Andrew Morton wrote:
> On Tue, 9 Sep 2008 11:55:36 +0100 Alan Cox <alan@lxorguk.ukuu.org.uk> wrote:
>> lock_kernel can sleep.
> 
> That used to be the case, but the mutex_lock() version of the bkl got
> removed.  It may of course come back.
> 
>> Its not allowed...
>>
> 
> yup.

OK, attached is a new version of this patch:

	tty-fix-echo-tab-erase-and-locking.patch

I have removed lock_kernel() from n_tty.c completely (which is a nice
benefit).  As we discussed, this is now easier due to my echo patch
isolating the column state code.  In its place, I created two new mutex
locks: "output_lock" (protects column state and and driver buffer space
left) and "echo_lock" (protects the echo buffer).  I could have used
just one to protect all of this, but there was no need to make the lock
that wide-ranging (e.g., it's OK to add stuff to the echo buffer while
regular output is happening).

All echo buffer operations take the echo lock, and all process_out*
functions take output lock.  The one function that takes both locks is
process_echoes (this is the one that erroneously took both the echo spin
lock and the BKL before).  Note that the process_output functions are at
least sometimes (and maybe all of the time) locked by tty_write_lock (as
Alan mentioned), but this does not protect echo from interfering with
regular output.

Let me know what you think of this patch.  I am using it now, and it
appears to work well so far.

						Thanks again, Joe