_STRUCT:
>> +      align = (*ptr)->alignment;
>> +      if (align > 16)
>> +        align = 16;
>> +      next_arg.ul = ALIGN (next_arg.ul, align);
>>      words = ((*ptr)->size + 7) / 8;
>>      if (next_arg.ul >= gpr_base.ul && next_arg.ul + words > gpr_end.ul)
>>        {
>> @@ -1349,6 +1354,7 @@ ffi_closure_helper_LINUX64 (ffi_closure *closure,
>>   long i, avn;
>>   ffi_cif *cif;
>>   ffi_dblfl *end_pfr = pfr + NUM_FPR_ARG_REGISTERS64;
>> +  unsigned long align;
>> 
>>   cif = closure->cif;
>>   avalue = alloca (cif->nargs * sizeof (void *));
>> @@ -1399,6 +1405,10 @@ ffi_closure_helper_LINUX64 (ffi_closure *closure,
>>      break;
>> 
>>    case FFI_TYPE_STRUCT:
>> +      align = arg_types[i]->alignment;
>> +      if (align > 16)
>> +        align = 16;
>> +      pst = ALIGN (pst, align);
>> #ifndef __LITTLE_ENDIAN__
>>      /* Structures with size less than eight bytes are passed
>>         left-padded.  */
> 

Re: [PATCH, PowerPC] Fix PR57949 (ABI alignment issue)

[Alan Modra]

On Wed, Sep 11, 2013 at 07:55:43AM -0500, Bill Schmidt wrote:
> On Wed, 2013-09-11 at 21:08 +0930, Alan Modra wrote:
> > On Wed, Aug 14, 2013 at 10:32:01AM -0500, Bill Schmidt wrote:
> > > This fixes a long-standing problem with GCC's implementation of the
> > > PPC64 ELF ABI.  If a structure contains a member requiring 128-bit
> > > alignment, and that structure is passed as a parameter, the parameter
> > > currently receives only 64-bit alignment.  This is an error, and is
> > > incompatible with correct code generated by the IBM XL compilers.
> > 
> > This caused multiple failures in the libffi testsuite:
> > libffi.call/cls_align_longdouble.c
> > libffi.call/cls_align_longdouble_split.c
> > libffi.call/cls_align_longdouble_split2.c
> > libffi.call/nested_struct5.c
> > 
> > Fixed by making the same alignment adjustment in libffi to structures
> > passed by value.  Bill, I think your patch needs to go on all active
> > gcc branches as otherwise we'll need different versions of libffi for
> > the next gcc releases.
> 
> Hm, the libffi case is unfortunate. :(
> 
> The alternative is to leave libffi alone, and require code that calls
> these interfaces with "bad" structs passed by value to be built using
> -mcompat-align-parm, which was provided for such compatibility issues.
> Hopefully there is a small number of cases where this can happen, and
> this could be documented with libffi and gcc.  What do you think?

We have precedent for compiling libffi based on gcc preprocessor
defines, eg. __NO_FPRS__, so here's a way of making upstream libffi
compatible with the various versions of gcc out there.  I've taken the
condition under which we align aggregates from
rs6000_function_arg_boundary, and defined a macro with a value of the
maximum alignment.

Bootstrapped and regression tested powerpc64-linux.  OK for mainline?

gcc/
	* config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): Define
	__STRUCT_PARM_ALIGN__.
libffi/
	* src/powerpc/ffi.c (ffi_prep_args64): Align FFI_TYPE_STRUCT.
	(ffi_closure_helper_LINUX64): Likewise.

Index: gcc/config/rs6000/rs6000-c.c
===================================================================
--- gcc/config/rs6000/rs6000-c.c	(revision 202428)
+++ gcc/config/rs6000/rs6000-c.c	(working copy)
@@ -473,6 +473,12 @@ rs6000_cpu_cpp_builtins (cpp_reader *pfile)
   if (TARGET_SOFT_FLOAT || !TARGET_FPRS)
     builtin_define ("__NO_FPRS__");
 
+  /* Whether aggregates passed by value are aligned to a 16 byte boundary
+     if their alignment is 16 bytes or larger.  */
+  if ((TARGET_MACHO && rs6000_darwin64_abi)
+      || (DEFAULT_ABI == ABI_AIX && !rs6000_compat_align_parm))
+    builtin_define ("__STRUCT_PARM_ALIGN__=16");
+
   /* Generate defines for Xilinx FPU. */
   if (rs6000_xilinx_fpu) 
     {
Index: libffi/src/powerpc/ffi.c
===================================================================
--- libffi/src/powerpc/ffi.c	(revision 202428)
+++ libffi/src/powerpc/ffi.c	(working copy)
@@ -462,6 +462,9 @@ ffi_prep_args64 (extended_cif *ecif, unsigned long
     double **d;
   } p_argv;
   unsigned long gprvalue;
+#ifdef __STRUCT_PARM_ALIGN__
+  unsigned long align;
+#endif
 
   stacktop.c = (char *) stack + bytes;
   gpr_base.ul = stacktop.ul - ASM_NEEDS_REGISTERS64 - NUM_GPR_ARG_REGISTERS64;
@@ -532,6 +535,12 @@ ffi_prep_args64 (extended_cif *ecif, unsigned long
 #endif
 
 	case FFI_TYPE_STRUCT:
+#ifdef __STRUCT_PARM_ALIGN__
+	  align = (*ptr)->alignment;
+	  if (align > __STRUCT_PARM_ALIGN__)
+	    align = __STRUCT_PARM_ALIGN__;
+	  next_arg.ul = ALIGN (next_arg.ul, align);
+#endif
 	  words = ((*ptr)->size + 7) / 8;
 	  if (next_arg.ul >= gpr_base.ul && next_arg.ul + words > gpr_end.ul)
 	    {
@@ -1349,6 +1358,9 @@ ffi_closure_helper_LINUX64 (ffi_closure *closure,
   long i, avn;
   ffi_cif *cif;
   ffi_dblfl *end_pfr = pfr + NUM_FPR_ARG_REGISTERS64;
+#ifdef __STRUCT_PARM_ALIGN__
+  unsigned long align;
+#endif
 
   cif = closure->cif;
   avalue = alloca (cif->nargs * sizeof (void *));
@@ -1399,6 +1411,12 @@ ffi_closure_helper_LINUX64 (ffi_closure *closure,
 	  break;
 
 	case FFI_TYPE_STRUCT:
+#ifdef __STRUCT_PARM_ALIGN__
+	  align = arg_types[i]->alignment;
+	  if (align > __STRUCT_PARM_ALIGN__)
+	    align = __STRUCT_PARM_ALIGN__;
+	  pst = ALIGN (pst, align);
+#endif
 #ifndef __LITTLE_ENDIAN__
 	  /* Structures with size less than eight bytes are passed
 	     left-padded.  */

-- 
Alan Modra
Australia Development Lab, IBM

Re: [PATCH, PowerPC] Fix PR57949 (ABI alignment issue)

[Andrew Haley]

On 09/12/2013 03:11 AM, Alan Modra wrote:
> We have precedent for compiling libffi based on gcc preprocessor
> defines, eg. __NO_FPRS__, so here's a way of making upstream libffi
> compatible with the various versions of gcc out there.  I've taken the
> condition under which we align aggregates from
> rs6000_function_arg_boundary, and defined a macro with a value of the
> maximum alignment.

I would have thought that a runtime or configure-time test in libffi
would be better.

Andrew.

Re: [PATCH, PowerPC] Fix PR57949 (ABI alignment issue)

[Alan Modra]

On Thu, Sep 12, 2013 at 09:33:05AM +0100, Andrew Haley wrote:
> On 09/12/2013 03:11 AM, Alan Modra wrote:
> > We have precedent for compiling libffi based on gcc preprocessor
> > defines, eg. __NO_FPRS__, so here's a way of making upstream libffi
> > compatible with the various versions of gcc out there.  I've taken the
> > condition under which we align aggregates from
> > rs6000_function_arg_boundary, and defined a macro with a value of the
> > maximum alignment.
> 
> I would have thought that a runtime or configure-time test in libffi
> would be better.

I don't see how a runtime check can be done, and a configure test just
adds another layer of indirection.  How is testing a cpp macro at
configure time better than testing the same at compile time?

-- 
Alan Modra
Australia Development Lab, IBM