[PATCH v2 3/6] mailbox: Add support for ST's Mailbox IP

[Lee Jones]

On Fri, 14 Aug 2015, Jassi Brar wrote:

> On Fri, Aug 14, 2015 at 12:03 PM, Lee Jones <lee.jones at linaro.org> wrote:
> > On Thu, 13 Aug 2015, Jassi Brar wrote:
> >> On Mon, Jul 27, 2015 at 3:14 PM, Lee Jones <lee.jones at linaro.org> wrote:
> >>
> >> > +
> >> > +static bool sti_mbox_tx_is_ready(struct mbox_chan *chan)
> >> > +{
> >> > +       struct sti_channel *chan_info = chan->con_priv;
> >> > +       struct sti_mbox_device *mdev = chan_info->mdev;
> >> > +       unsigned int instance = chan_info->instance;
> >> > +       unsigned int channel = chan_info->channel;
> >> > +       void __iomem *base = MBOX_BASE(mdev, instance);
> >> > +
> >> > +       if (!(chan_info->direction & MBOX_TX))
> >> > +               return false;
> >> >
> >> Here the 'direction' is gotten via DT node of the client i.e, you
> >> expect consumer drivers to tell the provider what its limitations are?
> >>
> >> IMO if some physical channel can't do TX then that should be either
> >> hardcoded inside the controller driver or learnt via DT node of the
> >> _controller_.
> >
> > That's a fair point.
> .
> 
> >  I need to create a new property similar to the
> > already existing 'read-only'.  I guess 'tx-only' is equivalent.
> >
> Just to be clear, if you must have such a property it should come from
> the _controller_ node.

Correct.  That's the plan.

> However at one point you said,  "Only the A9 (Mbox 0) can Rx."
>   Which sounds like the 'simplex' constraint is not coming from the
> mailbox controller but from the remote endpoints that don't RX+TX
> except for one of them. That makes more sense than a controller with
> differently capable physical channels. If that is indeed the
> situation, then the controller is actually 'duplex' and there should
> be no tx-only/rx-only property anywhere. Everything automatically
> falls into place because client drivers are written for specific
> targets and, unless you write some code, there can be no TX call to a
> remote that doesn't listen.

Unfortunately it's a restriction of the hardware (or the controller as
you call it, although it's not really a controller).  There is only
one IRQ for Rx'ing and that's wired up to the A9's mailbox (Mailbox
0).  If one of the remote processors attempted to send a message
through any of the other mailboxes (other than the Mailbox 0), then no
one would hear the doorbell ring and the message would go unserviced.

> >> > +
> >> > +       for (i = 0; i < mbox->num_chans; i++) {
> >> > +               chan_info = mbox->chans[i].con_priv;
> >> > +
> >> > +               /* Is requested channel free? */
> >> > +               if (direction != MBOX_LOOPBACK &&
> >> >
> >> Consider this example when 2 clients ask for same physical channel but
> >> in different modes.
> >>            mboxes = <&mboxA 0 1 MBOX_TX>;
> >>            mboxes = <&mboxA 0 1 MBOX_LOOPBACK>;
> >>
> >> You happily assign 2 virtual channels backed by one physical channel
> >> {mboxA, 0, 1}. The 2 clients think they can freely do startup(),
> >> shutdown() and send_data() on the channels. But obviously we are
> >> screwed with races like
> >>    client1.startup()
> >>     -> client2.startup()
> >>         -> client2.send_data()
> >>             -> client2.shutdown()
> >>                 -> client1.send_data()  XXXX
> >
> > Good catch and a fair point.  As you say, it's unlikely to happen, but
> > I would like to prevent it in any case.
> >
> No, such races are a practical problem. We must own them.
> 
> I was talking about problems that arise because someone wrote bad
> DT... those are not 'practical' problems because there are too many
> ways to screw up with bad DT properties that if we try to check for
> them we'll go insane.

The only thing I'm having trouble with protecting at the moment is
other clients _also_ requesting a LOOPBACK channel.  I would like to
check which clients have already requested one/them, however that
information is not available until _after_ xlate() has been called,
which is pretty frustrating.  Perhaps I'll put a comment in instead.

> >>  Now you can shove in some more checks to 'fix' the race OR you can
> >> simply expose only physical channels.
> >
> > We can't expose all of the channels.  There are too many and would
> > take up too much *unused* memory.
> >
> I am aware of that. I said expose _only_ physical channels, not _all_ :)

It's impossible to know which physical channels will be used by
clients and subsequently which physical channels to expose.

-- 
Lee Jones
Linaro STMicroelectronics Landing Team Lead
Linaro.org │ Open source software for ARM SoCs
Follow Linaro: Facebook | Twitter | Blog