> This makes it sound like "interruptible sleep" is some kind of wedged
> state waiting for a debugger.  But, this simply means your process is just
> running as normal, and your test's case, normal means idling.  Most likely,
> your test program had a "sleep" call in it.  IOW, it'd be in sleep state
> even if had been started outside gdb.  Another test program could go
> to "T (stopped)", "R (running)", etc.

You're right. Rev log updated.

> This test must be called unconditionally.  current_ptrace_options is
> only initialized once:
> 
>  void
>  linux_enable_event_reporting (pid_t pid)
>  {
>    /* Check if we have initialized the ptrace features for this
>       target.  If not, do it now.  */
>    if (current_ptrace_options == -1)
>      linux_check_ptrace_features ();
> 
> so if the first process gdbserver debugs is an "attach" process,
> but the second one is not, you'll miss setting PTRACE_O_EXITKILL on
> the second process.
> 
> Instead, always call linux_test_for_exitkill, and then in
> linux_enable_event_reporting mask out PTRACE_O_EXITKILL
> from current_ptrace_options if 'attached' is false.

Right again.

Here is an attached patch. Re-tested on x86_64-linux as previously
done.

gdb/ChangeLog:

        * nat/linux-ptrace.h (PTRACE_O_EXITKILL): Define if not
        already defined.
        (linux_enable_event_reporting): Add parameter "attached".
        * nat/linux-ptrace.c (linux_test_for_exitkill): New advance
        declaration.  New function.
        (linux_check_ptrace_features): Add linux_test_for_exitkill call.
        (linux_enable_event_reporting): Add new parameter "attached".
        Do not call ptrace with the PTRACE_O_EXITKILL if ATTACHED is
        nonzero.
        * linux-nat.c (linux_init_ptrace): Add parameter "attached".
        Use it.  Update function description.
        (linux_child_post_attach, linux_child_post_startup_inferior):
        Update call to linux_enable_event_reporting.

gdb/gdbserver/ChangeLog:

        * linux-low.c (linux_low_filter_event): Update call to
        linux_enable_event_reporting following the addition of
        a new parameter to that function.

Thanks, Pedro.
-- 
Joel