HW Watchpoint events and the inferior

HW Watchpoint events and the inferior

[Luis Machado]

Folks,

I've been investigating watchpoints on GDB and i have a question on how
GDB handles Hardware Watchpoint events and which TID it uses to grab
information about the stopped_data_address, considering a multithreaded
application.

The stopped_data_address functions seem to use the current
"inferior_ptid" variable to do a PTRACE request in order to get
information from the debug registers. The "inferior_ptid" holds ID's for
the task that was currently running when GDB stopped. 

Is it always true that, if we have a triggered hw watchpoint event, the
inferior_ptid represents the task that actually triggered it?

What if we did trigger a watchpoint but it was a different task (other
than the current inferior_ptid) that triggered it? Will GDB eventually
switch to the other task and confirm that we had a hw watchpoint
trigger?

It could also happen that two or more threads triggered a watchpoint.
Does GDB handle those events sequentially?

Just to make the scope clear, i'm trying to associate a specific hw
watchpoint event with the thread that actually triggered it. It believe
that would be useful information.

Regards,

-- 
Luis Machado
Software Engineer 
IBM Linux Technology Center
LoP Toolchain/Debuggers' team
Phone: +55 19-2132-2218
T/L: 839-2218
e-mail: luisgpm@vnet.linux.ibm.com

Re: HW Watchpoint events and the inferior

[Daniel Jacobowitz]

On Wed, Aug 01, 2007 at 12:18:01PM -0300, Luis Machado wrote:
> Folks,
> 
> I've been investigating watchpoints on GDB and i have a question on how
> GDB handles Hardware Watchpoint events and which TID it uses to grab
> information about the stopped_data_address, considering a multithreaded
> application.

GDB does not contain support for watchpoints in multi-threaded
programs.  You can find several patches and discussions of this issue
in the list archives as recently as the end of June.

In reply to a message from you, in fact!

-- 
Daniel Jacobowitz
CodeSourcery

Re: HW Watchpoint events and the inferior

[Luis Machado]

Hi,

On Wed, 2007-08-01 at 12:00 -0400, Daniel Jacobowitz wrote:
> On Wed, Aug 01, 2007 at 12:18:01PM -0300, Luis Machado wrote:
> > Folks,
> > 
> > I've been investigating watchpoints on GDB and i have a question on how
> > GDB handles Hardware Watchpoint events and which TID it uses to grab
> > information about the stopped_data_address, considering a multithreaded
> > application.
> 
> GDB does not contain support for watchpoints in multi-threaded
> programs.  You can find several patches and discussions of this issue
> in the list archives as recently as the end of June.

Yes, in fact i'm looking forward to improve that situation. That's why i'm going through some of the inferior handling code. Unfortunatelly there isn't much documentation there, and sometimes it gets rather bleak as to what GDB is really trying to do, and how it would handle multiple traps.


Regards,

-- 
Luis Machado
Software Engineer 
IBM Linux Technology Center
LoP Toolchain/Debuggers' team
Phone: +55 19-2132-2218
T/L: 839-2218
e-mail: luisgpm@vnet.linux.ibm.com

Re: HW Watchpoint events and the inferior

[Daniel Jacobowitz]

On Wed, Aug 01, 2007 at 02:13:27PM -0300, Luis Machado wrote:
> Yes, in fact i'm looking forward to improve that situation. That's
> why i'm going through some of the inferior handling
> code. Unfortunatelly there isn't much documentation there, and
> sometimes it gets rather bleak as to what GDB is really trying to
> do, and how it would handle multiple traps.

It's not sensible to ask how GDB handles this sort of thing, until it
actually does.  Did you mean how it _should_ handle it?

Are you starting from Jan's patches?  If not, why did you decide not
to?  I have been assuming that it would be a simple matter of merging
them.

-- 
Daniel Jacobowitz
CodeSourcery