Re: [DynInst_API:] Accessing more x86 registers

[Andrew Bernat <bernat@xxxxxxxxxxx>]

Quick reply :)

Those two definitely should be in the list; I know we can access them internally. The flags registers would be weirder. 

And you can get the original pc with a snippet type, so hopefully that will work for you. 

Drew

--
Drew Bernat
bernat@xxxxxxxxxxx
abernat@xxxxxxxxxxx

On Sep 10, 2012, at 5:17 PM, Josh Stone <jistone@xxxxxxxxxx> wrote:

> On 09/08/2012 09:12 AM, Andrew Bernat wrote:
>> What are you trying to accomplish with this? The SPRs are not accessible
>> via instrumentation, which is why the BPatch registers method didn't
>> return them. 
> 
> Hmm - well as I said, I haven't tried it yet, but I assumed the SPRs
> must be working for Mike at least for him to have this patch.
> 
> My intent is to allow systemtap scripts to access program state, like
> parameters and locals near the place they choose to probe.  GPRs are
> good for a lot of this, but stack and frame registers are also commonly
> needed in DWARF expressions.
> 
> I suppose it's possible, but I hope unlikely, that the IP could be
> referenced too, but I can see how that would especially challenging to
> reason about under Dyninst.  We do have our own unwinding runtime code
> which would need the IP, and almost certainly wouldn't work here; I have
> yet to consider this much.  I remember that Dyninst has an unwinder, but
> that's all mutator-side, right?
> 
> Anyway, just adding %rbp and %rsp to the available x86 register set
> would go a long way toward completeness.  Could those be made accessible?
> 
> Thanks,
> Josh