Hardware watchpoint support for MIPS?

Hardware watchpoint support for MIPS?

[David Daney]

GDB has had support for i386 debug registers on linux for quite some 
time now.

I was wondering if anyone has worked on similar support for the hardware 
debug support available in some MIPS processors?

Thanks in advance,
David Daney

Re: Hardware watchpoint support for MIPS?

[Daniel Jacobowitz]

On Thu, Mar 20, 2008 at 09:22:55AM -0700, David Daney wrote:
> GDB has had support for i386 debug registers on linux for quite some time 
> now.
>
> I was wondering if anyone has worked on similar support for the hardware  
> debug support available in some MIPS processors?

The issue's not GDB support, but kernel support.  I've seen several
proposals to expose watchpoint registers over ptrace on MIPS, but so
far none of them have merged.  The usual place they bog down is how to
explain to GDB what capabilities are available; x86 has it easy,
there's basically only one type.

-- 
Daniel Jacobowitz
CodeSourcery

Re: Hardware watchpoint support for MIPS?

[Paul Koning]

>>>>> "David" == David Daney <ddaney@avtrex.com> writes:

 David> GDB has had support for i386 debug registers on linux for
 David> quite some time now.

 David> I was wondering if anyone has worked on similar support for
 David> the hardware debug support available in some MIPS processors?

Yes, I have it on MIPS/sb1.

     paul

Re: Hardware watchpoint support for MIPS?

[Thiago Jung Bauermann]

On Thu, 2008-03-20 at 13:01 -0400, Daniel Jacobowitz wrote:
> On Thu, Mar 20, 2008 at 09:22:55AM -0700, David Daney wrote:
> > GDB has had support for i386 debug registers on linux for quite some time 
> > now.
> >
> > I was wondering if anyone has worked on similar support for the hardware  
> > debug support available in some MIPS processors?
> 
> The issue's not GDB support, but kernel support.  I've seen several
> proposals to expose watchpoint registers over ptrace on MIPS, but so
> far none of them have merged.  The usual place they bog down is how to
> explain to GDB what capabilities are available; x86 has it easy,
> there's basically only one type.

That's a problem in Power as well... What's currently there is
simplistic and hardly adequate.
-- 
[]'s
Thiago Jung Bauermann
Software Engineer
IBM Linux Technology Center

Re: Hardware watchpoint support for MIPS?

[David Daney]

Paul Koning wrote:
>>>>>> "David" == David Daney <ddaney@avtrex.com> writes:
> 
>  David> GDB has had support for i386 debug registers on linux for
>  David> quite some time now.
> 
>  David> I was wondering if anyone has worked on similar support for
>  David> the hardware debug support available in some MIPS processors?
> 
> Yes, I have it on MIPS/sb1.
> 

Great!

Could you be more specific?  Like where are the kernel patches to expose 
the debugs registers to ptrace and the gdb patches to take advantage of 
them?

David Daney

Re: Hardware watchpoint support for MIPS?

[David Daney]

Paul Koning wrote:
>>>>>> "David" == David Daney <ddaney@avtrex.com> writes:
> 
>  David> Paul Koning wrote:
>  >>>>>>> "David" == David Daney <ddaney@avtrex.com> writes:
>  >>
>  David> GDB has had support for i386 debug registers on linux for
>  David> quite some time now.
>  >>
>  David> I was wondering if anyone has worked on similar support for
>  David> the hardware debug support available in some MIPS processors?
>  >>  Yes, I have it on MIPS/sb1.
>  >> 
> 
>  David> Great!
> 
>  David> Could you be more specific?  Like where are the kernel patches
>  David> to expose the debugs registers to ptrace and the gdb patches
>  David> to take advantage of them?
> 
> As Daniel pointed out, the kernel and ptrace changes are really the
> issue here.
> 
> We modified a NetBSD 1.6.2 kernel, adding watchpoint primitives to its
> gdb stub (for kernel debug) and also to its ptrace.  The ptrace stuff
> was a bit hairy because you have to context switch that state.  
> 
> I don't think I had to change gdb at all, except perhaps for setting
> the #define symbols to turn the feature on and describe its behavior
> correctly with respect to single stepping.
> 
> Well, there was a bug at least in the earlier 6.x gdb releases where
> the code to sort out breakpoints and watchpoints would come up with
> the wrong answer if you hit a watchpoint while a breakpoint was also
> active (even at some entirely different address).
> 
> And yes, the notion of watchpoint is in the MIPS architecture but the
> specifics vary among implementations.  I have seen several, and they
> are close enough that a single gdb talking the same remote protocol
> works with all of them, but the target end does change somewhat from
> one MIPS flavor to the next.
> 

OK, thanks.  I mistakenly thought that you already had it working on 
mips[el]-linux and I might just be able to drop it in (It's OK to dream 
isn't it?).

David Daney