Bug#804536: gcc-5: [mips,mipsel] regression in optimization causes FTBFS for gsoap

Aurelien Jarno <[email protected]> Tue, 10 Nov 2015 13:22:42 +0100
Newsgroups	gmane.linux.debian.ports.mips
Message-ID	<[email protected]>
On 2015-11-09 11:53, Mattias Ellert wrote:
> Package: gcc-5
> Version: 5.2.1-23
> Severity: serious
> Justification: causes gsoap to FTBFS
> Control: affects -1 gsoap
> Control: block 804455 by -1
> X-Debbugs-Cc: [email protected]
> 
> Hi!
> 
> The binnmu of gsoap 2.8.22-1 due to the openssl transition failed on
> mips and mipsel, but succeeded on the other architectures.
> 
> https://buildd.debian.org/status/package.php?p=gsoap
> 
> (It also succeeded on mip64le - but the mips64el build used gcc-5
> 5.2.1-21 while mips and mipsel used 5.2.1-23. I am not sure if this is
> relevant information.)
> 
> The failure is a segmentation fault when running the soapcpp2 binary
> that has been compiled as part of the build. The soapcpp2 binary is not
> linked to openssl, so the issue is not due to the new openssl library
> that triggered the binnmu rebuild.
> 
> I can reproduce the failure on the eder.debian.org porterbox.
> 
> However, if I on the porterbox reduce the optimization from -O2 to -O1,
> the build succeeds. This therefore looks like a regression in
> mips/mipsel optimization.

The wrongly compiled code is in soapcpp2_yacc.y (generated from
soapcpp2.y) on line 2272 and following:

|   case 34:
| #line 436 "soapcpp2_yacc.y" /* yacc.c:1646  */
|     { if ((yyvsp[-1].rec).sto & Stypedef)
|                           {     sprintf(errbuf, "invalid typedef qualifier for '%s'", (yyvsp[0].sym)->name);
|                                 semwarn(errbuf);
|                           }
|                           printf("%p\n", (yyvsp[0].sym));
|                           p = enter(sp->table, (yyvsp[0].sym));
|                           p->info.typ = (yyvsp[-1].rec).typ;
|                           p->info.sto = (yyvsp[-1].rec).sto;
|                           p->info.hasval = False;
|                           p->info.offset = sp->offset;
|                           if (sp->grow)
|                                 sp->offset += p->info.typ->width;
|                           else if (p->info.typ->width > sp->offset)
|                                 sp->offset = p->info.typ->width;
|                           sp->entry = p;
|                         }
| #line 2290 "soapcpp2_yacc.c" /* yacc.c:1646  */
|     break;

The call to enter is translated into the following code:

| $L129:  
|         .loc 1 436 0
|         lw      $2,-60($20)
|         sw      $2,188($sp)
|         andi    $2,$2,0x10
|         bne     $2,$0,$L384
|         lui     $7,%hi($LC36)
| 
|         lw      $6,0($20)
| $LVL335 = .
| $L385:  
|         .loc 1 440 0
|         lw      $22,%got(sp)($28)
|         lw      $25,%call16(enter)($28)
|         lw      $2,0($22)
|         .reloc  1f,R_MIPS_JALR,enter
| 1:      jalr    $25
|         lw      $4,0($2)

Register a2 ($6) is loaded with (yyvsp[0].sym) instead of a1 ($5).

I still have to check if it is reproducible with trunk and to find the
optimization option causing the issue. Then we can report the bug
upstream.

Aurelien

-- 
Aurelien Jarno                          GPG: 4096R/1DDD8C9B
[email protected]                 http://www.aurel32.net
signature.asc (application/pgp-signature, 819 B)