Re: [PATCH] Fix AIX thread NULL assertion failure during fork

[Ulrich Weigand <Ulrich.Weigand@de.ibm.com>]

Aditya Kamath1 <Aditya.Kamath1@ibm.com> wrote:

>So this was the issue. Thank you for suggesting this. I appreciate your knowledge
>to tell us this. Please see Program 1 and the GDB outputs pasted below this email
>for set detach-on-fork on/off and set follow-fork-mode=parent/child. This works
>in this test case and it looks like many other test cases as well are fixed from
>the GDB.base and GDB.threads test suite numbers mentioned below in a section.

Excellent, that's good to hear!

>Kindly let me know if this is okay.

I see one possible issue with the current patch - there is one
other existing caller to pd_activate in the ::wait routine:

      if (regcache_read_pc (regcache)
          - gdbarch_decr_pc_after_break (gdbarch) == data->pd_brk_addr)
        return pd_activate (ptid.pid ());
    }

  return pd_update (ptid.pid ());

This is the one place other than update_thread_list where you do
in fact need to call pd_update.  However, now that you've changed
pd_activate to no longer call pd_update, in the case where the
code above executes "return pd_activate", that call is now missing.

I think a straightforward fix might be to simply remove the "return"
and just fall through into the pd_update after pd_activate returned.

(Note that the return value of pd_activate should then no longer
be used anywhere and can be removed.)


Also, it looks like your're only updating the *current* inferior
in update_thread_list.  Other targets update the thread lists of
*all* inferiors.  E.g. have a look at the linux-thread-db.c
version of update_thread_list.


>Having said that we are missing some corner cases that I want to solve
>in a separate patch or the same depending on whatever you suggest.

Do the above change fix those corner cases?

Bye,
Ulrich