Re: [PATCHv7 2/5] mailbox: Introduce framework for mailbox

[Matt Porter <mporter@linaro.org>]

On Fri, Jun 20, 2014 at 01:59:30AM +0530, Jassi Brar wrote:
> On 20 June 2014 00:33, Matt Porter <mporter@linaro.org> wrote:
> > On Thu, Jun 19, 2014 at 07:17:11PM +0100, Sudeep Holla wrote:
> >
> >> >+ * After startup and before shutdown any data received on the chan
> >> >+ * is passed on to the API via atomic mbox_chan_received_data().
> >> >+ * The controller should ACK the RX only after this call returns.
> >>
> >> Does this mean we can't support asynchronous messages from the remote.
> >> One possible scenario I can think is if the remote system power controller
> >> has feature to configure the bounds for thermal sensors and it can send
> >> async interrupt when the bounds are crossed. We can't just block one channel
> >> for this always. Again this might have been discussed before and you might have
> >> solution, I could not gather it with my brief look at older discussions.
> >
> > The way I see it we are simply putting the burden on the client to
> > implement very little in the rx_callback. In my case, we will have a
> > single client which is the IPC layer. The controller driver will notify
> > the IPC client layer which will do as little as possible in the
> > rx_callback before returning. We'll handle asynchronous dispatch of
> > events within our IPC layer to the real client drivers rather than in
> > the controller driver.
> >
> Yes. So do I.
> 
> >> >+/**
> >> >+ * mbox_client_peek_data - A way for client driver to pull data
> >> >+ *                     received from remote by the controller.
> >> >+ * @chan: Mailbox channel assigned to this client.
> >> >+ *
> >> >+ * A poke to controller driver for any received data.
> >> >+ * The data is actually passed onto client via the
> >> >+ * mbox_chan_received_data()
> >> >+ * The call can be made from atomic context, so the controller's
> >> >+ * implementation of peek_data() must not sleep.
> >> >+ *
> >> >+ * Return: True, if controller has, and is going to push after this,
> >> >+ *          some data.
> >> >+ *         False, if controller doesn't have any data to be read.
> >> >+ */
> >> >+bool mbox_client_peek_data(struct mbox_chan *chan)
> >> >+{
> >> >+       if (chan->mbox->ops->peek_data)
> >> >+               return chan->mbox->ops->peek_data(chan);
> >> >+
> >> >+       return false;
> >> >+}
> >> >+EXPORT_SYMBOL_GPL(mbox_client_peek_data);
> >>
> >> I am unable to understand how this API will be used. IIUC when the controller
> >> receives any data from remote, it calls mbox_chan_received_data to push data to
> >> client.
> >
> > Good question.
> >
> > That function is a no-op if your client chooses not to populate
> > rx_callback. It's not explicitly stated, but the implementation is a
> > no-op if rx_callback is NULL so rx_callback seems to be intended as an
> > optional field in the client data.
> >
> > I'm also not clear of the scenario where this could be used. I
> > originally thought .peek_data() was an alternative to the callback for
> > polling purposes except it clearly states it needs the callback to carry
> > the data.
> >
> > I probably missed earlier discussion that explains this.
> >
> peek_data is just a trigger for controller to flush out any buffered
> RX via mbox_chan_received_data() to upper layer. Intended usecase is
> irq-mitigation for QMTM driver, as Arnd pointed out a few months ago.

Ok, that makes much more sense now.

Thanks,
Matt