https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87702
--- Comment #5 from Uroš Bizjak <ubizjak at gmail dot com> --- (In reply to Uroš Bizjak from comment #4) > Program received signal SIGSEGV, Segmentation fault. > 0xf7e24064 in __run_exit_handlers (status=1, listp=0xf7fca3fc > <__exit_funcs>, run_list_atexit=true, run_dtors=true) > at exit.c:114 > warning: Source file is more recent than executable. > 114 if (__glibc_unlikely (new_exitfn_called != > __new_exitfn_called)) > > (gdb) list > 109 break; > 110 } > 111 /* Re-lock again before looking at global state. */ > 112 __libc_lock_lock (__exit_funcs_lock); > 113 > 114 if (__glibc_unlikely (new_exitfn_called != > __new_exitfn_called)) > 115 /* The last exit function, or another thread, has > registered > 116 more exit functions. Start the loop over. */ > 117 goto restart; > 118 } Here is what happens: 0xf7e2405c <+188>: movq 0x3720(%ebp),%xmm0 2> 0xf7e24064 <+196>: movdqa (%esp),%xmm1 0xf7e24069 <+201>: pxor %xmm0,%xmm1 0xf7e2406d <+205>: movdqa %xmm1,%xmm0 0xf7e24071 <+209>: movd %xmm0,%ecx 0xf7e24075 <+213>: psrlq $0x20,%xmm0 0xf7e2407a <+218>: movd %xmm0,%eax 0xf7e2407e <+222>: or %ecx,%eax 0xf7e24080 <+224>: jne 0xf7e23ffe <__run_exit_handlers+94> 0xf7e24086 <+230>: mov 0x4(%edi),%ecx 0xf7e24089 <+233>: test %ecx,%ecx 0xf7e2408b <+235>: je 0xf7e24110 <__run_exit_handlers+368> 0xf7e24091 <+241>: lea -0x1(%ecx),%eax 0xf7e24094 <+244>: mov 0x3724(%ebp),%edx 0xf7e2409a <+250>: mov %eax,0x4(%edi) 0xf7e2409d <+253>: mov 0x3720(%ebp),%eax 1> 0xf7e240a3 <+259>: mov %edx,0x4(%esp) 0xf7e240a7 <+263>: mov %eax,(%esp) At 1>, new_exitfn_called is copied from __new_exitfn_called to a stack location. At point 2>, the (unaligned) location is accessed with full 128 bit movdqa access (which seems wrong, probably movq should be emitted here). Please provide a preprocessed source of exit.c (the file is called exit.i and can be generated by adding --save-temps to the compile command that compiled exit.c) from the build and the exact compile flags that produce the above assembly.
