[patch] prevent .cfi_personality/.cfi_lsda on !dwarf eh configurations

[patch] prevent .cfi_personality/.cfi_lsda on !dwarf eh configurations

[Olivier Hainque]

[-- Attachment #1: Type: text/plain, Size: 1013 bytes --]

Hello,

Bootstrap is currently broken with Ada on x86_64-windows using SEH exceptions,
with several stage2/stage3 object comparison failures like

  Bootstrap comparison failure!
  gcc/ada/ali-util.o differs
  gcc/ada/ali.o differs
  ...

(PR 81470)

The differences come from variations between -g and -g0 on the placement of
data attached to LDFCM labels, in particular those related to the emission of
.cfi_personality and .cfi_lsda by dwarf2out_do_cfi_startproc.

These aren't needed for SEH anyway, so the proposed patch just prevents the
generation of those .cfi notes if the target eh format isn't dwarf2.

We have used this for a couple of years on gcc-6, it cures the same
bootstrap failure on gcc-7, and bootstraps + regression-tests fine
with mainline on x86_64-linux.

OK to commit ?

2017-11-30  Tristan Gingold  <gingold@adacore.com>

        * dwarf2out.c (dwarf2out_do_cfi_startproc): Only emit
        .cfi_personality or .cfi_lsda if the eh data format is dwarf2.


[-- Attachment #2: cfi_startproc-nondwarf2.diff --]
[-- Type: application/octet-stream, Size: 682 bytes --]

diff --git a/gcc/dwarf2out.c b/gcc/dwarf2out.c
index 3d619b8..62b5c77 100644
--- a/gcc/dwarf2out.c
+++ b/gcc/dwarf2out.c
@@ -958,10 +958,16 @@ dwarf2out_do_cfi_startproc (bool second)
 {
   int enc;
   rtx ref;
-  rtx personality = get_personality_function (current_function_decl);
 
   fprintf (asm_out_file, "\t.cfi_startproc\n");
 
+  /* .cfi_personality and .cfi_lsda are only relevant to DWARF2
+     eh unwinders.  */
+  if (targetm_common.except_unwind_info (&global_options) != UI_DWARF2)
+    return;
+
+  rtx personality = get_personality_function (current_function_decl);
+
   if (personality)
     {
       enc = ASM_PREFERRED_EH_DATA_FORMAT (/*code=*/2, /*global=*/1);

Re: [patch] prevent .cfi_personality/.cfi_lsda on !dwarf eh configurations

[Eric Botcazou]

> 2017-11-30  Tristan Gingold  <gingold@adacore.com>
> 
>         * dwarf2out.c (dwarf2out_do_cfi_startproc): Only emit
>         .cfi_personality or .cfi_lsda if the eh data format is dwarf2.

PR ada/81470 must be put on the first line to trigger xref in Bugzilla.

-- 
Eric Botcazou

Re: [patch] prevent .cfi_personality/.cfi_lsda on !dwarf eh configurations

[Olivier Hainque]

> On 30 Nov 2017, at 13:04, Eric Botcazou <ebotcazou@adacore.com> wrote:
> 
>> 2017-11-30  Tristan Gingold  <gingold@adacore.com>
>> 
>>        * dwarf2out.c (dwarf2out_do_cfi_startproc): Only emit
>>        .cfi_personality or .cfi_lsda if the eh data format is dwarf2.
> 
> PR ada/81470 must be put on the first line to trigger xref in Bugzilla.

Oh, indeed. Will add if the patch gets approved. Thanks!

Re: [patch] prevent .cfi_personality/.cfi_lsda on !dwarf eh configurations

[Olivier Hainque]

Hello,

ping for https://gcc.gnu.org/ml/gcc-patches/2017-11/msg02537.html

> On Nov 30, 2017, at 10:18 , Olivier Hainque <hainque@adacore.com> wrote:
> 
> Bootstrap is currently broken with Ada on x86_64-windows using SEH exceptions,

>        PR ada/81470 

>        * dwarf2out.c (dwarf2out_do_cfi_startproc): Only emit
>        .cfi_personality or .cfi_lsda if the eh data format is dwarf2.

Thanks much in advance!

With Kind Regards,

Olivier

Re: [patch] prevent .cfi_personality/.cfi_lsda on !dwarf eh configurations

[Jeff Law]

On 11/30/2017 02:18 AM, Olivier Hainque wrote:
> Hello,
> 
> Bootstrap is currently broken with Ada on x86_64-windows using SEH exceptions,
> with several stage2/stage3 object comparison failures like
> 
>   Bootstrap comparison failure!
>   gcc/ada/ali-util.o differs
>   gcc/ada/ali.o differs
>   ...
> 
> (PR 81470)
> 
> The differences come from variations between -g and -g0 on the placement of
> data attached to LDFCM labels, in particular those related to the emission of
> .cfi_personality and .cfi_lsda by dwarf2out_do_cfi_startproc.
> 
> These aren't needed for SEH anyway, so the proposed patch just prevents the
> generation of those .cfi notes if the target eh format isn't dwarf2.
> 
> We have used this for a couple of years on gcc-6, it cures the same
> bootstrap failure on gcc-7, and bootstraps + regression-tests fine
> with mainline on x86_64-linux.
> 
> OK to commit ?
> 
> 2017-11-30  Tristan Gingold  <gingold@adacore.com>
> 
>         * dwarf2out.c (dwarf2out_do_cfi_startproc): Only emit
>         .cfi_personality or .cfi_lsda if the eh data format is dwarf2.
> 
OK.
jeff

Re: [patch] prevent .cfi_personality/.cfi_lsda on !dwarf eh configurations

[Olivier Hainque]

> On Dec 7, 2017, at 23:57 , Jeff Law <law@redhat.com> wrote:
> 
>>        * dwarf2out.c (dwarf2out_do_cfi_startproc): Only emit
>>        .cfi_personality or .cfi_lsda if the eh data format is dwarf2.
>> 
> OK.

Now in, thanks Jeff!