rt/tst-aio7 weirdness on ppc

[Roland McGrath]

> When this happens you want to do the attach, then manually send a
> SIGCONT to the process GDB is attaching to; then it should successfully
> attach and stop immediately.

Ok, gdb stops crashing when I get it to use the right libthread_db.  So,
the process is stopped in its wait4 call as you'd expect, because the
signal went to the whole pgrp.  In the child process, there is a thread
running handle_fildes_io (aio code) that, lo and behold, is doing a read on
fd 0.  It looks like it's looking at a struct aiocb.  There must be some
kind of race bug in the aio stuff.  I will look some more.