[panda-users] taint inlining

[Brendan Dolan-Gavitt]

Correct – failing to inline will not change the semantics of the code.

Whether or not something can be inlined is determined by LLVM (we're just
calling its InlineFunction method). Looking through the ways it can return
false (http://llvm.org/doxygen/InlineFunction_8cpp_source.html) my best
guess is that it was trying to inline an external function. It might be
useful to modify that "Inlining failed" message to print out what it was
trying to inline, which could give you some clues as to why it failed.

Other plugins that use taint2 shouldn't have any effect on taint2's ability
to inline. But, of course, it can only inline something that it has LLVM
bitcode for, so you can't (for example) inline calls to your own functions
(unless you compile them to LLVM).

On Mon, Feb 19, 2018 at 9:52 AM, Manolis Stamatogiannakis <mstamat at gmail.com
> wrote:

> Hello,
>
> I have a couple of questions about taint operations inlining. I have
> provided inline=y as an argument to taint2. However, all I get is a ton of
> "Inlining failed!" printouts.
>
> If I understand correctly, this only means that a regular function call
> will be inserted instead. I.e. this is informational only, the correctness
> of the analysis you run is not affected. Is this right?
>
> What could be preventing inlining of taint operations from working? Are
> there any restrictions when building plugins that use taint2 in order to
> allow for taint operations inlining?
>
> Thanks,
> Manolis
>
> _______________________________________________
> panda-users mailing list
> panda-users at mit.edu
> http://mailman.mit.edu/mailman/listinfo/panda-users
>
>


-- 
Brendan Dolan-Gavitt
Assistant Professor, Department of Computer Science and Engineering
NYU Tandon School of Engineering
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.mit.edu/pipermail/panda-users/attachments/20180219/7b96ee89/attachment.html