[Bug target/50678] [4.7 Regression] FAIL: c52104y on x86_64-apple-darwin10

["ebotcazou at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50678

--- Comment #30 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2011-10-15 20:49:03 UTC ---
> however I've not got far through Raise_From_Signal_Handler  () - if one
> continues from there it ends with a loop on x86-64/darwin9 and another segv on
> x86-64/darwin10.

You need to compile with -fstack-check.  Then you'll be able to debug the
unwinding phase.  Here are some possible breakpoints:

0x00007fff85b7a260 in unw_init_local () from /usr/lib/libSystem.B.dylib
(gdb) disass
Dump of assembler code for function unw_init_local:

Breakpoint 3, 0x00007fff85b9fe32 in unwind_phase2 ()
   from /usr/lib/libSystem.B.dylib
(gdb) disass
Dump of assembler code for function unwind_phase2:

0x00007fff85ba0150 in libunwind::Registers_x86_64::jumpto() ()
   from /usr/lib/libSystem.B.dylib
(gdb) disass
Dump of assembler code for function _ZN9libunwind16Registers_x86_646jumptoEv:
=> 0x00007fff85ba0150 <+0>:     mov    0x38(%rdi),%rax

The context from which jumpto restores the registers has the wrong %rbx line.

> The thing that's itching slightly is that there is a syscall (in
> __gnat_error_handler ) to switch the signal stack - before the raise, and I
> wonder if the use of the alt sigstack is what's causing the problem.

This call doesn't actually do anything.  The stack is switched by the system
before __gnat_error_handler is entered and all the processing is done using the
alternate stack, until execution is resumed in the exception handler.

> One would imagine that anything as radical as a missing reg. save/rest in eh
> would be spotted outside Ada ;-)

Java is the only other language that supports this kind of things though.