Re: [RFC 1/6] mailbox: add core framework

[Rob Herring]

On Mon, Feb 10, 2014 at 1:59 PM, Courtney Cavin
<courtney.cavin@xxxxxxxxxxxxxx> wrote:
> On Mon, Feb 10, 2014 at 08:09:34PM +0100, Josh Cartwright wrote:
>> On Mon, Feb 10, 2014 at 11:52:05AM -0600, Rob Herring wrote:
>> > On Mon, Feb 10, 2014 at 8:11 AM, Arnd Bergmann <arnd@xxxxxxxx> wrote:
>> > > On Friday 07 February 2014 16:50:14 Courtney Cavin wrote:
>> [..]
>> > >> +int mbox_channel_notify(struct mbox_channel *chan,
>> > >> +             const void *data, unsigned int len)
>> > >> +{
>> > >> +     return atomic_notifier_call_chain(&chan->notifier, len, (void *)data);
>> > >> +}
>> > >> +EXPORT_SYMBOL(mbox_channel_notify);
>> > >
>> > > What is the reason to use a notifier chain here? Isn't a simple
>> > > callback function pointer enough? I would expect that each mailbox
>> > > can have exactly one consumer, not multiple ones.
>> >
>> > It probably can be a callback, but there can be multiple consumers. It
>> > was only a notifier on the pl320 as there was no framework at the time
>> > and to avoid creating custom interfaces between drivers. On highbank
>> > for example, we can asynchronously receive the events for temperature
>> > change, power off, and reset. So either there needs to be an event
>> > demux somewhere or callbacks have to return whether they handled an
>> > event or not.
>>
>> I'm not familiar with highbank IPC, but with these requirements should
>> the mailbox core even bother with asynchronous notifier chain?  It
>> sounds like a better fit might be for the mailbox core to implement a
>> proper adapter-specific irqdomain and used a chained irq handler to
>> demux (or have consumers request with IRQF_SHARED in the shared case).
>
> Although modeling this using irqdomains makes sense for the notification
> bit, and would probably suit most adapters, there's the issue of data
> being passed around which doesn't quite fit.  "Ok, I have mail... where
> is it?"  Did you have something in mind for that?
>
> Frankly, I don't see the notifier chain as being extraneous or
> over-complicated here core-wise or implementation-wise, and unless I
> understand Rob incorrectly, should suit the existing use-cases.  Am I
> missing something?

Well, I think notifiers are not liked very much. I don't know that irq
handlers would be the right answer either as these are not h/w events
really and we may not want handlers to run in irq context. I would say
a callback similar to how the dma engine framework works is the right
answer. On the send side, you may want to have completion callbacks as
well.

Rob
--
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/