[non-stop] 07/10 non-stop inferior control

[Pedro Alves]

A Wednesday 25 June 2008 20:44:09, Daniel Jacobowitz wrote:
> On Sun, Jun 15, 2008 at 10:05:16PM +0100, Pedro Alves wrote:
> > It is also needed to have support to interrupt/suspend a simple
> > thread so a new target_stop_ptid method was added to request
> > the target to interrupt a single thread.
>
> How about adding a ptid argument to to_stop instead?  I think the code
> will be shared in most cases.
>

Sure, I'll do that.  It was a way to not have to update all
targets before review.  :-)

> > @@ -1810,11 +1847,16 @@ handle_inferior_event (struct execution_
> >        && ecs->ws.kind != TARGET_WAITKIND_SIGNALLED &&
> > ecs->new_thread_event) add_thread (ecs->ptid);
> >
> > -  /* Mark all threads as not-executing.  In non-stop, this should be
> > -     adjusted to only mark ecs->ptid.  */
> > -  if (ecs->ws.kind != TARGET_WAITKIND_IGNORE
> > -      && stop_soon != STOP_QUIETLY)
> > -    set_executing (pid_to_ptid (-1), 0);
> > +  if (ecs->ws.kind != TARGET_WAITKIND_IGNORE)
> > +    {
> > +      /* Mark the stopped threads accordingly.  */
> > +      if (!non_stop
> > +	  || ecs->ws.kind == TARGET_WAITKIND_EXITED
> > +	  || ecs->ws.kind == TARGET_WAITKIND_SIGNALLED)
> > +	set_executing (pid_to_ptid (-1), 0);
> > +      else
> > +	set_executing (ecs->ptid, 0);
> > +    }
> >
> >    switch (ecs->ws.kind)
> >      {
>
> Are we going to miss the stop_soon check here?

Indeed, I'll recheck carefully.  It used to be needed, and the
symptom was a nice internal_error.  It isn't happening, which
probably means I either fixed the real issue, or I missed
something.  :-)

>
> > -	switch_to_thread (tp->ptid);
> > +	if (non_stop)
> > +	  context_switch_to (tp->ptid);
> > +	else
> > +	  switch_to_thread (tp->ptid);
> > +
>
> Ow ow my head, I really wish we didn't need both of these.  We'll
> never keep track of which is which.  

Easy, It's always both, except when it's a temporary switch to another
thread and back.  :-)

> Will we eventually be able to merge them?

Oh, yeah.  context_switch(_to) is a dying species.  I'm working on getting
rid  of it.  Then we'll only have switch_to_thread (which should go to,
if/when we switch to have a frame cache per thread).  stop_pc (the other
job of switch_to_thread) isn't doing its job in some cases either,
and we could remove it.  Say:

<stop at thread 1> (sets stop_pc)
p $pc = xxx
thread 2 (sets stop_pc)
thread 1 (sets stop_pc, but it's no longer the original...)

We can switch to always using read_pc() (or equivalent).

I have all-stop mode / linux mostly covered, now I'm making sure remote
always has a thread, monitor targets will be easy, and win32-nat.c
is also taken care of.  Will need to convert non-stop too, but
that'll take a bit.  I prefer that this series goes in first,
otherwise, we'll never get this done.  Baby steps, please.  :-)

>
> > -  add_prefix_cmd ("thread", class_run, thread_command, _("\
> > +  c = add_prefix_cmd ("thread", class_run, thread_command, _("\
> >  Use this command to switch between threads.\n\
> >  The new thread ID must be currently known."),
> >  		  &thread_cmd_list, "thread ", 1, &cmdlist);
>
> Indentation on that last line.
>
> Otherwise OK.

Thanks a bunch.  I'll get back once I resolve the issues
you mentioned.

-- 
Pedro Alves