Re: [PATCH v2 5/7] dt-bindings: i3c: Document core bindings

[Rob Herring]

On Thu, Dec 21, 2017 at 4:41 AM, Boris Brezillon
<boris.brezillon@xxxxxxxxxxxxxxxxxx> wrote:
> On Wed, 20 Dec 2017 12:06:45 -0600
> Rob Herring <robh@xxxxxxxxxx> wrote:
>
>> On Sat, Dec 16, 2017 at 07:35:37PM +0100, Boris Brezillon wrote:
>> > On Sat, 16 Dec 2017 11:20:40 -0600
>> > Rob Herring <robh@xxxxxxxxxx> wrote:
>> >
>> > > On Thu, Dec 14, 2017 at 04:16:08PM +0100, Boris Brezillon wrote:
>> > > > A new I3C subsystem has been added and a generic description has been
>> > > > created to represent the I3C bus and the devices connected on it.
>> > > >
>> > > > Document this generic representation.
>>
>> [...]
>>
>> > > So please define the node
>> > > name to be "i3c-controller". That's more inline with other node names
>> > > than i3c-master that you used below.
>> >
>> > Hm, not sure i3c-controller is appropriate though, because you can have
>> > slave controllers. Maybe i3c-host, but I'd prefer to keep the term
>> > master since it's employed everywhere in the spec. I can also be
>> > i3c-master-controller if you prefer.
>>
>> Okay, i3c-master is fine. Just make it explicit.
>
> Okay.
>
>>
>> > > > +I3C devices
>> > > > +===========
>> > > > +
>> > > > +All I3C devices are supposed to support DAA (Dynamic Address Assignment), and
>> > > > +are thus discoverable. So, by default, I3C devices do not have to be described
>> > > > +in the device tree.
>> > > > +This being said, one might want to attach extra resources to these devices,
>> > > > +and those resources may have to be described in the device tree, which in turn
>> > > > +means we have to describe I3C devices.
>> > > > +
>> > > > +Another use case for describing an I3C device in the device tree is when this
>> > > > +I3C device has a static address and we want to assign it a specific dynamic
>> > > > +address before the DAA takes place (so that other devices on the bus can't
>> > > > +take this dynamic address).
>> > > > +
>> > > > +Required properties
>> > > > +-------------------
>> > > > +- i3c-pid: PID (Provisional ID). 64-bit property which is used to match a
>> > > > +          device discovered during DAA with its device tree definition. The
>> > > > +          PID is supposed to be unique on a given bus, which guarantees a 1:1
>> > > > +          match. This property becomes optional if a reg property is defined,
>> > > > +          meaning that the device has a static address.
>> > >
>> > > What determines this number?
>> >
>> > Part of it is fixed (manufacturer and part id) and the last few bits
>> > represent the device instance on the bus (so you can have several
>> > identical devices on the same bus). The manufacturer and part ids
>> > should be statically assigned during production, instance id is usually
>> > configurable through extra pins that you drive high or low at reset
>> > time.
>>
>> Sounds like an I2C address at least for the pin strapping part...
>
> The address space of this instance-id is smaller (4bits) than the I2C
> one (7bits), and more importantly, the instance-id is not required to
> be unique, it's the aggregation of the vendor-id, part-id and
> instance-id that has to be unique. So, if you were thinking about using
> this id to uniquely identify the device on the bus it's not a good idea.

No, no. I was just commenting how I2C devices typically do the same
pin strapping to make addresses unique.


>> > > > +Optional properties
>> > > > +-------------------
>> > > > +- reg: static address. Only valid is the device has a static address.
>> > > > +- i3c-dynamic-address: dynamic address to be assigned to this device. This
>> > > > +                      property depends on the reg property.
>> > >
>> > > Perhaps "assigned-address" property would be appropriate. I'm not all
>> > > that familiar with it though.
>> >
>> > Again, the spec use the term "dynamic address" everywhere, and I'd like
>> > to stay as close as possible to the spec.
>>
>> I looked at assigned-addresses a bit more and that won't really fit
>> because it should be the same format as reg. So I think reg should
>> always be the PID as that is fixed and always present. Then the DAA
>> address is separate and can be the i3c-dynamic-address property.
>>
>> However, there's still part I don't understand...
>>
>> > > > +               /* I3C device with a static address. */
>> > > > +               thermal_sensor: sensor@68 {
>> > > > +                       reg = <0x68>;
>> > > > +                       i3c-dynamic-address = <0xa>;
>>
>> I'm confused as to how/why you have both reg and dynamic address?
>
> Some I3C devices have an I2C address (also called static or legacy
> address in a few places). The static/I2C/legacy address is used until
> the I3C device is assigned a dynamic address by the master. The whole
> point of specifying both an I2C address (through the reg property) and
> a dynamic address (through the i3c-dynamic-address) is to tell the
> controller that a specific dynamic address should be assigned to this
> device using the SETSADA (Set Dynamic Address from Static Address)
> command before a DAA (Dynamic Address Assignment) procedure is started.
> This way, the device will not participate to the DAA (because it
> already has a valid DA) and the dynamic address can't be assigned to
> a different device (which is one of the problem with the automatic DAA
> procedure).

Okay, think I got it now.

I think we should extend "reg" to have either I2C address, I3C PID, or
both (in a defined order). I'm assuming you can always distinguish a
static I2C address and an I3C PID just by upper bits all being 0s for
I2C addresses. Maybe both is not needed? This means we'd have to allow
64-bit I2C addresses (#address-cells=2), but that should be easily
fixed if that causes problems in the kernel.

So i3c-pid would go away and i3c-dynamic-address stays.

Rob