Re: [Valgrind-developers] Spill Registers Help

[Bryan M. <om...@br...>]

On Friday 11 Nov 2005 23:44, Nicholas Nethercote wrote:
> On Fri, 11 Nov 2005, Bryan Meredith wrote:
> 
> > I need to be able to see writes to spill registers.
> >
> > I can trap writes to memory and the shadow processor registers but I cant 
work
> > out how to trap writes to the spill registers.
> >
> > Can someone point me at the relevent bit of documentation or provide a 
small
> > sample of instrumentation please?
> >
> > The basic state of Omega at the moment is that it mostly works but it 
tends to
> > declare a leak and then out of nowhere, a new reference pops up (which is
> > kind of annoying). I added some instrumentation so I could see loads and
> > right before the last reference on the stack goes out of scope, there is a
> > load of the contents. Unfortunately, I cant see where the data is stored 
but
> > I suspect that it is going into a spill register as it doesn't get written
> > into a memory location nor any shadow processor register. This typically
> > happens with C++ code - at least I haven't seen it occur with pure C but I 
am
> > still running a limited set of apps in order to test.
> 
> I'm missing something...  what do you mean by "spill registers"?  Do you 
> mean spills introduced by Valgrind's code generator?
> 
> N
> 

Yes - there is a memory load into a temporary register. Once that happens, I 
dont see the value it holds until it is put into a processor register on the 
simulated cpu. I can only assume that it is being stored in the "spills" 
area. I need to be able to track this as it is the final live reference to an 
allocated block and without it, I raise a false leak alert as the stack 
pointer is moved between the load and the put (then sheepishly free the block 
later when the code calls free()).

Bryan "Brain Murders" Meredith