watchpoints and threads

watchpoints and threads

[PAUL GILLIAM]

The manual states:

     _Warning:_ In multi-thread programs, watchpoints have only limited
     usefulness.  With the current watchpoint implementation, GDB can
     only watch the value of an expression _in a single thread_.  If
     you are confident that the expression can only change due to the
     current thread's activity (and if you are also confident that no
     other thread can become current), then you can use watchpoints as
     usual.  However, GDB may not notice when a non-current thread's
     activity changes the expression.

Yet the test testsuite/gdb.threads/watchthreads.exp contains the
following:

        # Verify that we hit first watchpoint in child thread.
        set message "watchpoint on args\[0\] hit in thread"
        if { $args_1 > 1 } {
          pass $message
        } else {
          fail $message
        }
        
This seems like a conflict.

What brought this up is a user who wants this scenario to work:

        * The target program has a global variable.
        * GDB is started and a 'watch' command is give for that
        variable.
        * Some number of threads are started.
        * If _any_ thread changes the value of the global variable,
        _all_ threads stop.
        
The manual seems to say that this scenario will not work, but the
testcase seems to say that it will.

What's the story?

-=# Paul #=-

Re: watchpoints and threads

[Daniel Jacobowitz]

On Wed, Apr 26, 2006 at 11:03:02AM -0700, PAUL GILLIAM wrote:
> The manual seems to say that this scenario will not work, but the
> testcase seems to say that it will.
> 
> What's the story?

The patches to implement this (for some limited set of architectures?)
were approved but never committed, I believe.  Like last January.

-- 
Daniel Jacobowitz
CodeSourcery

Re: watchpoints and threads

[Wu Zhou]

Hi Daniel,

On Wed, 26 Apr 2006, Daniel Jacobowitz wrote:

> On Wed, Apr 26, 2006 at 11:03:02AM -0700, PAUL GILLIAM wrote:
> > The manual seems to say that this scenario will not work, but the
> > testcase seems to say that it will.
> > 
> > What's the story?
> 
> The patches to implement this (for some limited set of architectures?)
> were approved but never committed, I believe.  Like last January.

I guess this thread initiated by Jeff Johnston is what you were referring to:
http://sources.redhat.com/ml/gdb-patches/2004-12/msg00256.html

It was really a long thread, spanning several monthes.  My reading of 
these post tell me that it is partly approved at some point, and later 
held off for observer issue.  Is that right?  Does that kind of issue 
still hold in current gdb source code?  Does anybody has interest in 
looking into these question again?

I am recently working on h/w watchpoint support for ppc440 board.  
Customers are also inquirying if multi-threaded program works ok with 
h/w watchpoint.  I believe this feature is quite desirable for multi
threaded program.

Regards
- Wu Zhou