On 07/03/2014 10:48 AM, Dong Aisheng wrote:
> On Thu, Jul 03, 2014 at 09:12:49AM +0200, Marc Kleine-Budde wrote:
>> On 07/03/2014 05:48 AM, Dong Aisheng wrote:
>>> On Wed, Jul 02, 2014 at 09:13:07PM +0200, Marc Kleine-Budde wrote:
>>>> On 07/02/2014 07:54 PM, Oliver Hartkopp wrote:
>>>>> I'm not really familiar with the naming concept in device trees.
>>>>>
>>>>> What is your opinion about the remarks below?
>>>>
>>>> The entries in the DT, at least on freescale baords, follow the naming
>>>> scheme of the reference manual. E.g. on the mx25 it's can1 and can2:
>>>>
>>>>     can1: can@43f88000 { ... }
>>>>     can2: can@43f8c000 { ... }
>>>>
>>>> And on the mx28, its:
>>>>
>>>>     can0: can@80032000 { ... }
>>>>     can1: can@80034000 { ... }
>>>>
>>>> Because the imx25 datasheet uses a "1" based counting scheme, while the
>>>> imx28 uses a "0" based one.
>>>>
>>>> So it's best practise to follow the naming and numbering scheme of the
>>>> hardware reference manual.....and if you have access to the
>>>> documentation of the m_can core, use clock names of the m_can core for
>>>> the clock-names property.
>>>>
>>>
>>> Based on my knowledge, device tree allows define phandle name according to
>>> the real device name of HW according spec while the device node name should
>>> be general(e.g can@80032000 rather than flexcan@80032000).
>>> For imx6sx, there are already following entries in
>>> arch/arm/boot/dts/imx6sx.dtsi
>>> flexcan1: can@02090000 {...}
>>> flexcan2: can@02094000 {...}
>>> So i'd prefer to define as:
>>> m_can1: canfd@020e8000 {...}
>>> m_can2: canfd@020f0000 {...}
>>>
>>>
>>> One problem is there're can alias already.
>>> aliases {
>>> 	can0 = &flexcan1;
>>> 	can1 = &flexcan2;
>>> 	...
>>> }
>>> I'm not sure adding can2&can3 for mcan is properly:
>>> aliases {
>>> 	can0 = &flexcan1;
>>> 	can1 = &flexcan2;
>>> 	can2 = &m_can1;
>>> 	can3 = &m_can2;
>>> 	...
>>> }
>>> Since the m_can driver does not need to use aliases,
>>> so i will not add them.
>>
>> IMHO It's fine too add the can{2,3} aliases to m_can, too.
>>
> 
> I think the main problem for doing this way is that the meaning of id
> return by of_alias_get_id may be not persistent.
> e.g
> For MX6SX
> aliases {
> 	can0 = &flexcan1;
> 	can1 = &flexcan2;
> 	can2 = &m_can1;
> 	can3 = &m_can2;
> 	...
> }
> 
> For other platform, it could be:
> aliases {
> 	can0 = &m_can1;
> 	can1 = &m_can2;
> 	...
> }
> It's hard for driver to use.

The driver doesn't make use of it, does it?

> And actually the M_CAN driver does not need to use the alias.
> So i wonder if it makes sense to add the alias for m_can devices
> like that.

For example the imx53 has two different SPI units, in the alias section
we see:

                spi0 = &ecspi1;
                spi1 = &ecspi2;
                spi2 = &cspi;

Marc

-- 
Pengutronix e.K.                  | Marc Kleine-Budde           |
Industrial Linux Solutions        | Phone: +49-231-2826-924     |
Vertretung West/Dortmund          | Fax:   +49-5121-206917-5555 |
Amtsgericht Hildesheim, HRA 2686  | http://www.pengutronix.de   |