Re: [DynInst_API:] [dyninst/dyninst] aa0880: Correct interpretation of Intel instructions requires

[Barton Miller <bart@xxxxxxxxxxx>]

This is such a classic bug. We teach about this in our Software Security 
class (where it can often lead to vulnerabilities). Good catch.
--bart

On 2/16/2023 3:51 PM, Bolo -- Josef Burger wrote:
>    Branch: refs/heads/bolo/x86_immed
>    Home:   https://github.com/dyninst/dyninst
>    Commit: aa088045d3721819b705f6505608bd229dbc64d1
>        https://github.com/dyninst/dyninst/commit/aa088045d3721819b705f6505608bd229dbc64d1
>    Author: Bolo <bolo@xxxxxxxxxxx>
>    Date:   2023-02-16 (Thu, 16 Feb 2023)
>
>    Changed paths:
>      M common/src/arch-x86.h
>      M instructionAPI/h/Result.h
>
>    Log Message:
>    -----------
>    Correct interpretation of Intel instructions requires
> sign extension.   On intel it works because chars are
> signed on that platform.   On other platforms (arm64, powerpc64)
> intel code was not correctly analyzed because incorrect
> values were generated from 8 bit immediate constants;
> which were interpreted as unsigned on those systems.
>
> These changes ensure intel code will be correctly interpreted
> on any system, regardless of the signedness of chars.
>
> A n-way validation was used to validate this change, to
> ensure that correct code interpretation of (intel, arm,
> power) code happens on all three platforms.    No other
> issues of this nature were located -- at this time.
>
>
> _______________________________________________
> Dyninst-api mailing list
> Dyninst-api@xxxxxxxxxxx
> https://lists.cs.wisc.edu/mailman/listinfo/dyninst-api