Created attachment 1374985 [details]
unwind.i cleaned up for GCC 4.8

This also seems to affect gcc-4.8.5-25.el7.x86_64.  Based on the disassembly of /lib/libpthread.so.0, glibc-2.17-220.el7.i686 is miscompiled as well.

+++ This bug was initially created as a clone of Bug #1529979 +++

Description of problem:

The attached program, when compiled using

gcc -fPIC -O2 -fgnu89-inline-fstack-clash-protection -c unwind.i

produces invalid CFI instructions for __GI___pthread_unwind:

__GI___pthread_unwind:
.LFB55:
        .cfi_startproc
        pushl   %esi
        .cfi_def_cfa_offset 8
        .cfi_offset 6, -8
        pushl   %ebx
        .cfi_def_cfa_offset 12
        .cfi_offset 3, -12
        pushl   %esi
        .cfi_def_cfa_offset 16
        .cfi_offset 6, -16
        popl    %esi
        .cfi_restore 6
        .cfi_def_cfa_offset 12
        call    __x86.get_pc_thunk.bx
        addl    $_GLOBAL_OFFSET_TABLE_, %ebx
        subl    $8, %esp
        .cfi_def_cfa_offset 20
        movl    %eax, %esi

The problem is that the .cfi_restore for %esi does not restore the original location at offset 8, but essentially declares that the register will not be clobbered further below, while in fact it is used to store the function argument (__pthread_unwind is compiled with __attribute__ ((regparm (1)))).

As a result, the libgcc unwinder restores the wrong value for %esi, leading to bug 1529549.

Version-Release number of selected component (if applicable):

gcc-7.2.1-4.fc28.i686