Re: [RFC 3/6] mailbox: pl320: migrate to mbox framework

[Courtney Cavin]

On Mon, Feb 10, 2014 at 07:28:54PM +0100, Rob Herring wrote:
> On Fri, Feb 7, 2014 at 6:50 PM, Courtney Cavin
> <courtney.cavin@xxxxxxxxxxxxxx> wrote:
> > We don't remove the legacy methods here, but we mark them as deprecated
> > in the hopes that people with the ability to properly test modifications
> > can adapt its users.
> 
> The DT for highbank is pretty much fixed at this point. So adopting
> this will need a way to register without DT. Unfortunately, I don't
> have access to h/w either ATM.

That's fine. The lookup table stuff (see mbox_channel_lookup() and
mbox_add_table()) should solve this.

Hopefully we'll find someone with a test-setup for this.

> I should note that this driver is very much highbank specific and not
> really a generic pl320 driver. The pl320 has up to 8 mailboxes and 8
> interrupts. How it is used from there is a software decision. I've
> never seen any other user, but it could be done quite differently from
> how it is used in highbank. In the case of highbank, we assigned a tx
> and rx mailbox. While both the management core and linux side have all
> 8 interrupts wired up, we have assigned an interrupt to each side. I
> suppose you could have the interrupt tied to each mailbox, but really
> they are unrelated in the pl320 as each mailbox message could have
> multiple targets (interrupts). Probably splitting this between a pl320
> lib and platform specific drivers would be the right split if there
> are ever other users.

I'm not exactly sure I follow, and I probably should lookup the pl320
spec, but from the way you describe it and the simplicity of the
existing implementation, it seems to me that one could easily implement
this using DT to decribe how the hardware is hooked up.

> > -       ipc_irq = adev->irq[0];
> > -       ret = request_irq(ipc_irq, ipc_handler, 0, dev_name(&adev->dev), NULL);
> > +       pl->adapter.dev = &adev->dev;
> > +       pl->adapter.ops = &pl320_mbox_ops;
> > +       pl->adapter.nchannels = 1;
> 
> Shouldn't this be 2? The 2 channels here are not a single
> bi-directional channel in any way. They are completely independent and
> have unrelated events. For example we originally defined having 3
> mailboxes where we had 2 tx mailboxes for fast and slow messages, but
> we ultimately decided everything could be a single tx mailbox. Event
> completion is handled synchronously via the pl320's handshake
> mechanism. I'd imagine you could have a protocol where you have async
> completions via an rx mailbox instead.

I generally see a mailbox as having both rx & tx, and I thought that
this was what this driver was attempting to model, so I implemented it
that way.  If the channels are completely independent, there's no reason
why we can't model this as one rx, and one tx.  The proposed API for
this should be suitable.

If I understand what you mean, "event completion" in this case is an ACK
for each event.  Async event ACKing sounds pretty dirty, and I don't
really want to touch on that in the core implementation, but there
should be nothing stopping you from exposing a mailbox which uses
.put_message() to ACK an RX event.

> Rob

I'm glad you found this, as apparently I got your email address all
wrong.

Thanks for the comments! 

-Courtney
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/