| Issue |
208433
|
| Summary |
[lldb] lldb should reject invalid DW_OP_mod results instead of continuing with a void Scalar
|
| Labels |
new issue
|
| Assignees |
|
| Reporter |
firmiana402
|
LLDB's `DW_OP_mod` evaluator stores the result of `Scalar::operator%` back onto the DWARF stack without checking whether it is valid. When the modulo is undefined — for example a zero divisor, which the accepted [DWARF issue 250924.2](https://dwarfstd.org/issues/250924.2.html) clarifies should raise an error, like division — `operator%` returns a `Scalar` of type `e_void`, and evaluation continues with that invalid entry on the stack. Later opcodes can then consume or discard it (e.g. `DW_OP_drop`), so the invalid intermediate is silently absorbed instead of being reported where it arose.
## Source Evidence
`operator%` (in `Scalar.cpp`) returns a void `Scalar` when the modulo is undefined (zero divisor, or operands that do not promote to an integer):
```cpp
const Scalar lldb_private::operator%(Scalar lhs, Scalar rhs) {
Scalar result;
if ((result.m_type = Scalar::PromoteToMaxType(lhs, rhs)) != Scalar::e_void) {
if (!rhs.IsZero() && result.m_type == Scalar::e_int) {
result.m_integer = lhs.m_integer % rhs.m_integer;
return result;
}
}
result.m_type = Scalar::e_void;
return result;
}
```
The `DW_OP_mod` handler (in `DWARFExpression.cpp`) stores that result with no validity check:
```cpp
case DW_OP_mod:
tmp = stack.back();
stack.pop_back();
stack.back().GetScalar() = stack.back().GetScalar() % tmp.GetScalar();
break;
```
The neighbouring `DW_OP_div` handler, by contrast, guards both the zero divisor and the result validity:
```cpp
case DW_OP_div: {
tmp = stack.back();
if (tmp.GetScalar().IsZero())
return llvm::createStringError("divide by zero");
...
stack.back() = dividend / divisor;
if (!stack.back().GetScalar().IsValid())
return llvm::createStringError("divide failed");
} break;
```
`DW_OP_shr` and `DW_OP_plus_uconst` likewise return an evaluator error on failure. `DW_OP_mod` is the outlier that keeps the invalid `Scalar` and keeps going.
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
