This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
More SJLJ exception breakage
- From: "Ulrich Weigand" <uweigand at de dot ibm dot com>
- To: gcc-patches at gcc dot gnu dot org
- Cc: hubicka at ucw dot cz
- Date: Thu, 16 Apr 2009 20:24:00 +0200 (CEST)
- Subject: More SJLJ exception breakage
Hello,
it seems SJLJ exception handling is broken again (at least on SPU). The
most obvious symptom is that libstdc++ "operator new" is miscompiled.
A reduced test case would be:
typedef long unsigned int size_t;
extern "C" void * malloc (size_t __size);
class bad_alloc
{
public:
bad_alloc() throw() { }
virtual ~bad_alloc() throw();
virtual const char* what() const throw();
};
void *
operator new (size_t sz) throw (bad_alloc)
{
void *p;
p = (void *) malloc (sz);
if (p == 0)
throw bad_alloc();
return p;
}
Building this with -O2 results in:
brsl $lr,malloc ;# 7 _call_value/2 [length = 4]
brz $3,.L8 ;# 10 *spu.md:3590 [length = 4]
#note result of malloc is in register $r3 (the branch goes to the throw case)
ai $3,$sp,48 ;# 87 addsi3/2 [length = 4]
brsl $lr,_Unwind_SjLj_Unregister ;# 88 _call/2 [length = 4]
#which is clobbered here with the argument passed to _Unwind_SjLj_Unregister
ai $sp,$sp,176 ;# 160 addsi3/2 [length = 4]
lqd $lr,16($sp) ;# 162 _movv4si/5 [length = 4]
lqd $127,-16($sp) ;# 161 _movv4si/5 [length = 4]
bi $lr ;# 164 _return [length = 4]
#and never restored on its way out
The RTL is broken already in ...133r.eh:
(code_label 32 31 41 10 1 "" [1 uses])
(note 41 32 33 10 [bb 10] NOTE_INSN_BASIC_BLOCK)
(insn 33 41 39 10 new_op.ii:23 (set (reg/i:SI 3 $3)
(reg:SI 138 [ <result> ])) -1 (nil))
(insn 39 33 86 10 new_op.ii:23 (use (reg/i:SI 3 $3)) -1 (nil))
(insn 86 39 87 10 (set (reg:SI 186)
(plus:SI (reg/f:SI 132 virtual-stack-vars)
(const_int -112 [0xffffffffffffff90]))) -1 (nil))
(insn 87 86 88 10 (set (reg:SI 3 $3)
(reg:SI 186)) -1 (nil))
(call_insn 88 87 0 10 (parallel [
(call (mem:QI (symbol_ref:SI ("_Unwind_SjLj_Unregister") [flags 0x41]) [0 S1 A8])
(const_int 0 [0x0]))
(clobber (reg:SI 0 $lr))
(clobber (reg:SI 130 hbr))
]) -1 (expr_list:REG_EH_REGION (const_int 0 [0x0])
(nil))
(expr_list:REG_DEP_TRUE (use (reg:SI 3 $3))
(nil)))
It would appear the call to _Unwind_SjLj_Unregister is simply inserted
too late; it needs to take place *before* the result register is set up
(insn 33), not afterwards.
Any idea why this might happen or suggestions what else to check?
Thanks,
Ulrich
--
Dr. Ulrich Weigand
GNU Toolchain for Linux on System z and Cell BE
Ulrich.Weigand@de.ibm.com