Re: minor fix for i386v-nat.c

Re: minor fix for i386v-nat.c

[Andrew Cagney]

Did anyone ever look at this?



> In i386_insert_aligned_watchpoint the actual address being watched was saved
> in address_loookup.  This is fine for aligned watchpoints, the address gdb
> wants to watch is the same as the actual address.  For non-aligned
> watchpoints
> we have to ask the hadware to watch up to 3 different addresses.  We need to
> remember the address gdb wants to watch so i386_remove_watchpoint knows
> which
> registers to clean up.
> 
> Changelog:
> 
> 2000-10-30  John Hughes  <john@CalvaEDI.COM>
> 
> * i386v-nat.c (i386_insert_aligned_watchpoint) save waddr (the
>         watched address) in address_lookup[] rather than addr.  This allows
>         clean removal of non-aligned watchpoints.
> 
> Patch (against released gdb-5.0) attached.
> 
> --
> John Hughes <john@Calva.COM>,
>         CalvaEDI SA.                            Tel: +33-1-4313-3131
>         66 rue du Moulin de la Pointe,          Fax: +33-1-4313-3139
>         75013 PARIS.
> 
> 
> 
> --- i386v-nat.c.orig	Wed Dec  8 03:50:38 1999
> +++ i386v-nat.c	Sun Oct 29 12:53:16 2000
> @@ -213,7 +213,7 @@
>  	  addr);
>  
>    /* Record where we came from.  */
> -  address_lookup[register_number] = addr;
> +  address_lookup[register_number] = waddr;
>    return 0;
>  }
>  
> 

RE: minor fix for i386v-nat.c

[John Hughes]

Andrew Cagney <ac131313@cygnus.com> writes:
> 
> Did anyone ever look at this?
> 

You mean appart from me?  No idea.

(I don't think I'm on the mailing list any more, CC me if you
have questions).

RE: minor fix for i386v-nat.c

[John Hughes]

Andrew Cagney <ac131313@cygnus.com> writes:
>
>Did anyone ever look at this?

The right thing to do would be to zap the watchpoint handling from
i386v-nat.c and use the more sophisticated code in i386-nat.c.

Re: minor fix for i386v-nat.c

[Eli Zaretskii]

On Thu, 28 Jun 2001, Andrew Cagney wrote:

> Did anyone ever look at this?

Is any x86 port still using this code?  I'd expect them all to switch
to the new watchpoint support in i386-nat.c, by now.

> > In i386_insert_aligned_watchpoint the actual address being watched
> > was saved in address_loookup.  This is fine for aligned
> > watchpoints, the address gdb wants to watch is the same as the
> > actual address.  For non-aligned watchpoints we have to ask the
> > hadware to watch up to 3 different addresses.  We need to remember
> > the address gdb wants to watch so i386_remove_watchpoint knows
> > which registers to clean up.

Won't this do the wrong thing in i386_stopped_by_watchpoint?  You'll
report the wrong address there (not the one the debug register
actually watched).

I think a cleaner way to do this is to store the actual address being
watched by the DRi, but to modify i386_remove_watchpoint to break up
its act into several individual removals like
i386_insert_aligned_watchpoint does.  This is what the code in
i386-nat.c does.

RE: minor fix for i386v-nat.c

[Eli Zaretskii]

On Thu, 28 Jun 2001, John Hughes wrote:

> The right thing to do would be to zap the watchpoint handling from
> i386v-nat.c and use the more sophisticated code in i386-nat.c.

Yep.

Re: minor fix for i386v-nat.c

[Andrew Cagney]

> On Thu, 28 Jun 2001, John Hughes wrote:
> 
> 
>> The right thing to do would be to zap the watchpoint handling from
>> i386v-nat.c and use the more sophisticated code in i386-nat.c.
> 
> 
> Yep.


In that case I declare the patch dead :-)

	Andrew