This is the mail archive of the mailing list for the GDB project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [patch] Fix for internal-error: linux_nat_post_attach_wait: Assertion `pid == new_pid && WIFSTOPPED (status)' failed.

On Wednesday 14 October 2009 19:21:03, Paul Pluzhnikov wrote:

> It's very hard to test, since it only happens once in about 50
> attaches, but I did catch this several times, and GDB appears to
> have done the right thing.

Ugh.  Happened much more often for me, like 1 in 5...

On Wednesday 14 October 2009 19:21:03, Paul Pluzhnikov wrote:
>  /* Wait for the LWP specified by LP, which we have just attached to.
> -   Returns a wait status for that LWP, to cache.  */
> +   Returns 1 if LWP has been stopped, else 0.  */
>  static int
> -linux_nat_post_attach_wait (ptid_t ptid, int first, int *cloned,
> -                           int *signalled)
> +linux_nat_post_attach_wait (ptid_t ptid, int first, int *status,
> +                           int *cloned, int *signalled)
>  {

I don't think you needed the interface change.  The new interface
has two ways of saying the LWP exited, the 0 return, and
the !WIFSTOPPED(status), which looks strange.  With such an
interface, I would expect the return code to indicate a real failure,
but an exit indication isn't really a failure.  Any reason you'd not
just do ...

> -      status = linux_nat_post_attach_wait (ptid, 0, &cloned, &signalled);
> +      if (!linux_nat_post_attach_wait (ptid, 0, &status, &cloned, &signalled))
> +       return -1;
> +

      status = linux_nat_post_attach_wait (ptid, 0, &cloned, &signalled);
      if (!WIFSTOPPED (status))
        return -1;

?  Anyway, I was just surprised.  Either way is fine really.

> > I would believe that we can also see the main LWP exit right
> > after attach (in linux_nat_attach).  I'm not sure what exactly
> > is the best to do UI wise in that case.  If we want to store
> > the event pending to report later, we'll have to use
> > the lwp->waitstatus field, not lwp->status, due to the
> > fact that lwp->status == 0 is ambiguous with
> > "no-stored-pending-event" (see status_callback).
> I believe I did this now. Not sure whether lp->resumed should also
> be set here.

Yes, it should.  Otherwise, the pending event is ignored (see
status_callback), and worse, in this case, you'll hit this


  /* Make sure there is at least one LWP that has been resumed.  */
  gdb_assert (iterate_over_lwps (ptid, resumed_callback, NULL));

Oh, my bad.  I now took a look at what
infcmd.c:attach_command|attach_command_post_wait would do
in the case of the target_wait right after attach returning a
process exit, and it seems it would misbehave, and try to poke
at the inferior's memory...  Best not to go there for such a rare 
use case.  So, leaving the exit event pending ends up
fixing nothing afterall...

> > The simple
> > alternative is to again just pretend that the process had
> > exited before we managed to attach to it, get rid of it,
> > and error out like we would if the process didn't exist
> > at all when we tried to attach.
> This appears harder to do, since to_attach doesn't return anything.

Back to the "simple" variant: The inferior just added is
always inferior_ptid/current_inferior().  If you look
at fork-child.c:startup_inferior, you'll see bits of
code doing exactly what you'd need to do.  E.g.:

	    target_terminal_ours ();
	    target_mourn_inferior ();
            if (WIFEXITED (status))
              error (_("During startup program exited with code %d."),
   	             WIFEXITCODE (status));
            else if (WIFSIGNALED (status))
              error (_("During startup program exited with signal ..."),

But if this looks too much trouble for what it's worth, that's
probably right :-).  I'd be quite fine to ignore that use case
for now.

Pedro Alves

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]