Re: [PATCH] pxa serial auto flow control

[Robert Jarzmik <robert.jarzmik@xxxxxxx>]

Russell King - ARM Linux <linux@xxxxxxxxxxxxxxxx> writes:

>> In your example, I aggree that tty layer action on RTS won't work.
>
> But it _has_ to.  There's no way around that.
Now I've wandered in the tty layers, I understand that.
>
> It will always read from the FIFO as long as the FIFO says there's data
> available.  If the tty buffers fill up after the tty layer has asked
> the driver to throttle the remote end, characters _will_ be dropped on
> the floor.
Yep.

> So, the tty layer call back into the driver to deassert the RTS signal
> _must_ continue work.  If it can't work with the AFE bit enabled, then
> the PXA uarts are basically broken.
Yes.

> It's rather unfortunate that the available documentation doesn't indicate
> what happens when AFE=1 RTS=0 other than saying that auto-RTS flow is
> disabled.  It doesn't say what state the RTS line ends up in.  Sensible
> UARTs place the RTS line in a deasserted state, which means that
> everything works just fine as is (and you don't touch the RTS bit when
> setting AFE.)

I'll make some tests this week-end. I'll put my GPS chip in AFE=1, RTS=1, see
it's data, and then toggle RTS=0. If data stops, I reactivate RTS=1 several
seconds later. If I see all the frames that were blocked in the meantime, then
the mecanism works, in the fashion I described in the reply to Uli.

Then I'll repost the patch, without the RTS bit activation when setting AFE (as
Uli suggested).

--
Robert

-------------------------------------------------------------------
List admin: http://lists.arm.linux.org.uk/mailman/listinfo/linux-arm-kernel
FAQ:        http://www.arm.linux.org.uk/mailinglists/faq.php
Etiquette:  http://www.arm.linux.org.uk/mailinglists/etiquette.php