[LLVMdev] Info on byval attributes

[John Criswell]

On 5/12/12 6:17 PM, Santosh Nagarakatte wrote:
> LLVM developers,
>
>
> I was wondering if the program would still be safe if I strip the
> byval attributes from the parameters in the entire bitcode.

Can you explain why you want to remove the byval attribute?  It is far 
better if you can handle code with the byval argument instead of trying 
to strip it away.

If it's for memory safety checking in SoftBound, it might help to know 
that SAFECode treats byval arguments as local alloca's in the callee; it 
registers the bounds of the byval argument on function entry.

>
> LLVM language reference manual states that "The attribute implies that
> a hidden copy of the pointee is made between the caller and the
> callee, so the callee is unable to modify the value in the callee.
> This attribute is only valid on LLVM pointer arguments."
>
> Is there a way I can avoid the byval attribute by a custom
> transformation? I was thinking of a custom transformation that
> introduces an alloca and maintains the hidden copy of the pointee?

I would think that you could have either the caller or callee make a 
copy of the pointed-to memory object.  However, there are some gotchas 
to consider:

1) You will probably need to remove byval from both call sites and 
function arguments.  I'm not sure how the code generator inserts the 
hidden alloca and copy, and it wouldn't surprise me if different code 
generators do it differently.

2) You need to handle indirect call sites.

3) You may have ABI compliance issues with functions that escape to 
external code (either because they have externally visible linkage or 
because their address is taken and passed to external code).

Whatever your project is, supporting byval is probably a more robust 
solution than trying to remove byval from code.

-- John T.

>
> Thanks,
> Santosh
>