With GCC: (GNU) 5.0.0 20140917

give the following testcase


  typedef double t;

  void bar (t*);

  t g ()
  {
    t data[8192];

    data[4293] = data[4266] = 0;
    bar(data);
    return data[4293] + data[4266];
  }

Compiling with "-O2 -mno-lra -fomit-frame-pointer" we get:

        .cpu generic+fp+simd
        .file   "test.c"
        .text
        .align  2
        .global f
        .type   f, %function
f:
        add     x1, x0, 4093
        add     x0, x0, 4096
        ldr     d1, [x1]
        ldr     d0, [x0, 170]
        fadd    d0, d1, d0
        ret
        .size   f, .-f
        .align  2
        .global g
        .type   g, %function
g:
        sub     sp, sp, #65536
        fmov    d0, xzr
        str     x30, [sp, -16]!
        add     x1, sp, 32768
        add     x0, sp, 16
        str     d0, [x1, 1376]
        str     d0, [x1, 1592]
        bl      bar
        add     x0, sp, 32768
        ldr     x30, [sp], 16
        ldr     d0, [x0, 1376]
        add     sp, sp, 65536
        ldr     d1, [x0, 1592]
        fadd    d0, d1, d0
        ret

  Note that at the end we have x0=sp+32768; sp+=65536; ldr [x0+1592]
  Which means the last load is from deallocated stack space.

  This is a silent wrong-code bug of the worst kind; programs could fail  sporadically with this if an interrupt happens at the wrong instant in time and data was written onto the current stack.