Description of Problem:

From the thread "Fix fflush starvation [#57014]" on ecc-target-changes:

> But I don't see how your change manages this at all.  It seems to me
> that if you enter this [loop] with some other thread holding the lock,
> you'll try once and fail and then immediately just do the
> lock-with-wait.  Or, am I missing something?


The idea was that you may go enter this loop with a stream already locked,
and someone else may do the same, resulting in a potential deadlock.

However, I've just noticed that this is a difference between my[1] code and
the pthread code. In my code Cyg_StdioStream::refill_read_buffer will
already have flushed the stream, whereas in the pthread code, the code in
question deals with all the streams (including the one already locked).

I'll open a bugzilla bug, as I'm not sure exactly what would happen with
!CYGSEM_LIBC_STDIO_WANT_BUFFERED_IO. Right now it would break, but flushing
should be mostly a nop anyway. At the very least, flush_output_unlocked
should call cyg_stdio_flush in that case which it doesn't.