Re: [PATCH/RFC] Improve the "thread_step_needed" logic

[Daniel Jacobowitz <dmj+@andrew.cmu.edu>]

On Fri, Jun 15, 2001 at 06:13:56PM -0700, Michael Snyder wrote:
> > <sigh>
> > 
> > While you're fixing this, could you please address my issue from last
> > week with this code?  Checking "step" at this point in resume() is not
> > correct, since it will have been cleared if SOFTWARE_SINGLE_STEP_P ().
> > 
> > > --- 879,913 ----
> > >     if (should_resume)
> > >       {
> > >         ptid_t resume_ptid;
> > > +
> > > +       resume_ptid = RESUME_ALL;             /* Default */
> > >
> > > !       if ((step || singlestep_breakpoints_inserted_p) &&
> > > !       !breakpoints_inserted && breakpoint_here_p (read_pc ()))
> > 
> > > +
> > > + #ifdef CANNOT_STEP_BREAKPOINT
> > > +       /* Most targets can step a breakpoint instruction, thus executing it
> > > +      normally.  But if this one cannot, just continue and we will hit
> > > +      it anyway.  */
> > > +       if (step && breakpoints_inserted && breakpoint_here_p (read_pc ()))
> > > +     step = 0;
> > > + #endif
> > 
> > Specifically, those bits.
> 
> Oh, sorry, I missed this part of your msg.  
> Well, this isn't related to SOFTWARE_SINGLESTEP_P, 
> it's a separate issue.  But if you'll notice, I believe this is also 
> handled correctly, because my "thread_step_needed" test includes
> (NOT breakpoints_inserted), while this CANNOT_STEP_BREAKPOINT clause
> will only be taken if breakpoints_inserted is true.  So they are
> mutually exclusive, if you will.  I tested this on Alpha Linux, which
> does NOT use software single step, but DOES use CANNOT_STEP_BREAKPOINT,
> and it works.

My apologies.  I need to read closer before replying :)

I'm assuming that singlestep_breakpoints_inserted_p is zero normally
upon entering resume - that only makes sense.  In that case, your
change preserves the exact behavior I was asking for.  I'm going to run
it through the testsuite on MIPS/Linux, which I've finally gotten into
(almost) functional shape, and let you know if something breaks - I
expect that it will not break, though.

-- 
Daniel Jacobowitz                           Debian GNU/Linux Developer
Monta Vista Software                              Debian Security Team