Function argument no longer in register

Function argument no longer in register

[Orjan Friberg]

Hi all,

What should happen when a function argument goes out of scope because 
the register used to pass it is now used for something else?  I 
encountered the following situation:

foo (char *str)
{
   ...
   i = bar ();
   ...
}

The str argument is passed in register r10, which is then overwritten 
with the return value from bar.  If I stop somewhere after the call to 
bar and try to "print str", GDB will try to read from the address 
pointed out by r10 (i.e. sends an 'm' packet to the remote stub).

I don't know what kind of problem I'm looking at here.  Could it be that 
the debug information is wrong?  I'm guessing GDB somehow fails to 
realize that the function argument is out of scope at this point.

-- 
Orjan Friberg
Axis Communications

Re: Function argument no longer in register

[Daniel Jacobowitz]

On Mon, Mar 29, 2004 at 04:21:19PM +0200, Orjan Friberg wrote:
> Hi all,
> 
> What should happen when a function argument goes out of scope because 
> the register used to pass it is now used for something else?  I 
> encountered the following situation:
> 
> foo (char *str)
> {
>   ...
>   i = bar ();
>   ...
> }
> 
> The str argument is passed in register r10, which is then overwritten 
> with the return value from bar.  If I stop somewhere after the call to 
> bar and try to "print str", GDB will try to read from the address 
> pointed out by r10 (i.e. sends an 'm' packet to the remote stub).
> 
> I don't know what kind of problem I'm looking at here.  Could it be that 
> the debug information is wrong?  I'm guessing GDB somehow fails to 
> realize that the function argument is out of scope at this point.

Nothing's wrong - eventually, versions of GCC with proper location list
will tell GDB that the variable is out of scope, but no released
version does that.  I think 3.5 snapshots will.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

Re: Function argument no longer in register

[Orjan Friberg]

Daniel Jacobowitz wrote:
>>
>>I don't know what kind of problem I'm looking at here.  Could it be that 
>>the debug information is wrong?  I'm guessing GDB somehow fails to 
>>realize that the function argument is out of scope at this point.
> 
> 
> Nothing's wrong - eventually, versions of GCC with proper location list
> will tell GDB that the variable is out of scope, but no released
> version does that.  I think 3.5 snapshots will.

Ah, ok.  Thanks.  The setup is a remote stub debugging a Linux kernel, 
and the first page (8k for CRIS) is read/write-protected.  So, I guess I 
need to filter out these bogus reads in the stub to prevent the kernel 
from oopsing.

-- 
Orjan Friberg
Axis Communications