Re: [DynInst_API:] Inserting multiple snippets at an insertion point?

[Bill Williams <bill@xxxxxxxxxxx>]

On 07/21/2014 01:15 PM, Wayne Motycka wrote:
> Oh, I had forgot about the insertBefore and insertAfter flags, that is
> probably what I'll need to use.
>
> Thanks Bill.
As an aside, call/return edge instrumentation and pre/post call 
instrumentation (for particular callers) or entry/exit instrumentation 
(for all callers) *should* be entirely equivalent on x86, and should 
provide sane addresses in the instpoints. The only reason to use edge 
instrumentation is to correctly handle conditional calls/returns (or for 
modification, rather than instrumentation).
If you're working on PPC or with a cross-platform mutator, or there's 
some reason other than engineering ones to use edge instrumentation, 
then obviously, carry on.
> -Wayne
>
> On Mon, July 21, 2014 11:53 am, Bill Williams wrote:
> > On 07/21/2014 11:45 AM, Wayne Motycka wrote:
> >
> > > I've been trying to find ways to instrument the edges in the Dyninst
> > > CFG
> > > but instrumenting the back edge from a system library call, e.g. calling
> > >   printf(), doesn't give insertion points that seem reasonable/valid
> > > (the
> > > insertion point addresses given by getExitPoints() and getAllPoints()
> > > appears to be out of the local programs address ranges.)
> > It's possible (and if memory serves, it's likely) that when you're
> > inserting edge instrumentation, its instpoint address will be somewhere in
> > relocated code and will not correspond neatly to an original address.
> > (This is particularly true for call/return edges, as a particular
> > interprocedural edge doesn't live anywhere in the original binary.) We
> > *should* hand back NULL instpoints if they're really unusable.
> >
> >
> > > So, I'm thinking of moving that edge instrumenting code block into the
> > > edge that connects the calling block to the subsequent block along with
> > > the instrumenting code for that connecting edge, effectively marking
> > > that connecting edge as both the return path from the printf() system
> > > call and the inter-block connecting edge (which is EdgeTypeEnum type of
> > > "post-call
> > > fallthrough")
> > This is also a reasonable approach, and call fallthrough edges exist for
> > this sort of simplification.
> >
> > > My question is if it is possible to insert two snippets referencing
> > > calls to my edge marking library routine?
> > Yes, it's valid to insert multiple snippets at the same instrumentation
> > point; the "before/after" parameter to insertSnippet governs the relative
> > order of execution, if that's relevant to you.
> >
> > > Thanks,
> > > Wayne Motycka
> > > E2 Lab @ Univ. of Nebraska - Lincoln
> > >
> > >
> > > _______________________________________________
> > > Dyninst-api mailing list
> > > Dyninst-api@xxxxxxxxxxx
> > > https://lists.cs.wisc.edu/mailman/listinfo/dyninst-api
> >
> > --
> > --bw
> >
> >
> > Bill Williams
> > Paradyn Project
> > bill@xxxxxxxxxxx _______________________________________________
> > Dyninst-api mailing list
> > Dyninst-api@xxxxxxxxxxx
> > https://lists.cs.wisc.edu/mailman/listinfo/dyninst-api

--
--bw

Bill Williams
Paradyn Project
bill@xxxxxxxxxxx