[Bug gold/25034] New: gold creates incorrect executable with large arrays

[greened at obbligato dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=25034

            Bug ID: 25034
           Summary: gold creates incorrect executable with large arrays
           Product: binutils
           Version: 2.32
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: gold
          Assignee: ccoutant at gmail dot com
          Reporter: greened at obbligato dot org
                CC: ian at airs dot com
  Target Milestone: ---

$ cat big.c
#include <stdio.h>

struct {
  double a[512][512][512];
  double b[512][512][512];
  double c[512][512][512];
} bigarrays;

int x;

int main(void) {
  printf("&a: 0x%016x\n", bigarrays.a);
  printf("&b: 0x%016x\n", bigarrays.b);
  printf("&c: 0x%016x\n", bigarrays.c);
  printf("&x: 0x%016x\n", &x);

  x = 20;

  int i = 0;
  int j = 0;
  int k = 0;

  for (k = 0; k < 512; ++k) {
    for (j = 0; j < 512; ++j) {
      for (i = 0; i < 512; ++i) {
        bigarrays.a[k][j][i] = 0.0;
        bigarrays.b[k][j][i] = 0.0;
        bigarrays.c[k][j][i] = 0.0;
      }
    }
  }

  printf("x: %d\n", x);

  return 0;
}

$ gcc -fPIC big.c -Wl,--no-relax -o big.bfd
$ ./big.bfd
&a: 0x0000000000404080
&b: 0x0000000040404080
&c: 0x0000000080404080
&x: 0x0000000000404060
x: 20

$ gcc -fuse-ld=gold -fPIC big.c -o big.gold
$ ./big.gold 
&a: 0x0000000000402060
&b: 0x0000000040402060
&c: 0x0000000080402060
&x: 0x00000000c0402060
Segmentation fault

$ ld --version
GNU ld (GNU Binutils) 2.32
Copyright (C) 2019 Free Software Foundation, Inc.
This program is free software; you may redistribute it under the terms of
the GNU General Public License version 3 or (at your option) a later version.
This program has absolutely no warranty.

$ ld.gold --version
GNU gold (GNU Binutils 2.32) 1.16
Copyright (C) 2019 Free Software Foundation, Inc.
This program is free software; you may redistribute it under the terms of
the GNU General Public License version 3 or (at your option) a later version.
This program has absolutely no warranty.

gdb shows the fault here:

   0x0000000000400610 <+126>:   lea    -0x3fffe5b7(%rip),%rax        #
0xffffffffc0402060
=> 0x0000000000400617 <+133>:   movl   $0x14,(%rax)

That offset looks like a relocation overflow.  Is gold trying to relax the
reference and doing it incorrectly?  `--no-relax` doesn't help.  The code looks
the same with and without the option.

-- 
You are receiving this mail because:
You are on the CC list for the bug.